Method and system for determining the identity of an individual animal in a population of animals wi

专利摘要:
SYSTEM AND METHOD FOR INDIVIDUAL ANIMALS IDENTIFICATION BASED ON BACK IMAGES. The present disclosure relates to a system and method for identifying individual animals based on images, such as 3D images, of animals, especially cattle and cows. When animals live in areas or enclosures where they move freely, it can be tricky to identify the individual animal. In a first aspect, the present disclosure relates to a method for determining the identity of an individual animal in a population of animals with known identity, the method comprising the steps of acquiring at least an image of the backside of a preselected animal, extracting data from at least one image relating to the hindquarter anatomy and/or hindquarter topology of the pre-selected animal and comparing and/or combining the extracted data with reference data corresponding to hindquarter anatomy and/or hindquarter topology rear of animals with known identity, thus identifying the pre-selected animal. The method and system can be used to monitor food intake, such as feeding dairy cows and health status.
公开号:BR112018000046B1
申请号:R112018000046-9
申请日:2016-06-30
公开日:2022-01-25
发明作者:S0Ren Borchersen；Claus Borggaard；Niels Wors0E Hansen
申请人:Viking Genetics Fmba；
IPC主号:

专利说明:

[001] The present disclosure relates to a system and method for identifying individual animals based on images, such as 3D images, of animals, especially cattle and cows. Fundamentals of Invention
[002]The identification of individuals of livestock animals such as pigs, cattle and cows is usually performed by systems such as non-electronic identification tags, e.g. ear notch, ear tags, number tags on neck chains and electronic identification, the most common of which include electronic ear tags, microchips, and electronic collars. Each of these systems has advantages and disadvantages and the systems cannot be used exclusively to identify individuals in groups with simultaneous automatic collection of other relevant information for each animal.
[003] When producing milk from cows, up to 80% of the expenses are used to feed the cows. Optimizing feed intake in relation to milk production and cow health can reduce expenses not only used for food but also for medication or veterinary support. Cow health and welfare can be increased by having cows in a loose housing system where cows can move around and strengthen bones and muscles. In this loose housing system, it can be difficult to determine feed intake for each cow, as an estimate of feed intake must be correlated with the individual cow.
[004]WO 95/28807 ('Three-dimensional phenotypic measuring system for animals', Pheno Imaging Inc.) describes a three-dimensional phenotypic measuring system for animals such as dairy cows. The system uses a large amount of laser light beams modulated from a laser camera to measure approximately 100 dots per square inch of the animal. Each laser beam measures intensity, horizontal, vertical and depth dimensions, and by combining the measurements, the system composes a very accurate three-dimensional image of the animal. The system calculates the desired phenotypic measurements for animal conformation by combining measurements from selected points on the animal. The system stores measurements for each animal in a computer database for later use. The system also stores a light intensity image of the animal's marks which is compared with other stored images. The system takes side view photos of the animals and is used to classify the animals. The system can check the database for each new animal to ensure that the same animal is not processed more than once.
[005]EP 2027770 ('Method and apparatus for the automatic grading of condition of livestock', Icerobotics Limited) describes a method and apparatus for grading a characteristic of an animal. The animal is guided to a detection area, after which an image of the animal's back is captured. The animal's identity is also established when the animal is in the detection area. Identity is determined by reading an identification mark located on the animal. Image analysis identifies anatomical points and determines angles at these points. The angles are then used to calculate a rating for an animal characteristic. A modality is presented to automate the determination of body score condition in dairy cows using seven angles determined at three anatomical points from an image on the back of the cow.
[006] Thus, the identification of an individual animal is easy if it is possible to have access to the identification mark that is attached to each animal. But many animals live in a loose housing system where access to each animal's identification tag is not possible at any time. In addition, the animals can be located in an outdoor field. In both situations, it is impossible to monitor each individual animal if the identification mark cannot be accessed. Summary of the Invention
[007]If an individual animal in a loose housing system cannot be monitored constantly or frequently, it is virtually impossible to record the food intake of each animal. The presently disclosed invention therefore relates to a method for determining the identity of an individual animal from the natural appearance and/or topology of the animal's back. The present inventors have realized that each animal has unique characteristics associated with the natural configuration, appearance, topology and/or contours of the animal's back. The inventors also realized that these features can be extracted from one or more images that show at least one back of an animal. The favorable result is that animals can be identified from an image of the back of said animal if an earlier, and preferably substantially recent, image exists of the same animal, by comparing these images, such as by extracting corresponding features from the images that can be compared. By using backside images of animals, it is possible to identify and monitor the animals from above, for example based on camera systems mounted on the roof of a barn/stable or a camera system in the air, for example in the air for through a drone. Aerial camera systems can furthermore be applied to identify and monitor animals in an outdoor field.
[008] In one embodiment, the presently disclosed method therefore comprises the steps of: • obtaining at least one image of at least one back of an animal, for example, an unidentified animal, and • extracting data from the at least one image obtained, the extracted data, for example, predefined features, related to the natural appearance, anatomy, contour and/or topology of the animal's back.
[009]When the image(s) have been analyzed and the extracted data obtained, the animal can be identified if, for example, predefined features in the image correspond to predefined features in a previous image (reference ) of the same animal. A correspondence between two or more images of the same animal can therefore be established because the anatomy of an animal's back is unique to each animal, at least in a herd or population of animals with only a limited number of animals. The previous (reference) image can furthermore be associated with the animal's identity, for example with the animal's identity corresponding to the animal identification tag. Thus, once a correspondence has been established between the animal's identity, for example, through the identification mark and one or more predefined anatomical features of the animal's back, this animal can be identified only through images that show (at least one part of) the back of said animal.
[010] In an additional embodiment, the extracted data are compared with reference data extracted from at least one reference image of a back of an identified animal, where the identity information of the identified animal can be linked to at least one image of reference. Also, based on the comparison, it can be determined whether the unidentified animal matches the identified animal. The steps of comparing the extracted data with reference data and determining whether the unidentified animal corresponds to an identified animal can be repeated for a plurality of reference images of a plurality of identified animals until a match is obtained and the unidentified animal is identified. The extracted data can also be combined or compared against a database of predefined (anatomical) characteristics, the database, for example, which comprises predefined characteristics of each animal in the population or herd of animals that need to be distinguished, and a set of Predefined traits can be associated with exactly one animal of known identity. Once a match between sets of predefined traits is obtained, the unidentified animal is identified.
[011] The present disclosure further relates to a method for determining the identity of an individual animal in a population of animals with known identity, the method comprising the steps of: • acquiring at least one image of the back of a pre-selected animal, and• extract data from said at least one image relating to the anatomy, natural appearance and/or topology of the back of the pre-selected animal and•compare and/or combine said extracted data with reference data corresponding to anatomy, natural appearance and/or or topology of the back of animals with known identity, thus identifying the pre-selected animal.
[012]The system and method disclosed herein can therefore determine the individual animal based on the anatomy of an animal's back, whereby it is possible to estimate the intake of, for example, forage by combining the invention described herein with the system for determine food consumption as described for example. WO 2014/166498 ("System for determining feed consumption of at least one animal", Viking Genetics FMBA), where an imaging system is used to assess the amount of food consumed by each identified animal, determining the feed reduction in subsequent images of the feeding area in front of each identified animal.
[013]With the currently disclosed identification method, it may be feasible for animals not to need a visible identification mark because the animals are distinguishable based on the rear images. Thus, since the images are initially acquired on the back of all animals, they can be distinguished from each other based on the different images of the back of each animal and thus identified.
[014]The comparison of data extracted from at least one image with data extracted from a previous (reference) image may be performed by any possible method for comparing data and may be based on any data extracted directly from the images or any calculated data based on the images. Vectors can be calculated, scores can be determined, as principal components (PC scores) for a principal component analysis and these can be included in the comparison process and/or used to perform additional calculations, as a dot and the comparison is performed from the calculated product.
[015]Animals can be animal species, breeds or groups and, for example, be selected from the group of cattle, cows, dairy cows, bulls, calves, pigs, sows, wild boars, geldings, piglets, horses, sheep, goats , deer.
[016]Reference data can be extracted from at least one (reference) image acquired on the back of each of the animals in the animal population. A reference image of an animal can be obtained by simultaneously determining the identity of the animal by reading an identification marker affixed to said animal.
[017]Therefore, at least one reference image of the back of an identified animal can, for example, be obtained by •providing the identification number of an animal, hereby the animal being an identified animal, •providing at least one image of the back of the identified animal, and • storing in a database the identification number of the identified animal together with at least one image of the back of the identified animal, the image being here a reference image.
[018]At least one reference image of the back of an identified animal can be obtained frequently, for example, per day, but it can be determined due to the type of animals to be identified. A relatively short period of time, for example a day or two, can be important in identifying dairy cows.
[019] The method can be based on images and reference images that are topographic images of the back of animals, such images can be obtained as 3D images.
[020] The present disclosure also relates to an animal identification system for determining the identity of an individual animal among a population of animals with known identity, the system may comprise an imaging system configured to acquire at least one image of the animal. back of a pre-selected animal, and • a processing unit configured to-extract data from said at least one image relating to the anatomy, natural appearance and/or topology of the back of the pre-selected animal and-combine said extracted data with reference data corresponding to the anatomy, natural appearance and/or topology of the back of each of the animals with known identity, thus identifying the pre-selected animal.
[021] The system may further comprise a reference imaging unit for providing one or more reference images of an animal in the animal population, said reference imaging unit comprising at least one identity determining device configured to determine the identity of said animal, such as reading at least one identification marker affixed to said animal, and - at least one camera configured to acquire at least one image (reference) of the back of said animal.
[022]The system can be further configured to associate the determined identity of the animal with said at least one image acquired by said camera(s) and optionally store said at least one image as an image of reference.
[023] Thus, the pre-selected animal can be seen as unidentified because, at the time of image acquisition, the system may not know the identity of the animal. On the other hand, the identity of the pre-selected animal is not unknown per se, as it has already been identified and the reference data, possibly comprising features of the animal's anatomy, exist in such a way that the pre-selected animal can be identified automatically as soon as possible. after image acquisition. Reference data can be based on/extracted from one or more previous images of the pre-selected animal.
[024]The processing unit may be part of a computing device and images, extracted data, reference images and/or reference data may be exchanged with a database which may be part of the animal identification system or the system can have access to the database. The imaging system may include one or more cameras. The animal identification system can be configured in such a way that at least some of said cameras are arranged so that they are located above the animals to be identified so as to be able to image the back of the animals. Cameras can be in a fixed location, but can be configured so that the field of view can be varied to create images in different areas. The currently disclosed animal identification system may also be part of an aerial system, as indicated above.
[025] Another embodiment of the animal identification system refers to a system for determining the identity of an individual animal from the natural appearance and/or topology of the back of said animal, the system may comprise at least one camera to obtain at least one image of the back of an unidentified animal,• at least one database or entry into at least one database to store data related to at least one reference image of the back of an identified animal and to store related data to at least one image of the back of an unidentified animal, • data transmission means for transmitting data from said at least one camera to said database, and • at least one processing means connected to said database, said processing means being configured to compare data extracted from said at least one image of an unidentified animal with data extracted from at least one image of reference where said extracted data are related to the natural aspect and/or topology of the animal's back and based on this comparison, determine if said unidentified animal corresponds to said identified animal.
[026] Preferably, the images obtained from the back of the animals are 3D images and can be obtained by any suitable camera system capable of providing 3D images, such a system can be based, for example, on range cameras, stereo cameras and time-of-flight cameras.
[027] The method and system can be used not only to determine the identity of animals, but also, for example, to determine the amount of food consumed by an animal. Feeding images located in front of an animal that is eating can be analyzed by similar methods as described herein for animal identification to determine the amount of feed consumption. The invention makes it possible to determine the feed consumption of individual animals and to store such information in a database, for example in connection with that animal's file. Also the classification conditions or health conditions can be monitored with the system described here and this information can also be stored in the animal file, allowing to follow the development of an animal and/or to optimize its production, for example, milk production, controlling the type and amount of power consumption.
[028]The systems described here can be configured to perform any of the methods described here. Brief description of figures
[029] Fig. 1 illustrates cows that are eating in a barn in which a system of the present invention is installed.
[030] Fig. 2 illustrates examples of different preselected points on the back of a cow.
[031] Fig. 3 illustrates examples of established characteristics in relation to the back of an animal, here the back of a cow.
[032] Fig. 4 illustrates the height profile along the backbone of two cows.
[033] Fig. 5 illustrates a 3D Imaging Mesa reconstruction of part of a cow with a height above 90 cm from ground level.
[034] Fig. 6 illustrates the back of a cow.
[035] Fig. 7 illustrates the back of the cow in Fig. 6 with indications of some data/characteristics that can be used in the analysis.
[036] Fig. 8 illustrates the determination of the area based on resized data obtained from the part of a cow with a height above 90 cm from the ground.
[037] Fig. 9 and 10 illustrate different thickness profiles and height profiles at predetermined heights of two cows. Data is scaled.
[038] Fig. 11 illustrates a vertical height profile of a cow.
[039] Fig. 12 illustrates the determination of a cow based on the neural network, as a deep learning system. DETAILED DESCRIPTION OF THE INVENTION
[040] One aspect of the invention pertains to a method for determining the identity of an individual animal from the natural appearance and/or topology of the animal's back as described above. When comparing data extracted from at least one image of an (unidentified) animal with reference data extracted from at least one reference image, the data to be compared is obtained from the corresponding characteristics of the animals' back. The data to be compared are extracted from the characteristics of the animals' backs. Such features are based on the natural appearance and/or topology of the animal's back. Natural features may include any feature described herein, as well as any marks on the skin such as scratches, scars, etc. Preferably, the natural features do not include permanent ID tags applied to the animal by humans, such as applied identification marks or numbers, for example freeze-marking, heat-marking or tattooing.
[041]An animal's identity can be an identification number, a name or code used to uniquely identify the animal, for example, in the population, in a region, country and/or globally. An "identified animal" is therefore an animal with an identity.
[042] An "unidentified animal", as used herein, means an animal for which, at any given time, no identity is linked to an image of the animal's back, and where the identity may be an identification number. of the animal. An unidentified animal is preferably an animal belonging to a population of identified animals, e.g. each animal with an identification number, this population may be a herd of e.g. cows or cattle or other animals described elsewhere . By using the method and system as described here, animals can change status between identified and unidentified animals and come back again within a short period of time. An animal's status change can occur when an animal crosses a corral or shed and at least one new image of the animal's back is obtained. When data extracted from at least one image has been compared with data extracted from at least one reference image and a match is found, the animal changes status from unidentified to identified. An unidentified animal may therefore also be denoted as an animal to be identified.
[043]An image of an unidentified animal is preferably obtained in a location where it is not easy or impossible to unambiguously register an animal ID tag simultaneously with the acquisition of the image. This location may be in a field where the distance from an electronic ID tag to an antenna capable of recording IDs is too great to register and/or a non-electronic ID tag cannot be visualized through an image for too long. distance and/or position of the tags on the animal makes it impossible to see the ID tag. Location can also be where animals are too close together to register an individual ID which, of course, can be linked to an image of the animals taken at substantially the same time the animal ID is registered. This location can also be a field or a loose housing system, for example a loose housing system for cows, such as a feeding area for cows in loose housing systems.
[044] The term "back of an animal", as used herein as "back of an unidentified animal" or "back of an identified animal", is a reference to the anatomical part of the animal that contains the vertebral column, i.e. , the back. Thus, the term "back of an animal", as used herein, is not intended to refer to the hind or rear part of the animal, for example, the part of the cow which comprises the hind legs, as can be seen from a side behind the animal. Thus, the at least one image and the at least one reference image are taken from above the animal, for example directly from above or from an angle above the animal. Images and reference images taken from above an animal can, along with the back, also include the animal's head and neck and these parts of the animals can also be used to compare an image with at least one reference image.
[045] The present invention is based on the realization that the back of an animal can be used as a unique anatomical feature. Thus, by acquiring one or more images of at least a part of the back and extracting data related to the anatomy and/or topology of the back, the animal can be identified by comparing itself with previously mentioned characteristics. An image of the back of an animal, as used herein, must therefore comprise sufficient information so that the relevant features of the anatomy and/or topology of the back can be extracted from the image. In one embodiment, at least a portion of the spine is therefore included in the image. In yet another embodiment, an image of an animal's back includes the backbone from head to tail along and at least to the point where the neck begins. The beginning of the neck (viewed from the back to the animal's head) can be defined by a "neck point", which is the location between the animal's body and head where the body thickness is less than a predetermined part of the width. widest of the animal, for cows and cattle, the "neck point" may be where the neck is less than 38% of the widest width of the animal. The "neck point" for cows is illustrated in Fig. 7 as the area, including the left endpoints of the curves illustrated on the back of the cow. Preferably, also the position of at least one shoulder-blade (scapula) is included when taking an image of an animal's back.
[046] An image of the back of an animal preferably also includes at least the top 10, 15 or 20 cm of at least one side of the animal, where this distance is calculated from any highest point along the spine and to down, by this the vertebral column and a virtual lower line, for example, 15 cm below the vertebral column would have similar contours (be it parallel). For cows/cattle, an image of the back should preferably include at least the vertebral column from head to tail to neck and at least 15 cm below the vertebral column on at least one side of the cow/cattle.
[047] When obtaining at least one image of an animal's back, the ideal situation is to obtain at least one image substantially directly above the animal, where the image may include the spine and the area on either side of the spine that is visible from the point above. However, for practical reasons, it may be unfeasible to use an imaging system where each animal, for example in a stable, can be viewed directly from above. In practical implementation (a part of), the area on one side of the spine may be partially or totally blocked by the spine higher in the field of view in the image(s), for example if the imaging system does not is located high enough in relation to the corresponding animals.
[048]Thus, when obtaining at least one image of the back of an animal where the image is obtained from an angle such that it does not include data from both sides of the vertebral column, or if data from a part of one side of the vertebral column is missing, then the missing data can be calculated in such a way that the corresponding data from one side of the spine is mirrored to the other side of the spine to obtain a complete set of data on the animal's back. This "complete set of data" is to be understood as the term "imagery" as used herein, i.e. an "image" may be data obtained from an image obtained without mirroring of data or it may be data obtained from of an image obtained by mirroring some data. In practice, an image of an animal can be obtained, including the spine and area on only one side, for example, the left side of the animal, this image can be transformed into a "complete dataset", reflecting the data from the left side to the right side of the animal before using the image (ie, the entire dataset) to determine the animal's identification as described herein. Mirroring data from one side of an animal's back to the other side of the animal's back can be performed for all images obtained, such as images obtained at an angle less than ± 90° where the starting position is the location of the longitudinal direction vertebral column.
[049]Mirroring data step can be performed when data processing records missing information, so that missing information can be obtained by mirroring the corresponding data on the other side of the spine.
[050]Mirroring is not necessary if there is enough information in the image so that sufficient data concerning the anatomical and/or topological features of the back can be extracted from the image to identify the animal.
[051]Data obtained from an image may also include neck and/or head related data. However, this data may be used for purposes other than determining the identity of the animal, for example, to determine the location of the nose. When determining the location of the nose, this can correspond to whether the animal is eating and where the animal is eating, such information can be correlated with information that determines food intake. Thus, when identifying the nose of an animal that is eating, this corresponds to identifying the location of a virtual feeding channel from which food intake can be determined.
[052]The term "compare images" should be understood as comparison of data extracted from images.
[053] In the reference image of an animal, the identity of the animal shown in the image is known.
[054]One or more reference images of the back of an animal, such as an identified animal, may be taken at least once a month, such as at least every third week, for example, at least every second week or at least once a week. It is preferable that a reference image is taken at least twice a week, such as at least three times a week, for example, at least four times a week, such as at least five times a week. Preferably, at least one reference image of an animal is taken at least every second day, more preferably at least one reference image of an animal is taken at least once a day, such as twice a day, for example , three times a day.
[055] To determine an interval between obtaining at least one reference image on the back of an animal, possible changes in the natural appearance and/or topology of the back must be considered. The interval between obtaining subsequent reference images must be short enough to record changes in appearance and/or topology of the back for the individual animal and still be able to identify the animal based on the back images. For dairy cattle, the time interval between obtaining reference images is preferably shorter than for fattening cattle. The purpose of identifying an unidentified animal should also be considered when determining a time interval between obtaining reference images. Such purposes are described elsewhere herein and may relate to a request for information, for example, physiological status, stature, health, physical status, etc.
[056]A reference image of an animal can be obtained at a location where it is known that the animal must pass at least once a day if this is the determined interval between obtaining reference images. Such a location may be at the entrance or exit of the milking area if the animal is in a group of dairy cows. A location for taking a reference image may also be at a drinking trough, driveway, drinking station, or other location where the animal is likely to be or pass by every day or often.
[057]The suitable time and the longest acceptable time, that is, the interval between obtaining two reference images of a single animal can also be determined due to the characteristics of the animal, these characteristics can be breed, species, age, maturity, health etc. The range can also be determined due to the purpose of controlling the animal and the purpose of identifying the animal. The purpose of animal control may be for the production of milk, meat, juveniles (e.g. piglets) or semen or it may be for other purposes such as conservation or presentation eg ZOOs or use for competition eg racing of horses and jumps. Each purpose for keeping the animal can affect the animal's form, including rear appearance or rear topology differently and with different speed. An animal kept for milk production may have a negative energy balance and is usually getting bigger very fast during the milking period and therefore a short interval between obtaining reference images may be recommended, whereas an animal kept for meat production, although increasing in size, it does not change the appearance or topology of the back as fast as a dairy cow and for the animal kept for meat production, it may only be necessary to obtain a reference image once a month or once a fortnight. Other factors can also influence the appearance and/or topology of the animal's back, such as health.
[058]A reference image and/or reference data of an animal is an image of (the back) and animal or data, e.g. anatomical features, corresponding to the animal where the identity of the animal is known, i.e. if the image is stored in a database, the animal's identity is associated/connected to the image, and the data associated with the image comprises the animal's identity information.
[059] In one embodiment, at least one reference image of an animal's back is obtained by • providing the animal's identification number, hereby the animal being an identified animal and • acquiring at least one image of the identified animal's back .
[060]The identification number of the identified animal and at least one image of the back of the identified animal may subsequently be stored together in a database, the image being a reference image. Data can also be extracted from the image to provide reference data for the identified animal and reference data can be stored, for example, in a database. Storing the reference data only instead of the actual images is more efficient in terms of storage space.
[061]Providing an animal's identification number and providing at least one image of that animal's back can be done simultaneously or right after the other in any order. Soon can mean less than 60 seconds, for example less than 30 seconds, for example, less than 15 seconds, for example, less than 10 seconds, such as less than 5 seconds, for example, less than 1 second, such as less of 0.5 second.
[062]When the animal identification number is obtained and at least one image of the back of the same animal is obtained and these are stored together, this is a reference image of an identified animal, i.e. the animal identification and the appearance, anatomy and/or topography of its back is known or may be known by obtaining and processing data from at least one image and this data may be stored together with the animal's ID in a database. An animal's identification number can be obtained by any known method, for example, based on an electronic tag such as an electronic ear tag, an electronic tag on a collar or a microchip under the skin. Non-electronic labels are also possible.
[063]When the identity of an animal is obtained, for example, by an identity determination device, this can trigger a system to provide at least an image of the back of this identified animal. A reference image of the back of an identified animal can also be obtained shortly after the identification number of the identified animal has been provided.
[064] A reference image and/or animal identification can also be obtained manually, where the identification number is entered into a system by a human and/or a human can trigger a camera to obtain at least an image of the back of the animal. an animal with the identification number that is or must be entered into the system.
[065] In principle, any animal image, or data extracted therefrom, acquired as described here can become a reference image, because once an identification of the animal in the image is provided according to the method described here, there is an association/connection between the image of the animal and the identity of the animal in the image.
[066]When a new animal enters a population, for example when a new cow or cattle joins a herd, at least one reference image can be obtained on the back of that animal. The at least one reference image can initially be considered an image of an unknown animal and tested in the system to ensure that no match is obtained between this image and the reference images in the database. If a match is found between at least one image of the new animal and the reference images in the databases, the number of features used to compare images and reference images should preferably be increased until no match is obtained based on in the image of the new animal. Subsequently, the at least one image of the new animal can be considered a reference image or a group of reference images.
[067]For each animal, a series of reference images can be stored. When comparing at least one image of an unidentified animal with reference images, one may decide to only compare with the most recent reference images obtained for each identified animal, such reference images can be, for example, the last 2, 3 , 4, 5, 6, 7, 8, 9 or 10 reference images obtained for each animal or it could be an average of data extracted from reference images obtained later, e.g. 2, 3, 4, 5, 6, 7, 8, 9 or 10 times the animal was subjected to recording of reference images.
[068] In practice, each of at least one image of an unidentified animal can be compared with at least one reference image of a series of animals. The identity of an animal can be determined by comparing a series of images of the back of the animal. animal with a series of reference images of animals, for example in a herd, and the identity can be determined as the combination with reference images obtained most of the time. If, for example, 10 images of an unidentified animal are compared with reference images and 8 of these images match at least one reference image of animal A and the remaining 2 images correspond to at least one reference image of animal B, the unidentified animal can be determined to be animal A.
[069]The number of images of the back of an unidentified animal that must be compared with at least one reference image of a number of identified animals may be at least 5, such as at least 10, e.g. at least 15, such as at least 20, e.g. at least 25, such as at least 30, e.g. at least 35, such as at least 40, e.g. at least 45, such as at least 50, e.g. at least 75, such as at least 100. Preferably, the number of images of the back of an unidentified animal that is to be compared with at least one reference image of a number of identified animals is about 5, such as about 10, for example about 15, such as about 20, more preferably about 10, for example about 15.
[070]The image and the reference image can be topographical images of the animals' back, so both are 3D images. 3D images can be transformed into 3D image layers, hereby the image and reference image can be multiple 3D image layers, each including a number of pixels corresponding to the size (length and width) of the image. animal and the number of layers corresponding to the height of the animal.
[071] When determining the identity of an unidentified animal, the at least one image obtained is compared with the at least one reference image comparing data with respect to at least one characteristic obtained from the at least one image with a data in relation to at least one corresponding feature obtained from the at least one reference image.
[072] The at least one characteristic used to compare at least one image with at least one reference image may be multi-layer area values of said 3D image. The at least one characteristic can also be values selected from the group of: topographic profile of the animal, height of the animal, width of the animal, contour line or height profile along the animal's backbone, length of the back, contour plots for different heights of the animal, size of cavities, depth of cavities, the distance between two pre-selected points or features on the animal, angles between lines determined between predetermined points or features of the animal, vertical height profile(s) at different pre-selected points selected. Examples of using data extracted from images are described in Example 2, one or more of these data types can be used in conjunction with other data types mentioned here, as well as more data types extracted directly from the images or calculated from data extracted from the images and the type and number of data can be determined due to the number of animals and due to the animal species and/or breed in a herd.
[073]The height of the animal could be the average height of the contour line along the backbone or it could be the height at the legs, for example the average height at the legs or it could be the height at the head of the tail. Back length can be determined as the length at a height of 90% of the animal's total height, for example, for an animal with a maximum height of 165 cm, the back length is determined at a height of 148.5 cm. The animal's amplitude can be determined as the amplitude between two pre-selected points. A contour line length along the backbone can be determined as the distance from the neck to the head of the tail. A vertical height profile can be determined along the length of the backbone. When determining contour plots for different heights of an animal, the area of the animal's back at certain heights is determined, for example, the % height at 166-170 cm, the % height at 161-165 cm, the % height at 156-160 cm, height % at 151-155 cm, height % at 146-150 cm, etc., to obtain a group of areas for the animal. The height described may change due to the actual height of an animal to be identified from an identified animal. Examples of contour plots are presented in Example 2.
[074]When comparing image data to determine an animal's identity, this can be accomplished by comparing "masks" on the animal's back with corresponding "masks" on the back of animals in reference images. A "mask" may include the animal's back and, optionally, also the animal's neck and head. An animal's back "mask" is data that describes the topology of the animal's back and can be visualized as shown in Fig. 5.
[075]Preselected points can be selected from the group of right hip, left hip, right shoulder, left shoulder, tail head, neck, (1) left front rib, (2) left short rib start, (3) ) beginning of the left scapula, (4) anterior midpoint of the left scapula; (5) left scapula, (6) midpoint of left scapula, (7) tip of left scapula, (8) left hip joint, (9) left ischion, (10) nadir of left tail head, (11) articulation (12) tail head joint, (13) right tail head joint, (14) right tail head nadir, (15) right ischion, (16) right hip joint, (17) scapula tip right scapula, (18) right scapula midpoint, (19) right scapula, (20) right scapula anterior midpoint, (21) right scapula beginning, (22) right rib beginning, and (23) right front rib. The numbers indicated correspond to the numbers in Fig. 2. The location of these points and/or their height, for example above ground level, can themselves be given for image comparison, but preferably these points are used to calculate the distances between them. , to calculate angles between different lines between different points, to determine location of longitudinal and/or vertical height profiles, etc.
[076] The features to be used when comparing at least one image with at least one reference image can be any feature that is measurable and/or detectable. Preferably, the trait is a natural trait of the animal, as a part of the animal's phenotype, although wounds and/or scars may also be used as a trait. The characteristic is preferably not a tag applied to the animal by humans, such as a tag, for example an ID tag. The phenotype traits include the traits mentioned above and can also be skin color, color pattern, cavity location, cavity depth, and/or cavity areas.
[077]When comparing to at least one characteristic or data obtained from at least one image, this can be performed as a sequential identification procedure sequentially comparing a single characteristic of an unidentified animal with a corresponding characteristic of identified animals.
[078] A sequential identification procedure can be comparing a first characteristic, for example, the animal height obtained from an image of an unidentified animal with a corresponding first characteristic from images of identified animals, that is, from reference images , by this means closed on the identified animals that meet the characteristic (= a first closed population) and then proceeds to a second characteristic, for example, the length of the back of the unidentified animal, which is compared with the second characteristic of the identified animals of the closed population, which ends in that population until a second closed population. This procedure is continued with other traits until a combination of the unidentified animal and a single identified animal is obtained. The final combination of the unidentified animal with a single identified animal indicates that the unidentified animal corresponds to the identified animal and, therefore, the unidentified animal is identified.
[079] The comparison of the image with the reference image can also be performed by comparing feature vectors obtained from the at least one image with the corresponding feature vectors obtained from the at least one reference image. A feature vector can be based on at least two of the features described here.
[080] When comparing to at least one feature or data obtained from at least one image, this can also be accomplished by calculating a value for each image where that value is determined from a series of data. The value can be a dot product between vectors, as for example as described in Example 2.
[081] The at least one image and the reference image of the animal's back can be obtained within an angle between 0 and 50 degrees above the animal, where 0 is in a direct direction above the central part of the animal's back, such as as in a straight line above the animal's backbone. Preferably, the angle is between 0 and 40°, more preferably between 0 and 30°.
[082]By obtaining at least one image and/or at least one reference image within an angle other than 0, the system can automatically correlate to the deformation within the images and/or the comparison of at least one image can be performed with at least one reference image obtained from substantially the same angle measured according to any line drawn through the animal. Substantially the same angle can be a deviation of ± 5°, such as ± 4°, for example. ±3°. Preferred is ±2°, more preferred is a deviation of ±1°.
[083]The at least one reference image of the back of an unidentified animal is preferably obtained with only one animal present in an area covered by a reference imaging unit that provides at least one reference image of the animal's back.
[084]A trigger mechanism can be located near the reference imaging unit. The trigger mechanism can be located such that when an animal is activating the trigger mechanism, the trigger is triggered and sends a signal to the reference imaging unit to take at least an image of the animal's back. For example, a detector could be mounted on a gate that triggers when the cow makes contact with the gate.
[085] At least one image of the back of an unidentified animal may be obtained with one or more animals present in an area covered by an imaging unit to obtain images of the back of at least one unidentified animal. The system is preferably capable of distinguishing different animals from each other in an image, that is, when an image covers more than one animal, each of these animals can preferably be identified.
[086]The method as described here can be used to identify any type of animal. Preferably, the animal is selected from the group of bovines, cows, dairy cows, bulls, calves, pigs, sows, wild boars, barrows, piglets, horses, sheep, goats, deer.
[087]The animal can also be one or more animals that live in a zoo, a park, or in the wild. Such animals can be elephants, monkeys, giraffes, hippos, rhinos, wolves, foxes, bears, tigers, lions, cheetahs, pandas, leopards, tapirs, llamas, camels, reindeer, okapis, antelopes, wildebeests.
[088]The method of identifying an animal can be used to verify whether the identified animal is still among the population or may be dead. The method can also be used for additional analyzes as described herein, such as estimating the health or well-being of the animal, or be combined with other methods for estimating the animal's food intake, such as a system for determining fur food consumption. at least one animal as described in WO 2014/166498.
[089]The recorded health conditions can be used to assess different conditions, such as:• the physiological state of the animal, including detectable body scoring elements in images obtained from above the animal, i.e. the animal's back, neck and the head, • the general state of health of the animal, • state of reproduction, ie whether the animal, such as a cow, is ready to be inseminated/fertilized; this can be predicted from eating behavior such as reduced food consumption (combined with good health to ensure the animal is not sick),• behavior such as eating behavior eg how long the animal is in a feeding channel (in loose housing systems the feeding channel can be a virtual feeding channel as the animal can select different places to eat), how long is the animal actually eating, how often is the animal eating , how much the animal eats, when it eats and how much the animal eats per day, • indications of illness such as reductions and/or changes in feed consumption and/or feeding behavior.
[090]Another aspect of the invention relates to a system for determining the identity of an individual animal from the appearance and/or topology of the animal's back, the system comprising•A reference imaging unit to provide reference images of fur at least one identified animal, where the reference imaging unit comprises at least one identity determination device for determining the identity of the identified animal, at least one camera for obtaining at least one image of the back of the unidentified animal, at least one base of data to store at least identity information of at least one identified animal and at least one image of the identified animal's back, and data transmission means for transmitting data from the identity determination device and camera to the database,•A imaging unit to obtain at least one image of the back of at least one unidentified animal, where the imaging unit nt is connected to the database for transmitting data from the imaging unit to the database, and •At least one processing medium connected to the database to compare at least one image of an unidentified animal with at least one reference image and based on this comparison, determines whether the unidentified animal matches the identified animal.
[091] The image obtained by the system can be a 3D image and also the reference image can be a 3D image and therefore a 3D reference image.
[092]The system's imaging unit may include at least two cameras. These two cameras can be located at any distance from each other, enabling coverage of areas of interest. Preferably, the at least two cameras are located at mutual distances of within 15M, such as within 10M, e.g. within 5M of each other to simultaneously obtain at least one image by each camera of the back of at least 5M. least one unidentified animal, where the at least two cameras are database connection for transmitting data from the cameras to the database and where the database builds at least one 3D image of the animal based on the fur images least two cameras.
[093]The at least one camera of the reference imaging unit and the imaging unit may be one or more cameras selected from the group of range cameras, stereo cameras, time-of-flight cameras. Preferably, the reference imaging unit and the imaging unit comprise cameras of the same type.
[094]The reference imaging unit and/or the imaging unit may comprise at least a range camera with a depth sensor and a 2D camera, such as an RGB camera. The reference imaging unit and/or the imaging unit may also include at least one time-of-flight camera.
[095] Preferably, the reference imaging unit and the system imaging unit are configured to acquire topographical images.
[096] The system can be configured so that the reference imaging unit's camera is activated to take an image of the animals back when an animal is close to the identity determination device and the animal's identity has been recorded. A triggering mechanism as described elsewhere may be a part of the system.
[097]The system can also include ID tags. These ID tags can be linked to the animals to be identified. ID tags can be visual and/or electronic ID tags. Electronic ID tags can be electronic ear tags and/or electronic ID tags attached to an animal, such as on a collar. A single animal may be tagged with one or more ID tags, such as at least one visual ID tag and or at least one electronic ID tag. An example is at least one visual ear ID tag combined with at least one electronic ID tag on a collar. Another example is at least one visual ear ID tag combined with at least one electronic ear ID.
[098] The system may also comprise an identity determination device, such as a camera suitable for taking images of visual ID tags. The identity determining device may also comprise an ID reader capable of recording an animal identity based on electronic identity markers located on the animal or an animal.
[099] The system comprises a database that can store multiple reference images of a single animal. The database can store multiple reference images of a single animal each day. Such reference images can be taken at different time intervals over a period of one day, two days, three days, four days, five days, six days, a week or at longer intervals. The time between obtaining reference images of an animal can be determined such that each time the animal is in an area of an identity determining device, the system determines the identity of the animal and obtains at least one image of reference of the animal's back. The system may store reference images and/or other images of an animal, for example, for the entire life of the animal or for as long as the animal is kept on site, for example on the farm where the images are taken. Images can also be stored much longer and can be used as statistical data for different purposes, such as evaluating feeding types, feeding and breeding methods, eg value of specific crosses or values of specific male animals.
[0100]The system as described here can also be used to monitor individual animals, such as in relation to health status and disease risk. This monitoring can be based on any observed body changes, for example from day to day, or by comparing data obtained from several days, such as two days, three days or more. The system can automatically monitor each animal in a population, and certain threshold values based on changes in recorded information can be included in the system, so that an alarm message or information is created by the system when an animal's body changes too much within the system. of a certain period of time.
[0101] Preferably, the database stores at least reference images of a single animal for at least one month, such as at least two months, such as at least half a year, for example, at least one year. Preferably, the database stores at least reference images of a single animal until that animal is no longer within the animal population or is no longer present, for example due to being sold or killed.
[0102] The system comprises processing means that can select features from at least one image and the at least one reference image before comparing those features. Examples of feature types are described elsewhere here. The system's processing means can compare features of at least one image with features of at least one reference image by any known comparison method.
[0103]To compare traits, processing means may use a method in which an animal's predefined trait vectors for pre-selected distances calculated from the ground or ground are compared. By comparing at least one feature of at least one image with at least one corresponding feature of at least one reference image, the processing means can determine and compare areas of 3D image layers. Such areas may be part of feature vectors or they may constitute features for, for example, sequential comparison of at least one image with at least one reference image.
[0104]By establishing image features, that is, of at least one image of an unidentified animal, and these at least one image are in fact two or more images, these images can be obtained within a short period of time, as in less than 20 seconds, for example, in less than 10 seconds, as in less than 5 seconds, for example, in less than 3 seconds, as in less than 2 seconds. For such a series of images, a characteristic can be established based on a single image or it can be an average based on two or more images in the series.
[0105] When establishing reference image characteristics, i.e. from at least one reference image of an identified animal, these characteristics can be established from one or more series images of an identified animal and in a manner as described for images of unidentified animals.
[0106]Areas of layers of an animal can be determined for layers with a pre-selected planar distance. Such a preselected plane distance may be about 8 cm, such as about 7 cm, for example, about 5 cm, such as about 4 cm, for example, about 3 cm from a fixed point. default. Preferably, a preselected plane distance is about 5 cm. Hereby, the processing means can make a calculation of the area of an animal, such as the area of the back in horizontal planes with mutual distances from the pre-selected planar distance, for example 5 cm. Such layered areas may constitute features for comparing at least one image with at least one reference image.
[0107]Layer areas can also be used to determine the percentage of an animal above a pre-selected level. Different areas of the animal's back determined at pre-selected distances from the plane and calculated as percentages from a pre-selected level may constitute features for comparing at least one image with at least one reference image. Example: A pre-selected level can be 135 cm above ground level and at this level the area of a horizontal plane of the animal is calculated. A pre-selected flat distance can be 5 cm and the area of these levels, i.e. at 140 cm, 145 cm, 150 cm, 155 cm, etc. above ground level can be determined. Areas can be converted to percentages with respect to the area at the pre-selected level, i.e. in this example at 135 cm, and those percentages can constitute features for comparing at least one image with at least one reference image.
[0108]Determining the features to use when comparing at least one image with at least one reference image may be based on flat areas as described above and may be performed for pre-selected distances calculated from the ground or ground . Such pre-selected distances may be selected due to the height of the animal species, animal breed and/or animal type that must be identified. A pre-selected distance for animals with maximum height, for example. 180 cm can be from 140 to 180 cm and can be combined with a pre-selected flat distance, e.g. 5 cm in such a way that the areas of the animals or the back of the animals are determined for distances of 140 cm, 145 cm, 150 cm, 155 cm, 160 cm, 165 cm, 170 cm, 175 cm and 180 cm above ground level. Such areas can be used as exact numbers and/or as a percentage of the area at a pre-selected level, for example 140 cm above ground level and can be used as features to compare at least one image with at least one image of reference.
[0109]Instead of determining the areas in different planes, the planes can be assumed as a ground level to determine the volume of the animal's back from ground level i.e. the volume of the animal above different heights of the animal. Each plane, e.g. 120 cm, 125 cm, 130 cm, etc. above ground level can have its own ground level, and for each of these ground levels the volume above this ground level can be determined. One or more of these volumes can be used as a feature to compare at least one image with at least one reference image. Plans to determine volumes of animal hindquarters above plans may be selected due to maximum and average height and/or size of animal species, breed, type, etc. to be identified.
[0110]Reference images can be acquired in a location where the cows are well-positioned in relation to a 3D camera under which each cow in the herd passes one or more times a day. At this location, each cow's RFID tag is read, so the cow ID and 3D images can be paired. Over time, a large library with images of all the cows is built up. This image library can be used to identify cows from images of cow origins acquired elsewhere on the farm. The library can also be used to track the health status of each cow over time.
[0111]When determining the identity of an animal by comparing at least one feature from at least one image with at least one corresponding feature from at least one reference image, the process of determining the identity of an animal can be performed sequentially, for example , by first comparing gross or general features obtained from the image and reference images, and hereby classifying reference images that do not meet the general characteristics. The second comparison can be performed based on other less general and/or more specific characteristics obtained from the image and reference images. A third, fourth, etc. comparison of at least one feature obtained from at least one image may be compared with at least one corresponding feature obtained from at least one reference image until a match is obtained between the at least one image and the at least one image. reference image where the at least one reference image are images of a single animal.
[0112]An example of carrying out a sequential determination of an animal based on the invention as described herein may comprise comparing characteristics determined in at least one image with the corresponding characteristics determined in at least one reference image: 1o comparing: Height of the animal (Q),2nd comparing: Skin color pattern (U),3rd comparing: Length of the back (V),4th comparing: Contour line along the spine (W),5th comparing: distance between two pre- selected, for example, distance between the hind hips (X),6o comparing: location and/or sizes and/or depth of the cavities (Y),7o comparing: contour plots or flat areas for different planes of the animal (Z), 8o comparing volumes above selected animal planes.
[0113]The example described with sequentially determining the identity of an animal may include any suitable feature and be performed in any suitable order until all tested features obtained from at least one image of an unidentified animal match all of the corresponding characteristics obtained from the at least one reference image of an identified animal, and where the at least one reference image of an identified animal, if more than one reference image, all reference images are of the same individual.
[0114]Determining the identity of an animal can also be performed by comparing trait vectors. In the example above that indicates 7 comparisons in a sequential determination, the characteristics are indicated by a letter, each of these letters can correspond to a group of characteristics each comprising different possibilities, for example, for the height of the animal Q1 is different from Q2. A trait vector can thus comprise at least one trait from each trait group and these trait vectors can be compared to determine the identity of an animal.
[0115]As an example of comparing trait vectors and unidentified animals might have a trait vector of [Q, U, V, W, X, Y, Z] and presuming that only two possibilities exist within each group of features, a feature vector comparison can be performed as indicated below, where only a limited number of possible feature combinations are shown in feature vectors:
[0116]Feature vector obtained for unidentified animals: [Q1, U2, V1, W2, X1, Y2, Z1]
[0117]Feature vector obtained for the identified animal No. 1: [Q1, U1, V1, W2, X1, Y2, Z1]
[0118] Characteristic vector obtained for the identified animal No. 2: [Q1, U1, V2, W1, X2, Y1, Z2]
[0119] Characteristic vector obtained for the identified animal No. 3: [Q1, U1, V1, W2, X1, Y2, Z2]
[0120] Characteristic vector obtained for the identified animal No. 4: [Q1, U2, V2, W1, X2, Y1, Z2]
[0121] Characteristic vector obtained for the identified animal No. 5: [Q1, U2, V1, W2, X1, Y2, Z1]
[0122]Feature vector obtained for the identified animal No. 6: [Q2, U1, V2, W1, X2, Y1, Z1]
[0123]Feature vector obtained for the identified animal No. 7: [Q2, U1, V1, W2, X1, Y2, Z1]
[0124]Feature vector obtained for the identified animal No. 8: [Q2, U1, V2, W1, X2, Y1, Z2]
[0125]Feature vector obtained for the identified animal No. 9: [Q2, U2, V1, W2, X1, Y2, Z1]
[0126] Characteristic vector obtained for the identified animal No. 10: [Q2, U2, V2, W1, X2, Y1, Z2]
[0127]When comparing the trait vectors, the only match between the trait vector for the unidentified animal matches the trait vector for animal No. 5, it can be concluded that the unidentified animal is animal No. 5. Performing a sequential comparison with the traits mentioned in the trait vectors, the 1st comparison based on trait Q will match animal No. 1, 2, 3, 4 and 5, which are used for the next comparison. The 2nd comparison based on trait U will match animal No. 4 and 5, and of these, the 3rd comparison based on trait V will match only animal No. 5.
[0128]When an unidentified animal is identified as described herein, the system of the invention can itself be used to obtain different types of information for identified animals, the system can also be extended to provide additional information that can be stored together with the identity of an animal identified according to the method as described herein.
[0129]The comparison can also be performed using a neural network implemented as a deep learning system. Both neural networks and deep learning processes are known to be skilled in the art of image processing. For example: A cow and its orientation in the image can be found using pattern matching techniques, which are also known in the art. Once an unknown cow appears in the image, features such as height, color patterns, back length, backbone height contour, distances between pre-selected points, cavities, areas at various heights and volumes above these areas can be calculated. . A supervised or unsupervised neural network that has been trained on a large number of reference images of each cow in the herd can then be applied. The trained neural network can then identify the unknown cow by comparing the unknown cow with images from the library of all the cows.
[0130]The system may comprise means for determining the food consumption of at least one of said animals. Such means may comprise • a feeding area imaging unit to provide images of a feeding area and • processing means configured to assess the amount of food consumed by each identified animal determining the reduction in feeding in subsequent images of the feeding area in front of each identified animal.
[0131]Processes of determining feed intake or reducing feed in a feeding area based on comparing the amount of feed in subsequent images of the feeding area are described in WO2014/166498 ("System for determining feed consumption of at least one animal", Viking Genetics FMBA,).
[0132]The feeding area imaging unit may be the imaging unit for obtaining at least one image of the back of at least one unidentified animal such that the imaging unit obtains images of the back of at least one unidentified animal identified as well as a feeding area and where at least one unidentified animal is able to eat food from the feeding area. Preferably, at least one image covers the back of at least one unidentified animal along with a feeding area in front of that unidentified animal.
[0133]The system can determine the power consumption of at least two images from the same feed area and where the feed reduction is calculated as the difference in feed volume within a feed area established from at least two images .
[0134]The system imaging unit can be configured to continuously image at least a portion of a feed area. It is also possible to have an imaging unit that is configured to image an area including a feed area at predefined and/or selected time points.
[0135] The at least one camera of the system can be articulated around at least one axis, making it possible to adjust the at least one camera in different directions to obtain at least one image of at least one animal or of at least one animal and the feeding area in front of at least one animal.
[0136] The system may also comprise at least one camera rail and/or camera wire for positioning at least one camera with respect to the at least one animal and/or a feeding area in front of the at least one animal. The tracks and/or wires can be suspended or stretched above an area where the animals to be identified remain and this can be an indoor area and/or an outdoor area.
[0137]The system may also comprise at least one drone, the drone being connected to at least one camera and the drone being capable of flying above at least one animal to allow the at least one camera to obtain at least an image of the fur least one animal. The at least one camera on the drone can be fixed or pivotable. A pivotable camera can be rotated due to the input of the camera's position, obtaining information about the location of the animals. The location information of the animals can be based on signals from at least one electronic ID tag on an animal and/or may be based on signals obtained from an infrared camera capable of detecting live animals.
[0138]A drone can be used inside an animal protection shed or stable and/or can be used outdoors in areas where animals to be identified can be located such as in the field and/or in an enclosure. A drone can be used to image unidentified animals, and at other times it can be used to obtain reference images of animals, also obtaining animal information from at least one electronic ID tag.
[0139]A drone when used outdoors together with the invention described herein can be used for different purposes, such as identifying, for example, dairy cows in plantation systems, to determine the health status of an animal, etc. Detailed description of figures
[0140] Fig. 1 illustrates cows that are eating in a barn (1) in which a system of the present invention is installed. Cameras (4) mounted above the cows (3) take images of the backs of the cows and forward these images to a database and processing unit (6). Cows are tagged by ID tags such as ear tags (5), however if cows are walking freely in the stable it may not be possible to identify cows from ID tags. The system can be configured to image the back of the cows as well as the feed (2) in front of the cows. From the images obtained, it is possible to identify each cow and estimate the amount of food consumption for each of these cows.
[0141] Fig. 2 illustrates examples of different preselected points on the back of a cow. Such pre-selected points can be used to extract additional information from the images, such as lengths between different points, line angles between different points, etc.
[0142] Fig. 3 illustrates examples of data or characteristics established in relation to the back of an animal, here the back of a cow. The data or characteristics indicated are:• total area of the cow's back that is located above 70% of the maximum height of the cow (large ellipse indicated by a dotted line), • total area of the cow's back that is located above 90% of the maximum height of the cow (two small ellipses within the large ellipse), • length of a profile along the spine at a height greater than 70% of the maximum height of the cow (illustrated by a dotted line in the longitudinal direction of the cow's neck to the head of the tail)• distance between the hip bones at their maximum height (illustrated by a thick vertical line through the small ellipse on the cow's back),• body width greater than 70% of the maximum height of the cow, eg 7 locations along the cow's body (illustrated by thin vertical lines within the large ellipse), • color pattern, if any (not illustrated).
[0143] Fig. 4 illustrates the height profile along the backbone of two cows from the head of the tail (left part of the graph) to the neck (right part of the graph) of a cow slightly over 1.6 M (Fig. 4A) and a cow that is about 1.7 M (Fig. 4B).
[0144] Fig. 5 illustrates a 3D Imaging Mesa reconstruction of part of a cow with a height above 90 cm from ground level.
[0145] Fig. 6 illustrates the back of a cow.
[0146] Fig. 7 illustrates the back of the cow in Fig. 6 with indications of some data/characteristics that can be used in the analysis. Steps 1-6 are further explained in Example 2 and represent:• 1: Length of the backbone and a height profile along the backbone of the cow ie a longitudinal height profile.• 2: Contour line of the cow at a predetermined height of 90 cm from the ground.• 3: Contour plane for pixels located above the height of the cow corresponding to the 80% quantile height subtracted 8 cm.• 4: Contour plane for pixels located above a height corresponding to the 80% quantile height subtracted 2 cm.• 5: An arbitrary triangle made based on the location of the left and right hip bones and the head of the tail, where, for example, the angle at the head of the tail can be determined.• 6: The maximum width in the transverse direction of the cow at the place where the cow is widest and along this line a height profile can be determined, ie a transverse height profile
[0147] Fig. 8 illustrates the determination of the area based on resized data obtained from the part of a cow with a height above 90 cm from the ground. Areas below the graphs (and for example above 90 cm of line) can be determined.
[0148] Fig. 9 and 10 illustrate different thickness profiles and height profiles at predetermined heights of two cows. In each figure, data is scaled to 100 pixels (= X axis) and thickness measured in pixels (= Y axis) or height above ground measured in cm (= Y axis). The left end of the graph corresponds to the neck region and the right end of the graph corresponds to the tail region.• Fig. 9A and 10A: Thickness profile for a cow measured 90 cm above ground level. Each axis indicates pixels.• Fig. 9B and 10B: Thickness profile for a cow measured along the line indicated by step 3 in Fig. 7 that is, at a cow height corresponding to the 80% quantile height subtracted 8 cm.• Fig. 9C and 10C: Thickness profile for a cow measured along the line indicated by step 4 in Fig. 7 that is, at a cow height corresponding to the height of 80% amount subtracted 2 cm.•Fig. 9D and 10D: Longitudinal height profile along the backbone of a cow. The X axis indicates pixels, the Y axis indicates cm from the ground.
[0149] Fig. 11 illustrates a cross-sectional height profile of a cow in the position where the cow was thickest (measured above 90 cm from the ground). Data is resized to 40 pixels. The X axis indicates pixels, the Y axis indicates cm from the ground.
[0150] Fig. 12 illustrates the determination of a cow based on the neural network, as a deep learning system. A series of characteristics of a cow to be identified are entered into the system and an output is obtained with the estimated and ranked probability of different matches. Example 1
[0151]The method was developed by testing whether a number of Jersey and Holstein cows could be determined/identified from each other based on images of their hindquarters. On a Danish farm with dairy cows, three-dimensional images of the backs of cows were provided. The imaging system included a 3D camera (Swiss Ranger 4500 from Mesa Imaging, Switzerland, which is an IP 67 camera suitable for dusty and humid rooms). In parallel with the 3D camera, two black and white Basler industrial cameras were mounted. The cameras were mounted 4.5 meters above ground level. The distance from the camera to the upper back of the cows was approximately 2.7-3 meters depending on the height of the cows. The back images of the cows were taken when the cows were on their way to the milking station and in a position where the cows walked one after the other. The images were obtained with only one cow in each image. From the contour plots the 3D images obtained were performed as further described in Example 2, albeit at 148 cm, 153 cm, 158 cm, 165 cm and 172 cm above ground level. The area of the cow's back within each of the contour plots at the indicated heights was determined. Based on the area within the aforementioned contour plots, the 16 cows were easily identified without mixing up the identities. In this experiment to test whether the cows could really be identified from the images, the cows were also identified by different visible marks painted on the back of each cow. These marks were used only to verify that the identification based on the other characteristics was correct.
[0152] Fig. 4 illustrates other features that can be used when identifying animals. The figure illustrates a contour line along the backbone. The position of the backbone is illustrated in Fig. 7.
[0153]Fig. 4A: Height profile in the longitudinal direction of the cow along the backbone of a short cow.
[0154]Fig 4B: Height profile in the longitudinal direction of the cow along the backbone from a height.
[0155]Both the length of the backbone and the height profile along the backbones can be used as characteristics when identifying animals as cows as explained in Example 2.
[0156] In the experiment, about 6 images of each cow were obtained with about 1 second between each exposure. The analysis of each image as described above and the comparison of data obtained from the images for each cow and between cows clearly showed much less variation for the images of one cow than between different cows. Example 2
[0157]The identification method was tested in another experiment with dairy cows from the Jersey race. Three-dimensional images of the back of the cows were provided with a system including a 3D TOF (time of flight) camera (Swiss Ranger 4500 from Mesa Imaging, Switzerland). Two black and white Basler industrial cameras were also used. The three cameras were connected to a computer allowing images to be stored and analyzed. The 3D camera was located 3.2M above the ground at the entrance to the milking station and where the aisle is about 1M wide. On a wall along the aisle, an ID reader was located to obtain a signal. of the ear tag each time a cow passed the ID reader. A trigger signal was sent to the computer each time a cow passed the ID reader. The trigger signal induced the computer to store an image from each of the three cameras with 0.5 second between exposures. The ID reader also stored the cow ID obtained from the ear tag, and these IDs were only used to verify the identification method developed based only on the cow's back images. The two black and white cameras were only used to take images to see the cows and the environment to check if anything looked strange. The images from the black and white cameras were not used for the identification process.
[0158] Fig 5 is a 3D Imaging Mesa reconstruction of part of a cow with a height above 90 cm from ground level. The same cow data is shown on a 3D contour height plot in Fig. 6. For each 3D image obtained, the images were analyzed in different steps to obtain data and PCA scores to calculate a vector for each cow. Fig. 7 indicates where on the back of the cow the data were obtained. The steps in the analysis are described below and indicated in Fig. 7:a)Step 1: Calculation of a height profile in the longitudinal direction of the cow along the backbone. A curve was calculated to describe the height profile along the backbone from the "head to tail" to the "neck point", where these end positions in this measurement were determined by the point where the body thickness was less than 38%. of the widest width the cow. b)Step 2a (indicated as Step 2 in Fig. 7): Determination of a contour line of the cow at a predetermined height of 90 cm from the ground. The cow contour line was determined by the same length as for the height profile in step 1 ie from the "neck point" to the "head to tail". The area within this contour line was determined as the area below the "height" graph and above 90 cm in Fig. 8 as described further in Step 2b.c)Step 2b - Further analysis of the data from Step 2a: Height distribution in image pixels located within the 90 cm contour line. Different distributions are illustrated as graphs in Fig. 8, where all pixels within the 90 cm contour line are sorted according to the corresponding height of the cow and this is shown as a function of the percentage of pixels corresponding to the height of the cow between 90 cm and a predetermined height above 90 cm or the total height of the cow. In Fig. 8, this distribution or area determination is shown for a cow with a maximum height of 130 cm indicated by the "height" graph, where the graph illustrates the percentage of pixels below a certain height of the cow, but above 90 cm from the ground . It can be seen that about 40 percent of the pixels (in the range above 90 cm) are located below 120 cm.d)Step 2c - Further analysis of the data from Step 2b: From the height distribution determined in step 2b , a quantile height of 80% was determined to be 80% of the maximum height of the cow. This graph is shown as "80%". The maximum height of the cow was determined as an average of the value of the 50 pixels indicating the highest locations of the cow. In the example with the data in Fig. 8, the maximum height is 130 cm and the 80% quantile is 104 cm. The area below the "80%" graph and above 90 cm was determined.e)Step 3: Determination of a contour plane delimitation for the pixels located above the cow height corresponding to the 80% height quantile subtracted 8 cm . The area within this contour line has been determined. In the example with the data in Fig. 8, the contour plane is determined at a cow height of 104 cm - 8 cm = 96 cm. The area is determined as the area below the graph "80%" - 8 cm' and above 90 cm.f)Step 4: Determining a delimitation of a contour plane for the pixels located above the height of the cow corresponding to the height 80% quantile subtracted 2 cm. The area within this contour line has been determined. In the example with the data in Fig. 8, the contour plane is determined at a cow height of 104 cm - 2 cm = 102 cm. The area is determined as the area below the graph "80%" - 2 cm' and above 90 cm.g)Step 5: Determination of the points on the images corresponding to the location of the external part of the hip bones that was defined as the location in the image where the contour plane determined in step 3 is wider. A virtual arbitrary triangle was made based on the location of the left and right hip bones and the head of the tail as determined in step 1 and in this triangle the angle at the head of the tail was determined as well as the distance between the left hip bones and right.h)Step 6: Determination of the maximum width in the transverse direction of the cow and at the place where the cow is widest and calculation of a height profile along the maximum width, ie a transverse height profile. Data analysis
[0159]The data obtained as described in the eight items above were converted into data, enabling statistical analysis.
[0160]The contour planes determined in steps 2a (90 cm height), 2c (80% quantile height) and 4 (80% quantile height minus 2 cm) were transformed into thickness profiles. different lengths between cows as the length of the cows differs and therefore the thickness profile of each cow was scaled to a fixed length of 100 pixels. Similarly, the longitudinal height profile from step 1 was scaled to a fixed length of 100 pixels. The cross height profile from step 6 has been resized to a fixed length of 40 pixels. The scaling was performed as a simple ratio calculation based on the actual length or width of the cow and a length of 100 (or 40 if 40 pixels is the scaling dimension) through this a Zn value for a cow of length 80 cm is scaled by (Zn/ 80) x100 = 1.25Zn or if Zm is for a cow with a length of 115 cm the Zm value is scaled to (Zm/ 115) x100 = 0.87Zm.
[0161]The entire dataset for each image at this stage comprised 449 variables: 1.The determined area within the 90 cm contour line as described in step 2a (1 variable)2.The determined area within the delimited contour line by the 80% quantile height as described in step 3 (1 variable)3.The determined area within the contour line delimited by the 80% quantile height minus 2 cm as described in step 4 (1 variable)4.The quantile height of 80% (1 variable)5.The angle between the lines from the head of the tail to the right and left hip bone as described in step 5 (1 variable)6.The maximum width as described in step 6 (1 variable)7.O contour line length determined at cow height of 90 cm as described in step 2a (and step 1) (1 variable)8. The contour length delimited by the 80% quantile height as described in step 3 (1 variable)9 .The length of the contour delimited by the quantile height of 80% minus 2 cm as described in step 4 (1 variable l)10. Thickness profiles at cow height of 90 cm are scaled to 100 pixels (100 variables) and illustrated in Fig. 9A and 10A.11. Thickness profiles at cow height are determined at the 80% height quantile as described in step 3 and scaled to 100 pixels (100 variables) and illustrated in Fig. 9B and 10B.12. Thickness profiles at cow height are determined at the height quantile of 80% minus 2 cm as described in step 4 and scaled to 100 pixels (100 variables) and illustrated in Fig. 9C and 10C.13. The height profile in the longitudinal direction as described in step 1 and scaled to 100 pixels (100 variables) and illustrated in Fig. 9D and 10D.14. The height profile along the maximum width as described in step 6 and resized to 40 pixels (40 variables) and illustrated in Fig. 11.
[0162]To further compress the data, a 6 PCA model (PCA = principal component analysis) was developed with up to 15 principal components (PC scores) for each dataset (set of characteristics) with the following combination of data and where the variable number refers to the list above:a)Variable 1 to 9 (9 PC scores)b)Variable 7 + 10 (15 PC scores)c)Variable 8 + 11 (15 PC scores) d)Variable 9 + 12 (15 PC scores)e)Variable 10 + 13 (15 PC scores)f)Variable 11 + 14 (15 PC scores)
[0163] Those skilled in the art know how to perform a principal component analysis, and this will not be described.
[0164]The original lengths of the curves were included in the calculation of PC scores, so knowledge of the length of the individual cow was maintained.
[0165]With the PC scores, a total of 449 variables were reduced to 85 variables.Identification of the individual cow
[0166]The sequence of numbers, i.e. the PC scores for a cow to be identified was compared to the average PC characteristic of each of the cows in the herd. A cow was identified when the mean trait PC for that cow resembled a mean trait PC calculated for one cow more than it resembled mean trait PCs calculated for the other cows in the herd. In practice, the calculation was performed by creating the dot product between each mean vector Xk for each cow 'k' in the herd and the vector Xu for the unidentified cow, i.e. the cow to be identified:
where is the angle between the two vectors
are the length of each of the vectors. If the vector for an unidentified cow resembles a vector for a cow in the herd then .:. 2 will be close to +1(plus 1), whereas these two cows are very different .> • /... will be close to -1 (minus 1).
[0167]The model shown for analysis is very simple and overfitting is almost unlikely. The model can be extended and improved by continuing as more photos are taken for each cow. It is also simple to identify deficient images and avoid using them to identify a cow or to extend the calculation of an average vector for each of the cows.
[0168] The method described above was tested with 9 main components for characteristics indicated in item a) in the list above and 15, 14, 13, 12, 11, 10, 9, 8, 7 or 6 main components for each of the characteristics remaining indicated in item b) to f) in the list above. The best result was obtained using 9 scores for features in item a) and 7 scores for each of features in item b) to f).
[0169]The analysis as described in example 2 was performed for about 5 images for each cow representing a total of 27 cows, for a total of 137 images. The images representing a cow were obtained at different times of the day and on different days. Of the 137 images, 116 were immediately correctly connected to the right cow by using 9 scores for the features in item a) and 7 scores for each of the features indicated in items b) to f) in the list above. By averaging the 5-6 images taken for each cow, although taken on different days, the identification of all cows was correct. Extending the analysis to be based on more features taken from the images and/or features taken from more than one image of a cow where the images are taken, for example, with a very short period of time, for example, from 0.1-1, eg 0.5 second would ensure correct identification.More Details1.A method for determining the identity of an individual animal from the natural appearance and/or topology of the back of said animal, said method comprising• obtaining at least one image of the back of an unidentified animal,• extracting data from said at least one image obtained, said extracted data relating to the natural appearance and/or topology of the animal's back,• comparing said data extracted from at least one image of an unidentified animal with reference data extracted from at least one reference image of a back of an identified animal where the identity information of the identified animal is linked to the referenced at least us a reference image and • determine on the basis of said comparison whether said unidentified animal corresponds to said identified animal.2. The method according to item 1 in which at least one reference image of the back of an identified animal is obtained at least once a month, it is preferable that the reference image is obtained at least every second day, more preferably the reference image is obtained at least once a day and/or said animal is selected from the group of bovines , cows, dairy cows, bulls, calves, pigs, sows, wild boars, geldings, piglets, horses, sheep, goats, deer. reference image of the back of an identified animal is obtained by • providing the identification number of an animal, hereby the animal being an identified animal, • providing at least one image of the back of said identified animal, and • storing in a database said identification number of the identified animal together with said at least one image of the back of said identified animal, said image being here a reference image.4. The method according to any one of items 1 to 3 wherein said image and said reference image are topographical images of the back of animals, such as 3D images, e.g. multiple layers of 3D images.5.The method according to any one of items 1 to 3 4 wherein said comparison of data extracted from said image with data extracted from said reference image is performed by comparing at least one feature and/or at least one feature vector obtained from said image with a corresponding feature and/or vector of features obtained from said reference images, such features and/or feature vector may comprise or be based on values from the area of multiple layers of said 3D image and/or selected values from the animal's topographic profile group, such as the height of the animal, the amplitude of the animal, the contour line along the animal's backbone, the length of the back, the contours of different heights of the animal, the volume of the animal above different heights of the animal, the size of the cavities, the depth of the cavities, the distance between two pre-selected points on the animal, where the pre-selected points can be selected from the hip group right, left hip, right shoulder, left shoulder, tail head, neck, (1) left front rib, (2) left short rib beginning, (3) left scapula beginning, (4) left scapula anterior midpoint ; (5) left scapula, (6) midpoint of left scapula, (7) tip of left scapula, (8) left hip joint, (9) left ischion, (10) nadir of left tail head, (11) articulation left tail head, (12) tail, (13) right tail head joint, (14) right tail head nadir, (15) right ischion, (16) right hip joint, (17) scapula tip right scapula, (18) right scapula midpoint, (19) right scapula, (20) right scapula anterior midpoint, (21) right scapula start, (22) right rib start, and (23). system for determining the identity of an individual animal from the natural appearance and/or topology of the back of said animal, said system comprising at least one camera for obtaining at least one image of the back of an unidentified animal, at least least one database or admission to at least one database to store data related to pe at least one reference image of the back of an identified animal and to store data relating to at least one image of the back of an unidentified animal, • data transmission means for transmitting data from said at least one camera to said database data, • at least one processing means connected to said database, said processing means being configured to compare data extracted from said at least one image of an unidentified animal with data extracted from at least one reference image where said extracted data are related to the natural appearance and/or topology of the animal's back and based on that comparison, determine whether said unidentified animal corresponds to said identified animal.• A system for determining the identity of an individual animal from of the natural appearance and/or topology of the back of said animal, said system comprising•A re-imaging unit for providing reference images of at least one identified animal, said reference imaging unit comprising at least one identity determining device for determining the identity of said identified animal, and at least one camera for obtaining at least one back image of said unidentified animal,iii .at least one database or entry into at least one database to store at least one identity information of at least one identified animal and at least one back image of said identified animal iv. data transmission means for transmitting data from said identity determining device and said camera to said database, • An imaging unit configured to obtain at least one image of the back of at least one animal unidentified, wherein said imaging unit is connected to said database for data transmission from said imaging unit to said database, • at least one processing means connected to said database, said processing means being configured to compare data extracted from said at least one image of an unidentified animal with data extracted from at least one reference image where said extracted data are related to the natural aspect and/or topology of the animal's back and based on this comparison, determine whether said unidentified animal corresponds to said identified animal. 8. The system according to item 7, wherein said image is a 3D image and said reference image is a 3D reference image and/or said at least one camera of said reference imaging unit and said imaging unit each is one or more cameras selected from the group of range cameras, stereo cameras, time-of-flight cameras, such as a range camera comprising a depth sensor and a 2D camera, such as a camera RGB and/or.9. The system according to any one of items 7 to 8 wherein said camera of said reference imaging unit is activated to obtain an image of the rear parts of animals when an animal is close to said device determination of identity and the identity of the animal was recorded.10.The system according to any of items 7 to 9 wherein said database stores multiple reference images of a single animal, such as multiple reference images 11. The system according to any one of items 7 to 10 wherein said processing means determines characteristic vectors of an animal for pre-selected distances calculated from the distance from the ground or from the floor and/or said feature vectors are areas of 3D image layers and/or the pre-selected distances are between 70 and 180 cm.12. The system according to any one of items 7 to 11 further comprising means to determine the feed consumption of at least one of said animals, such as • a feeding area imaging unit to provide images of a feeding area, • processing means configured to assess the amount of food consumed by each identified animal determining the reduction of feeding in subsequent images of the feeding area in front of each identified animal.13. The system according to item 12 in which said imaging unit d and the feeding area is said imaging unit for obtaining at least one image of the back of at least one unidentified animal such that said imaging unit obtains images of the back of at least one unidentified animal, as well as an area 14.The system according to any of items 12 to 13 wherein the power consumption is determined from at least two images of the same power area and the power reduction is calculated as the difference in volume of feeding between the at least two images.15.The system according to any one of items 12 to 14, wherein the imaging unit is configured to continuously image at least a portion of a feeding area.

权利要求:
Claims (13)
[0001]
1. Method to determine the identity of an individual animal (3) in a population of animals with known identity, the method CHARACTERIZED by the fact that it comprises the steps of:- acquiring at least one 3D image of the back of an animal (3 ) pre-selected,- extracting data from said at least one 3D image relating to the back anatomy and/or topology of the back of the pre-selected animal (3), and- comparing and/or combining said extracted data with reference data corresponding to the anatomy of the back and/or topology of the back of animals with known identity, thus identifying the pre-selected animal (3), in which reference data of animals with known identity are obtained and/or updated at least once once a week when an animal passes a reference data location where an identification number (5) of said animal (3) is associated with at least one reference 3D image of the rear of said animal (3).
[0002]
2. Method, according to claim 1, CHARACTERIZED by the fact that said animal (3) is selected from the group of livestock animals, and/or the group of cattle, cows, dairy cows, bulls, calves, pigs, sows, wild boar, piglets, horses, sheep, goats, deer, and/or wherein said population of animals is a population of animals of the same type, species and/or breed selected from the group of cattle, cows, dairy cows , bulls, calves, pigs, sows, wild boars, geldings, piglets, horses, sheep, goats, deer.
[0003]
3. Method, according to claim 1 or 2, CHARACTERIZED by the fact that the extracted data and the reference data comprise values selected from the group of topographic profiles of the animals.
[0004]
4. Method, according to claim 3, CHARACTERIZED by the fact that the topographic profiles are selected from the group of: the height of the animal, the amplitude of the animal, the contour line along the backbone of the animal, the back length, contour plots for different animal heights, the volume of the animal above different animal heights, the size of the cavities, the depth of the cavities, the distance between two pre-selected points on the animal, where the pre-selected points -selected can be selected from the group of right hip, left hip, right shoulder, left shoulder, tail head, neck, (1) left front rib, (2) left short rib beginning, (3) scapula beginning left, (4) anterior midpoint of left scapula; (5) left scapula, (6) posterior midpoint of left scapula, (7) tip of left scapula, (8) left hip joint, (9) left ischion, (10) nadir of head of left tail, (11) left tail head joint, (12) tail, (13) right tail head joint, (14) right tail head nadir, (15) right ischion, (16) right hip joint, (17) extremity right scapula, (18) posterior midpoint of right scapula, (19) right scapula, (20) anterior midpoint of right scapula, (21) beginning of right scapula, (22) beginning of right short rib (23).
[0005]
5. Method, according to any one of the preceding claims, CHARACTERIZED by the fact that the extracted data and the reference data comprise at least one feature and/or at least one feature vector, preferably related to a feature of the back of the animal (3).
[0006]
6. Method, according to any one of the previous claims, CHARACTERIZED by the fact that said image and/or said reference image is a topographical image of the animals' back, such as multiple layers of 3D images.
[0007]
7. Method according to claim 6, CHARACTERIZED by the fact that the extracted data and reference data comprise at least one feature and/or at least one feature vector based on multi-layer area values of said image in 3D.
[0008]
8. Method according to any one of the preceding claims, CHARACTERIZED by the fact that the extracted data and the reference data comprise at least one feature vector for pre-selected distances calculated from the ground or ground distance that supports animals, the preselected distances are preferably between 70 and 180 cm.
[0009]
9. Method, according to any one of the previous claims, CHARACTERIZED by the fact that it also comprises the steps of determining the food consumption of said identified pre-selected animal (3).
[0010]
10. System (1) for determining the identity of an individual animal (3) among a population of animals with known identity, the system (1) CHARACTERIZED in that it comprises - a first imaging system configured to acquire at least one image pre-selected 3D back of an animal (3), - a different/separate reference imaging unit from the first imaging unit to provide one or more reference 3D images of an animal (3) in the animal population, comprising:- at least one identity determining device configured to determine the identity of said animal (3), and- at least one camera configured to acquire at least one reference 3D image of the back of said animal (3), and - a reference processing unit configured to associate the determined identity of the animal (3) with reference data extracted from said at least one reference 3D image, and- a first u processing unit (6) configured to:- extract data from said at least one 3D image relating to the back anatomy and/or back topology of the pre-selected animal (3), and- combine said extracted data with data from reference corresponding to the anatomy of the back and/or topology of the back of each of the animals with known identity, thus identifying the pre-selected animal (3).
[0011]
11. System, according to claim 10, CHARACTERIZED by the fact that said first imaging system and/or said reference imaging unit and comprises one or more cameras (4) selected from the group of range cameras, cameras stereo, time-of-flight cameras such as a range camera that includes a depth sensor and a 2D camera such as an RGB camera.
[0012]
12. System, according to claim 10 or 11, CHARACTERIZED by the fact that said reference imaging unit is configured to acquire at least one reference 3D image of the back of an animal, when the identity of said animal is determined by said at least one identity determining device and/or wherein said reference imaging unit is configured to acquire at least one (reference) image of an animal's back and/or determine the identity of an animal when said animal is within a predefined distance of said identity determining device.
[0013]
13. System, according to any one of claims 10 to 12, CHARACTERIZED in that it further comprises a feeding area imaging unit configured to acquire images, such as 3D images of a feeding area in front of the animal ( 3) pre-selected identified.

类似技术:

---

公开号 | 公开日 | 专利标题

---

BR112018000046B1|2022-01-25|Method and system for determining the identity of an individual animal in a population of animals with known identity

---

Nasirahmadi et al.2017|Implementation of machine vision for detecting behaviour of cattle and pigs

---

Bewley et al.2008|Potential for estimation of body condition scores in dairy cattle from digital images

---

Rushen et al.2012|Automated monitoring of behavioural-based animal welfare indicators

---

US8971586B2|2015-03-03|Apparatus and method for estimation of livestock weight

---

Flower et al.2009|Gait assessment in dairy cattle

---

Wurtz et al.2019|Recording behaviour of indoor-housed farm animals automatically using machine vision technology: A systematic review

---

Aydin et al.2010|Application of a fully automatic analysis tool to assess the activity of broiler chickens with different gait scores

---

Aydin2017|Using 3D vision camera system to automatically assess the level of inactivity in broiler chickens

---

Van Hertem et al.2013|Comparison of segmentation algorithms for cow contour extraction from natural barn background in side view images

---

US20170103257A1|2017-04-13|Non-invasive multimodal biometrical identification system of animals

---

Pallottino et al.2015|Comparison between manual and stereovision body traits measurements of Lipizzan horses

---

Achour et al.2020|Image analysis for individual identification and feeding behaviour monitoring of dairy cows based on Convolutional Neural Networks |

---

Hansen et al.2015|Non-intrusive automated measurement of dairy cow body condition using 3D video

---

Weber et al.2020|Cattle weight estimation using active contour models and regression trees Bagging

---

US20110054338A1|2011-03-03|Device, system and method for monitoring heat stress of a livestock animal

---

Dalton et al.2016|Validation of HOBO Pendant® data loggers for automated step detection in two age classes of male turkeys: growers and finishers

---

Dohmen et al.2021|Computer vision-based weight estimation of livestock: a systematic literature review

---

WO2021237144A1|2021-11-25|Systems and methods for automatic and noninvasive livestock health analysis

---

Nasirahmadi2017|Development of automated computer vision systems for investigation of livestock behaviours

---

WO2017105195A1|2017-06-22|Method for determining the optimum slaughter time and identifying patterns of behaviour of cattle

---

Condotta et al.2020|Development of method for lameness detection based on depth image analysis

---

Viazzi et al.2013|Using a 3D camera to evaluate the back posture of dairy cows

---

ES2807724T3|2021-02-24|Improved method and relevant apparatus for determining body condition score, body weight and fertility status

---

JP2021101668A|2021-07-15|Estrus diagnosis system, estrus diagnosis method, program and individual identification system

同族专利:

---

公开号 | 公开日

---

AU2016286295A1|2018-01-18|

---

EP3316680A1|2018-05-09|

---

WO2017001538A1|2017-01-05|

---

ES2761702T3|2020-05-20|

---

JP6824199B2|2021-02-03|

---

AU2016286295B2|2019-09-26|

---

CL2017003461A1|2018-05-11|

---

PT3316680T|2019-12-19|

---

UA124378C2|2021-09-08|

---

JP2018520680A|2018-08-02|

---

US20210406530A1|2021-12-30|

---

DK3316680T3|2019-11-18|

---

EA033347B9|2019-12-18|

---

IL256406D0|2018-02-28|

---

NZ738150A|2019-06-28|

---

AU2019283978A1|2020-01-23|

---

US11080522B2|2021-08-03|

---

BR112018000046A2|2018-09-04|

---

US20200143157A1|2020-05-07|

---

EA201890099A1|2018-06-29|

---

IL256406A|2021-05-31|

---

SI3316680T1|2020-01-31|

---

PL3316680T3|2020-04-30|

---

AU2019283978B2|2022-01-06|

---

CN107820616A|2018-03-20|

---

EA033347B1|2019-09-30|

---

MX2017016878A|2018-08-15|

---

CA2989258A1|2017-01-05|

---

EP3316680B1|2019-08-14|

---

LT3316680T|2019-12-10|

引用文献:

---

公开号 | 申请日 | 公开日 | 申请人 | 专利标题

---

---

US4881953A|1988-09-15|1989-11-21|Union Carbide Corporation|Prevention of membrane degradation|

---

US5412420A|1992-10-26|1995-05-02|Pheno Imaging, Inc.|Three-dimensional phenotypic measuring system for animals|

---

DE4420254A1|1994-06-10|1995-01-26|Jochen Stehr|Detecting edges|

---

DE60016767T2|1999-09-02|2006-01-12|Kristoffer Larsen Innovation A/S|METHOD FOR CONTROLLING THE RESTORATION OF FREE-RACING ANIMALS|

---

AUPQ603800A0|2000-03-03|2000-03-30|ID+Plus Pty Ltd|Method and apparatus for livestock identification|

---

JP3803750B2|2002-02-22|2006-08-02|防衛庁技術研究本部長|Volume measuring method and volume measuring program|

---

US7399220B2|2002-08-02|2008-07-15|Kriesel Marshall S|Apparatus and methods for the volumetric and dimensional measurement of livestock|

---

DE10239674A1|2002-08-26|2004-03-11|Hölscher & Leuschner GmbH & Co|Process for monitoring pigs for marketability|

---

US7841300B2|2002-11-08|2010-11-30|Biopar, LLC|System for uniquely identifying subjects from a target population|

---

US20080314324A1|2004-03-30|2008-12-25|Delaval Holding Ab|Arrangement and Method for Determining Positions of the Teats of a Milking Animal|

---

KR100601957B1|2004-07-07|2006-07-14|삼성전자주식회사|Apparatus for and method for determining image correspondence, apparatus and method for image correction therefor|

---

US7497894B2|2005-04-19|2009-03-03|Generon Igs, Inc.|Portable air separation/air dehydration system|

---

US7578871B2|2005-05-25|2009-08-25|Generon Igs, Inc.|Gas separation membrane with partial surfactant coating|

---

US7517388B2|2006-05-15|2009-04-14|Generon Igs, Inc.|Air separation membrane module with variable sweep stream|

---

US7662333B2|2006-08-14|2010-02-16|Generon Igs, Inc.|Vacuum-assisted potting of fiber module tubesheets|

---

DE102007036294A1|2007-07-31|2009-02-05|Gea Westfaliasurge Gmbh|Apparatus and method for providing information about animals when passing through an animal passage|

---

GB0716333D0|2007-08-22|2007-10-03|White Spark Holdings Ltd|Method and apparatus for the automatic grading of condition of livestock|

---

GB0813782D0|2008-07-28|2008-09-03|Delaval Holding Ab|Animal installation with height measurement device|

---

US9684956B2|2008-12-03|2017-06-20|Delaval Holding Ab|Arrangement and method for determining a body condition score of an animal|

---

DE202009008268U1|2009-06-10|2010-11-04|Big Dutchman Pig Equipment Gmbh|Optical based livestock locating device|

---

JP2011177116A|2010-03-02|2011-09-15|Mitsubishi Electric Corp|Rfid animal ear tag and method of manufacturing the same|

---

AU2010219406B2|2010-05-19|2013-01-24|Plf Agritech Pty Ltd|Image analysis for making animal measurements|

---

JP5756967B2|2011-03-30|2015-07-29|国立研究開発法人農業・食品産業技術総合研究機構|Milk cow health condition management method and management system|

---

WO2014083433A2|2012-12-02|2014-06-05|Agricam Ab|Systems and methods for predicting the outcome of a state of a subject|

---

TW201435749A|2013-03-01|2014-09-16|ting-chang Huang|Animal mark chip and its data processing system|

---

JP6556119B2|2013-04-10|2019-08-07|ヴァイキング ジェネティクス エフエムベーア|System for determining food consumption of at least one animal|

---

KR101527801B1|2013-05-22|2015-06-11|주식회사 아이싸이랩|Apparatus and Method of an Animal Recognition System using nose patterns|

---

US20160125276A1|2013-06-04|2016-05-05|Clic RWeight ,LLC|Methods and systems for marking animals|

---

CN103514422B|2013-08-29|2016-03-16|广西慧云信息技术有限公司|A kind of abnormity early warning method of searching for food of identity-based identification|

---

US10430942B2|2013-10-01|2019-10-01|University Of Kentucky Research Foundation|Image analysis for predicting body weight in humans|

---

JP6291927B2|2014-03-13|2018-03-14|富士通株式会社|Identification method, identification program, identification apparatus and identification system|

---

US9084411B1|2014-04-10|2015-07-21|Animal Biotech Llc|Livestock identification system and method|

---

KR102333463B1|2014-07-02|2021-12-03|한미약품 주식회사|Pharmaceutical Composition for Oral Administration Comprising Rivaroxaban And Method of Preparing the Same|

---

KR101676643B1|2014-10-16|2016-11-29|렉스젠|Apparatus for managing livestock and method thereof|

---

US10027179B1|2015-04-30|2018-07-17|University Of South Florida|Continuous wireless powering of moving biological sensors|

---

WO2017001538A1|2015-07-01|2017-01-05|Viking Genetics Fmba|System and method for identification of individual animals based on images of the back|EP3122173B1|2014-03-26|2021-03-31|SCR Engineers Ltd|Livestock location system|

---

US10986817B2|2014-09-05|2021-04-27|Intervet Inc.|Method and system for tracking health in animal populations|

---

US11071279B2|2014-09-05|2021-07-27|Intervet Inc.|Method and system for tracking health in animal populations|

---

WO2017001538A1|2015-07-01|2017-01-05|Viking Genetics Fmba|System and method for identification of individual animals based on images of the back|

---

EP3449719A4|2016-04-28|2020-03-18|Osaka University|Health condition estimation device|

---

US10485643B2|2017-06-08|2019-11-26|Wildlife Protection Management, Inc.|Animal control system|

---

CN107992903A|2017-09-20|2018-05-04|翔创科技（北京）有限公司|Livestock personal identification method, device, storage medium and electronic equipment|

---

NL2020025B1|2017-12-06|2019-06-18|Lely Patent Nv|Feed system|

---

CN108764045B|2018-04-26|2019-11-26|平安科技（深圳）有限公司|Livestock recognition methods, device and storage medium|

---

WO2019207041A1|2018-04-26|2019-10-31|F. Hoffmann-La Roche Ag|Method of and system for tracking an animal in a population of animals|

---

EP3574751A1|2018-05-28|2019-12-04|Bayer Animal Health GmbH|Apparatus for fly management|

---

JP2022514115A|2018-10-17|2022-02-09|グループ ロ－マン インク．|Livestock surveillance|

---

CN109583254A|2018-11-16|2019-04-05|北京中竞鸽体育文化发展有限公司|Moving object recognition methods, device and storage medium|

---

US11109576B2|2018-11-16|2021-09-07|International Business Machines Corporation|Livestock management|

---

CN109583400A|2018-12-05|2019-04-05|成都牧云慧视科技有限公司|One kind is registered automatically without intervention for livestock identity and knows method for distinguishing|

---

TWI714057B|2019-04-17|2020-12-21|國立臺灣大學|Analysis system and method for feeding milk-production livestock|

---

EP3756458A1|2019-06-26|2020-12-30|Viking Genetics FmbA|Weight determination of an animal based on 3d imaging|

---

CN110569735A|2019-08-13|2019-12-13|中国农业大学|Analysis method and device based on back body condition of dairy cow|

---

WO2021032890A2|2019-08-21|2021-02-25|Dairymaster|A method and apparatus for determining the identity of an animal of a herd of animals|

---

CN110569759B|2019-08-26|2020-11-03|王睿琪|Method, system, server and front end for acquiring individual eating data|

---

FR3101435B1|2019-09-30|2021-10-29|Uwinloc|aid system for the orientation of an antenna of a beacon with regard to a target position|

---

CN110796043B|2019-10-16|2021-04-30|北京海益同展信息科技有限公司|Container detection and feeding detection method and device and feeding system|

---

CN110762031B|2019-10-23|2020-08-14|台州辉腾泵业有限公司|Water pump driving device based on data quantitative control|

---

CN112189588A|2020-10-10|2021-01-08|东北农业大学|Cow image information collecting and processing method and system|

---

RU2754095C1|2020-10-29|2021-08-26|Федеральное государственное бюджетное учреждение науки Институт проблем управления им. В.А. Трапезникова Российской академии наук|Methodology for preparing sets of photos for machine analysis for personal identification of animals by face|

法律状态:
2019-10-01| B06U| Preliminary requirement: requests with searches performed by other patent offices: procedure suspended [chapter 6.21 patent gazette]|

---

2021-08-10| B07A| Application suspended after technical examination (opinion) [chapter 7.1 patent gazette]|

---

2021-12-07| B09A| Decision: intention to grant [chapter 9.1 patent gazette]|

---

2022-01-25| B16A| Patent or certificate of addition of invention granted [chapter 16.1 patent gazette]|Free format text: PRAZO DE VALIDADE: 20 (VINTE) ANOS CONTADOS A PARTIR DE 30/06/2016, OBSERVADAS AS CONDICOES LEGAIS. |

优先权:

---

申请号 | 申请日 | 专利标题

---

EM15174783.9|2015-07-01|

---

EP15174783|2015-07-01|

---

PCT/EP2016/065241|WO2017001538A1|2015-07-01|2016-06-30|System and method for identification of individual animals based on images of the back|

---