• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    The present invention relates to an assembly monitoring method based on deep learning, comprising steps of: creating a training set for a physical assembly body, 5 the training set comprising a depth image set Di and a label image set Li of a 3D assembly body at multiple monitoring angles, wherein i represents an assembly step, the depth image set Di in the ith step corresponds to the label image set Li in the ith step, and in label images in the label image set Li, different parts of the 3D assembly body are rendered by different colors; training a deep learning network 10 model by the training set; and obtaining, by the depth camera, a physical assembly body depth image C in a physical assembly scene, inputting the physical assembly body depth image C into the deep learning network model, and outputting a pixel segmentation image of the physical assembly body, in which different parts are represented by pixel colors to identify all the parts of the physical assembly body. In 15 the present invention, parts in the assembly body can be identified, and the assembly steps, as well as the occurrence of assembly errors and the type of errors, can be monitored for the parts
    公开号:NL2024682A
    申请号:NL2024682
    申请日:2020-01-16
    公开日:2020-09-04
    发明作者:Chen Chengjun；Zhang Chunlin；Li Dongnian；Hong Jun
    申请人:Univ Qingdao Technology；
    IPC主号:
    专利说明:

    ASSEMBLY MONITORING METHOD AND DEVICE BASED ON DEEP LEARNING, AND READABLE STORAGE MEDIUM
    TECHNICAL FIELD The present invention relates to an assembly monitoring method and device based on deep learning, and a readable storage medium, belonging to the field of computer monitoring based on image processing.
    BACKGROUND OF THE PRESENT INVENTION Personalized customization is one of main trends of development in the manufacturing industry. At present, the mechanical assembly industry in China is low in automation level. The assembly of a large number of customized products and complex mechanical products still relies on manual assembly. Due to the frequent change of product types and the large number of parts, the process and assembly accuracy are highly required. During the assembly, various events such as missing of parts and wrong assembly sequence occur easily, which affect the product assembly quality and efficiency. In order to better adapt to the new situation of the manufacturing industry and to meet the assembly of large-scale customized products, the conventional assembly workshop urgently needs an intelligent assembly monitoring system that can intelligently identify parts of the assembly body and monitor the assembly process.
    On the other hand, during the assembly of complex products, the assembly progresses of components are not coordinated. As a result, the components needed during the final assembly of products are not all in readiness, thereby affecting the production efficiency. Therefore, it is also necessary to monitor and coordinate the assembly progresses of components, thus to improve the assembly efficiency.
    The Invention Patent CN108491776A, entitled "METHOD AND DEVICE FOR ASSEMBLY PART IDENTIFICATION BASED ON PIXEL CLASSIFICATION, AND ASSEMBLY MONITORING SYSTEM", disclosed part identification by training a random forest model. The training process requires artificially designed deepdifferential features, and features cannot be learned automatically. This technical solution can only realize part identification and determine whether the part is wrong, and cannot further monitor whether the assembly position and assembly sequence of the part are wrong, so it is impossible to detect missing and misplacing in time.
    SUMMARY OF THE PRESENT INVENTION To solve the technical problems, the present invention provides an assembly monitoring method based on deep learning, by which parts in the assembly body can be identified, and the assembly steps during the assembly, as well as the occurrence of assembly errors and the type of errors, can be monitored for the parts. The present invention employs the following technical solution.
    An assembly monitoring method based on deep learning is provided, implemented by execution of a computer program by a processor, comprising steps of: creating a training set for a physical assembly body, the training set comprising a depth image set Di and a label image set Li of a 3D assembly body at multiple monitoring angles, wherein i represents an assembly step, the depth image set Di in the ith step corresponds to the label image set Li in the ith step, and in label images in the label image set Li, different parts of the 3D assembly body are rendered by different colors; the monitoring angle is an angle of view between a depth camera and an assembly body to be monitored, the depth camera is arranged in situ and changed in its capture angle within a preset rotation range to form the multiple monitoring angles; and the 3D assembly body is a 3D model of the physical assembly body; training a deep learning network model by the training set; and obtaining, by the depth camera, a physical assembly body depth image C in a physical assembly scene, inputting the physical assembly body depth image C into the deep learning network model, and outputting a pixel segmentation image of the physical assembly body, in which different parts are represented by pixel colors to identify all the parts of the physical assembly body.
    Preferably, the step of creating a training set comprises: establishing a 3D assembly body, comprising 3D models of parts and a positional relation between the parts; annotating different parts in the 3D assembly body by different colors to establish an annotation model of the 3D assembly body; defining an assembly sequence model, comprising defining parts contained in each assembly step; and according to the assembly sequence model, loading annotation models of different parts, generating a 3D model and an annotation model for an assembly body corresponding to each assembly step, synthesizing a depth image set Di of the 3D assembly body at multiple monitoring angles by a depth buffer, and synthesizing a label image set Li by an imaging model of the depth camera, wherein i represents an assembly step, the depth image set Di in the ith step corresponds to the label image set Li in the ith step, and in label images in the label image set Li, different parts of the 3D assembly body are rendered by different colors thus to establish a training set.
    Preferably, the deep learning network model is a fully convolutional deep network comprising a convolution process and a deconvolution process, and the deep learning network model is trained by using the depth image set Di in the training set as the input to the deep learning network model and the label image set Li in the training set as the output from the deep learning network model. Preferably, the step of obtaining, by the depth camera, a physical assembly body depth image C in a physical assembly scene comprises: in an initialization stage, fixing the depth camera, aligning the depth camera to an assembly zone, and capturing a physical assembly scene depth image A not containing the physical assembly body by the depth camera; in a circulation stage, capturing a physical assembly scene depth image B containing the physical assembly body by the depth camera; and in the physical assembly scene depth image A and the physical assembly scene depth image B, calculating the physical assembly body depth image C by subtracting depth values of corresponding pixel points, specifically: if the absolute value of the difference between depth values of corresponding pixel points in the depth image A and the depth image B is less than a preset threshold D, considering this pixel point in the depth image B as the background, or otherwise considering this pixel point in the depth image B as the assembly body; and if a pixel point in the depth image A and a pixel point in the depth image B have samecoordinates in an image coordinate system, considering the two pixel points as corresponding pixel points. Preferably, after identifying all the parts of the physical assembly body, the method further comprises assembly step monitoring: obtaining statistic information about the composition of pixels for parts in the pixel segmentation image of the physical assembly body, determining the composition of the parts, and inquiring a current assembly step according to the composition of the parts and based on the assembly sequence model to complete assembly step monitoring.
    Preferably, after identifying all the parts of the physical assembly body, the method further comprises parts assembly monitoring: determining whether a part is assembled incorrectly, specifically: according to the current assembly step, comparing each pixel point in the pixel segmentation image of the physical assembly body with each pixel point in a correct label image corresponding thereto, and respectively calculating a pixel overlap rate 9: petween pixels for a part, which is assembled in the current assembly step, in the pixel segmentation image of the physical assembly body with respect to pixels for the part in the correct label image, Co , Wherein Pe represents the number of pixel points, the coordinates of which are overlapped, for the assembled part in the correct label image and the pixel segmentation image, and He represents the number of pixel points for the assembled part in the pixel segmentation image of the physical assembly body; the correct label image is a label image synthesized by the imaging model of the depth camera by using coordinates of the depth camera in the physical assembly scene as coordinates of a viewpoint in a virtual scene, and each step corresponds to one correct label image; and determining that the part is not assembled incorrectly if Te is greater than or equal to a preset overlap threshold E, or otherwise presenting examination information. Preferably, the part assembly monitoring further comprises determining the type of assembly errors, comprising: according to the current assembly step, comparing each pixel point in the pixel segmentation image of the physical assembly body witheach pixel point in a correct label image corresponding thereto, and respectively calculating a pixel reduction rate 9s between pixels for a part, which is assembled in the current assembly step, in the pixel segmentation image of the physical assembly body with respect to pixels for the part in the correct label image, g, = 27 2 5 ) % wherein "2 represents the number of pixel points for the assembledpart in the correct label image, and Ha represents the number of pixel points for the assembled part in the pixel segmentation image of the physical assembly body; determining that the part is missed, when the dz for the part is less than or equal to a preset overlap threshold F and the absolute value of @h is greater than or equalto a preset missing threshold, that is, the part has a low overlap rate of pixel points and a great difference in the number of pixel points; and determining that the part is misplaced, when the de for the part is less than or equal to the preset overlap threshold F and the absolute value of © is less than or equal to a preset misplacing threshold, that is, the part has a low overlap rate of pixel points but asmal difference in the number of pixel points.
    The present invention further provides an assembly monitoring device based on deep learning, comprising a memory, a processor, and a computer program stored in the memory and can be run on the processor, wherein the program, whenexecuted by the processor, causes the processor to perform steps of: creating a training set for a physical assembly body, the training set comprising a depth image set Di and a label image set Li of a 3D assembly body at different monitoring angles, wherein i represents an assembly step, the depth image set Di in the ith step corresponds to the label image set Li in the ith step, and in label images in thelabel image set Li, different parts of the 3D assembly body are rendered by different colors; the 3D assembly body is a 3D model of the physical assembly body; training a deep learning network model by the training set; and obtaining, by the depth camera, a physical assembly body depth image C in a physical assembly scene, inputting the physical assembly body depth image C into the deep learning networkmodel, and outputting a pixel segmentation image of the physical assembly body,
    in which different parts are represented by pixel colors to identify all the parts of the physical assembly body. The present invention further provides a computer-readable storage medium storing a computer program that, when executed by the processor, causes the processor to perform steps of: creating a training set for a physical assembly body, the training set comprising a depth image set Di and a label image set Li of a 3D assembly body at different monitoring angles, wherein i represents an assembly step, the depth image set Di in the ith step corresponds to the label image set Li in the ith step, and in label images in the label image set Li, different parts of the 3D assembly body are rendered by different colors; the 3D assembly body is a 3D model of the physical assembly body; training a deep learning network model by the training set; and obtaining, by the depth camera, a physical assembly body depth image C in a physical assembly scene, inputting the physical assembly body depth image C into the deep learning network model, and outputting a pixel segmentation image of the physical assembly body, in which different parts are represented by pixel colors to identify all the parts of the physical assembly body. The present invention has the following beneficial effects:
    1. in the present invention, the composition of parts of the assembly body is calculated by a deep learning neural network and according to a depth image, to realize the identification of the parts of the assembly body;
    2. in the present invention, the deep learning network model is trained by a depth image set and a label image set, to realize the monitoring of the physical assembly and meet the assembly monitoring requirement for customized products; and
    3. in the present invention, the assembly step monitoring, the missing determination, the misplacing monitoring and the like can be realized, and it is of great application value for large-scale customization in workshops.
    BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a schematic flowchart of an assembly monitoring method based on deep learning.
    Fig. 2 is a schematic view of an assembly monitoring device based on deep learning, comprising a server computer system.
    DETAILED DESCRIPTION OF THE PRESENT INVENTION The present invention will be described below in detail with reference to the accompanying drawings by specific embodiments. Embodiment 1 The Embodiment 1 of the present application provides a schematic flowchart of a verification method, as shown in Fig. 1, comprising:
    1. establishing a 3D assembly body by CAD software or other three-dimensional software, comprising 3D models of parts and a positional relation between the parts;
    2. importing the 3D assembly body into virtual reality modeling software (for example, Multigen Creator), and annotating different parts in the 3D assembly body by different colors to establish an annotation model of the 3D assembly body;
    3. defining an assembly sequence model, comprising defining parts contained in each assembly step; and
    4. according to the assembly sequence model, loading annotation models of different parts, and generating a 3D model and an annotation model for an assembly body corresponding to each assembly step.
    5. A depth image set Di of the 3D assembly body at multiple monitoring angles synthesized by a depth buffer, and a label image set Li is synthesized by an imaging model of the depth camera, wherein i represents an assembly step, the depth image set Di in the ith step corresponds to the label image set Li in the ith step, and in label images in the label image set Li, different parts of the 3D assembly body are rendered by different colors thus to establish a training set. Generally, the images in the depth image set Di and the images in label image set Li have a same resolution. A depth image set corresponding to each step contains a set of depth images at multiple monitoring angles in this assembly step, and a label image set corresponding to each step contains a set of label images at multiple monitoring angles in this assembly step.
    6. The deep learning network model is trained by the training set. The deep learning network model is a fully convolutional deep network that replaces fully connected layers in the conventional convolutional neural network with convolutional layers, comprising a convolution process and a deconvolution process, and the deep learning network model is trained by using the depth image set Di in the training set as the input to the deep learning network model and the label image set Li in the training set as the output from the deep learning network model.
    7. A physical assembly body depth image C in a physical assembly scene is obtained by the depth camera. The step of obtaining, by the depth camera, a physical assembly body depth image C in a physical assembly scene comprises: in an initialization stage, fixing the depth camera, aligning the depth camera to an assembly zone, and capturing a physical assembly scene depth image A not containing the physical assembly body by the depth camera; in a circulation stage, capturing a physical assembly scene depth image B containing the physical assembly body by the depth camera; and in the physical assembly scene depth image A and the physical assembly scene depth image B, calculating the physical assembly body depth image C by subtracting corresponding pixels, specifically: if the absolute value of the difference between depth values of corresponding pixel points in the physical assembly scene depth image A and the physical assembly scene depth image B is less than a preset threshold d, considering this pixel point in the physical assembly scene depth image B as the background and considering the depth value of the corresponding pixel point in the physical assembly body depth image C as 0, or otherwise considering this pixel point in the physical assembly scene depth image B as the assembly body and considering the depth value of the corresponding pixel point in the physical assembly body depth image C as the depth value of the corresponding pixel point in the physical assembly scene depth image B. The pixel points in the two depth images, having same coordinates, are corresponding pixel points in the two depth images.
    8. The physical assembly body depth image C is input into the deep learning network model, and a pixel segmentation image of the physical assembly body is output, in which different parts are represented by pixel colors to identify all the parts of the physical assembly body.
    9. After identifying all the parts of the physical assembly body, the method further comprises assembly step monitoring: obtaining statistic information about thecomposition of pixels for parts in the pixel segmentation image of the physical assembly body, determining the composition of the parts, and inquiring a current assembly step according to the composition of the parts and based on the assembly sequence model to complete assembly step monitoring.
    10. After identifying all the parts of the physical assembly body, the method further comprises parts assembly monitoring: determining whether a part is assembled incorrectly, specifically: according to the current assembly step, comparing each pixel point in the pixel segmentation image of the physical assembly body with each pixel point in a correct label image corresponding thereto, and respectively calculating a pixel overlap rate 4: petween pixels for a part, which is assembled in the current assembly step, in the pixel segmentation image of the physical assembly body with respect to pixels for the part in the correct label image, =, wherein © represents the number of pixel points, the coordinates of which are overlapped, for the assembled part in the correct label image and the pixel segmentation image, and Bz represents the number of pixel points for the assembled part in the pixel segmentation image of the physical assembly body; the pixel segmentation image of the physical assembly body and the correct label image have a same resolution; the correct label image is a label image synthesized by the imaging model of the depth camera by using coordinates of the depth camera in the physical assembly scene as coordinates of a viewpoint in a virtual scene, and in the correct label image, different parts of the 3D assembly body are rendered by different colors. It is determined that the part is not assembled incorrectly if dz ig greater than or equal to a preset overlap threshold E, or otherwise examination information is presented. Generally, it is indicated that the partis not assembled incorrectly when Te approaches 1. Therefore, the overlap threshold E may be a numerical value approaching 1.
    11. The part assembly monitoring further comprises determining the type of assembly errors, specifically: according to the current assembly step, comparing each pixel point in the pixel segmentation image of the physical assembly body with each pixel point in a correct label image corresponding thereto, and respectively calculating a pixel reduction rate d: between pixels for a part, which is assembledin the current assembly step, in the pixel segmentation image of the physical assembly body with respect to pixels for the part in the correct label image, g = By ~ 7, i} 7% wherein ”s represents the number of pixel points for the assembled part in the correct label image, and fs represents the number of pixel points for the assembled part in the pixel segmentation image of the physical assembly body; determining that the part is missed, when the 4 for the part is less than or equal to a preset overlap threshold F and the absolute value of % is greater than or equal to a preset missing threshold, that is, the part has a low overlap rate of pixel points and a great difference in the number of pixel points; and determining that the part is misplaced, when the de for the part is less than or equal to the preset overlap threshold F and the absolute value of Gn is less than or equal to a preset misplacing threshold, that is, the part has a low overlap rate of pixel points but a small difference in the number of pixel points.
    The overlap threshold F is a numerical value approaching 0. When 9: is less than the overlap threshold F, it is indicated that the part has a low overlap rate of pixel points.
    The missing threshold may be a numerical value greater than 0 or approaching 1, indicating a great difference in the number of pixel points for the part between the pixel segmentation image of the physical assembly body and the correct label image.
    The misplacing threshold may be a numerical value approaching 0, indicating a small difference in the number of pixel points for the part between the pixel segmentation image of the physical assembly body and the correct label image.
    The absolute value of the obtained “= approaches 0. Embodiment 2
    As shown in Fig. 2, this embodiment is a schematic view of an assembly monitoring device based on deep learning, comprising a server computer system 20. The server computer system comprises a processor 201, a storage unit 202, and a computer program stored in the storage unit and can be run on the processor.
    The computer program comprises steps of:
    2010: creating a training set for a physical assembly body, the training set comprising a depth image set Di and a label image set Li of a 3D assembly body at multiple monitoring angles, wherein i represents an assembly step, the depth image set Di in the ith step corresponds to the label image set Li in the ith step, and in label images in the label image set Li, different parts of the 3D assembly body are rendered by different colors; and the 3D assembly body is a 3D model of the physical assembly body; 2020: training a deep learning network model by the training set; and 2030: obtaining, by the depth camera, a physical assembly body depth image C in a physical assembly scene, inputting the physical assembly body depth image C into the deep learning network model, and outputting a pixel segmentation image of the physical assembly body, in which different parts are represented by pixel colors to identify all the parts of the physical assembly body.
    Embodiment 3 A computer-readable storage medium is provided, storing a computer program that, when executed by the processor, causes the processor to perform steps of: creating a training set for a physical assembly body, the training set comprising a depth image set Di and a label image set Li of a 3D assembly body at multiple monitoring angles, wherein i represents an assembly step, the depth image set Di in the ith step corresponds to the label image set Li in the ith step, and in label images in the label image set Li, different parts of the 3D assembly body are rendered by different colors; and the 3D assembly body is a 3D model of the physical assembly body; training a deep learning network model by the training set; and obtaining, by the depth camera, a physical assembly body depth image C in a physical assembly scene, inputting the physical assembly body depth image C into the deep learning network model, and outputting a pixel segmentation image of the physical assembly body, in which different parts are represented by pixel colors to identify all the parts of the physical assembly body. The specific steps in the Embodiment 2 and Embodiment 3 may refer to the related descriptions of the embodiment of the method.
    The foregoing contents are merely embodiments of the present invention, and are not intended to limit the patent scope of the present invention.
    Any equivalent structures or equivalent flow transformations made on the basis of the descriptions and drawings of the present invention, or direct or indirect applications thereof in other related technical fields shall be within the patent scope of the present invention.
    权利要求:
    Claims (9)
    [1]
    An assembly monitoring method based on deep learning, performed by execution of a computer program by a processor, comprising the steps: creating a training set for a physical assembly body, the training set includes a depth image set, and a label image set Li of a 3D assembly body from multiple monitoring angles, where i represents an assembling step, the depth image set D; corresponds to the label image set Li in step i in step i, and in the label images in the label image set Li, different parts of the 3D assembly body are represented by different colors; the monitoring angle is a viewing angle between a depth camera and an assembly body to be monitored, the depth camera is arranged in situ and changed in its angle of view within a preset rotation range to form the multiple monitoring angles; and the 3D assembly body is a 3D model of the physical assembly body; training a deep learning network model through the training set; and obtaining, by the depth camera, a physical assembly body image C in a physical assembly environment, inputting the physical assembly body image C into the deep learning network model, and outputting a pixel segmentation image of the physical assembly body, in which different parts are represented by pixel colors to identify all parts of the physical assembly body.
    [2]
    The assembly monitoring method based on deep learning according to claim 1, wherein the step of creating a training set comprises: establishing a 3D assembly body, including 3D models of parts and a positional relationship between the parts;
    annotating different parts in the 3D assembly body by different colors to create an annotation model of the 3D assembly body; defining a sequential assembly model, including defining parts in each assembly step; and according to the sequential assembly model, loading annotation models of different parts, generating a 3D model and an annotation model for an assembly body corresponding to each assembly step, assembling a depth image set D; of the 3D assembly body from multiple monitoring angles through a depth buffer and assembling a label image set L; through an image model of the depth camera, in which i represents an assembling step, the set of depth images D; in step i corresponds to the label image set Li in step i, and in the label images in the label image set Li, different parts of the 3D assembly body are represented by different colors to beach a training set.
    [3]
    The assembly monitoring method based on deep learning according to claim 1, wherein the deep learning network model is a full convolutional deep network including a convolution process and a deconvolution process, and the deep learning network model is trained using the depth image set D; in the training set as the input to the deep learning network model and the label image set Li in the training set as the output of the deep learning network model.
    [4]
    The assembly monitoring method based on deep learning according to claim 1, wherein the step of obtaining, by the depth camera, a physical assembly body image C in a physical assembly environment comprises: in an initialization phase, fixing the depth camera, aligning the depth camera on an assembly zone, and capturing by the depth camera the physical assembly environment depth image A not including the physical assembly body;
    in a circulation phase, capturing a physical assembly environment depth image B with the physical assembly body by the depth camera; and in the physical assembly environment depth image A and the physical assembly environment depth image B, calculating the physical assembly body depth image C by subtracting depth values from corresponding pixel points, comprising: if the absolute value of the difference between the depth values of the corresponding pixel points in the depth image A and the depth image B is less than a preset threshold D, considering this pixel point in the depth image B as the background, or otherwise considering this pixel point in the depth image B as the assembly body; and if a pixel point in the depth image A and a pixel point in the depth image B have the same coordinates in an image coordinate system, considering the two pixel points as corresponding pixel points.
    [5]
    The assembly monitoring method based on deep learning according to claim 2, after identifying all parts of the physical assembly body, assembly step monitoring further comprising: obtaining statistical information about the composition of pixels for parts in the pixel segmentation image of the physical assembly body , determining the composition of the parts, and examining a current assembly step of the composition of the parts and based on the sequential assembly model to complete the assembly step monitoring.
    [6]
    The assembly monitoring method based on deep learning according to claim 1, after identifying all parts of the physical assembly body, further comprising part assembly monitoring: determining whether a part has been incorrectly assembled, comprising: according to the current assembly step, comparing each pixel point in the pixel segmentation image of the physical assembly body with each pixel point in a correct label image corresponding to it, and respectively calculating a pixel overlap ratio 1 between pixels for a part, which is assembled in the current assembly step, in the pixel segmentation image of the physical assembly body with respect to of the pixels for the part in the correct label 0 = e image 2, where “represents the number of pixel points, whose coordinates overlap, for the assembled part in the correct label image and the pixel segmentation image of the physical assembly body; the correct label image is a label image composed by the depth camera image model using coordinates of the depth camera in the physical assembly environment as coordinates of a viewing angle in a virtual environment and each step corresponds to a correct label image; and determining that the part is not incorrectly assembled if the is greater than or equal to a preset overlap threshold E, or otherwise presenting examination information.
    [7]
    The assembly monitoring method based on deep learning according to claim 6, wherein the part assembly monitoring further comprises determining the type of assembly errors, comprising: according to the current assembly step, comparing each pixel point in the pixel segmentation image of the physical assembly body with each pixel point in a calculate corresponding correct label image, and a pixel reduction degree 7, respectively, between pixels for a part, which is assembled in the current assembly step, in the pixel segmentation image of the physical assembly body with respect to the pixels of the part in the correct label image, : ”; where = represents the number of pixel points of the assembled part in the correct label image, and “represents the number of pixel points for the assembled part in the pixel segmentation image of the physical assembly body; determining that the part is missed, when the 4 for the part is less than or equal to a preset overlap threshold F and the absolute value of%: is greater than or equal to a preset missing threshold, i.e. part has low pixel point overlap ratio and large pixel point difference; and determining that the part is misplaced when the “= for the part is less than or equal to the preset threshold F and the absolute value of 92 is less than or equal to the preset misplacement threshold, that is, the part has a low pixel point overlap ratio but a small difference in the number of pixel points.
    [8]
    An assembly monitoring device based on deep learning, comprising a memory, a processor and a computer program stored in the memory and can be executed on the processor, wherein the program, when executed by the processor, causes the processor to perform the assembly monitoring method based on on deep learning according to any one of claims 1-7.
    [9]
    A computer readable storage medium for storing a computer program which, when executed by the processor, causes the processor to perform the deep learning assembly monitoring method according to any one of claims 1-7.
    类似技术:
    公开号 | 公开日 | 专利标题
    NL2024682B1|2020-11-05|Assembly monitoring method and device based on deep learning, and readable storage medium
    US9251582B2|2016-02-02|Methods and systems for enhanced automated visual inspection of a physical asset
    US9058669B2|2015-06-16|Incorporating video meta-data in 3D models
    CN106797458B|2019-03-08|The virtual change of real object
    JP6551184B2|2019-07-31|Simulation apparatus, simulation method, and simulation program
    US10997711B2|2021-05-04|Appearance inspection device
    Alvarez et al.2011|Providing guidance for maintenance operations using automatic markerless augmented reality system
    Bhattacharya et al.2019|Augmented reality via expert demonstration authoring |
    Zubizarreta et al.2019|A framework for augmented reality guidance in industry
    US10854006B2|2020-12-01|AR-enabled labeling using aligned CAD models
    Olbrich et al.2011|Augmented reality pipe layout planning in the shipbuilding industry
    CN111300416B|2021-07-13|Modularized reconfigurable robot planning simulation method and system based on augmented reality
    Wientapper et al.2011|Composing the feature map retrieval process for robust and ready-to-use monocular tracking
    Funke et al.2009|A framework for evaluating visual slam
    US20210279957A1|2021-09-09|Systems and methods for building a virtual representation of a location
    US20180126654A1|2018-05-10|Multi-part counting system for three-dimensional printed parts
    Schlette et al.2014|A new benchmark for pose estimation with ground truth from virtual reality
    CN110059436B|2021-07-06|Three-dimensional visualization software development of autonomous guarantee system of spacecraft
    Gimeno et al.2013|An easy-to-use ar authoring tool for industrial applications
    Gimeno et al.2012|An Occlusion-aware AR Authoring Tool for Assembly and Repair Tasks.
    Irgenfried et al.2017|A versatile hardware and software toolset for computer aided inspection planning of machine vision applications
    Gimeno et al.2012|An advanced authoring tool for augmented reality applications in industry
    Patibandla et al.2020|Augmented reality for assistive maintenance and real-time failure analysis in industries
    Ren et al.2017|26‐2: Invited Paper: Recent Advances in HMI for Automotive Aftermarket Applications
    US20220005260A1|2022-01-06|System for determining visually relevant design differences between 3d models
    同族专利:
    公开号 | 公开日
    US10964025B2|2021-03-30|
    CN109816049A|2019-05-28|
    US20200273177A1|2020-08-27|
    CN109816049B|2020-09-18|
    NL2024682B1|2020-11-05|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    CN108491776A|2018-03-12|2018-09-04|青岛理工大学|Assembly Parts Recognition method, apparatus based on pixel classifications and monitoring system|
    US9483584B2|2012-01-19|2016-11-01|Oracle International Corporation|Collaboration within a visualization application|
    DE112013003338B4|2012-07-02|2017-09-07|Panasonic Intellectual Property Management Co., Ltd.|Size measuring device and size measuring method|
    US10824310B2|2012-12-20|2020-11-03|Sri International|Augmented reality virtual personal assistant for external representation|
    US9292180B2|2013-02-28|2016-03-22|The Boeing Company|Locator system for three-dimensional visualization|
    JP6344890B2|2013-05-22|2018-06-20|川崎重工業株式会社|Component assembly work support system and component assembly method|
    US9965897B2|2013-07-08|2018-05-08|OPS Solutions, LLC|Eyewear operational guide system and method|
    US9305216B1|2014-12-15|2016-04-05|Amazon Technologies, Inc.|Context-based detection and classification of actions|
    WO2017160688A1|2016-03-14|2017-09-21|Siemens Aktiengesellschaft|Method and system for efficiently mining dataset essentials with bootstrapping strategy in 6dof pose estimate of 3d objects|
    CN108602191B|2016-03-14|2021-11-30|欧姆龙株式会社|Operation information generating device, operation information generating method, and recording medium|
    CN106897995B|2017-02-04|2019-08-06|同济大学|A kind of components automatic identifying method towards Automatic manual transmission process|
    CN109117822A|2018-08-31|2019-01-01|贵州大学|A kind of part case segmentation recognition method based on deep learning|US11107236B2|2019-04-22|2021-08-31|Dag Michael Peter Hansson|Projected augmented reality interface with pose tracking for directing manual processes|
    CN110543892A|2019-08-05|2019-12-06|青岛理工大学|part identification method based on multilayer random forest|
    CN110666793B|2019-09-11|2020-11-03|大连理工大学|Method for realizing robot square part assembly based on deep reinforcement learning|
    CN110738164A|2019-10-12|2020-01-31|北京猎户星空科技有限公司|Part abnormity detection method, model training method and device|
    CN111207875B|2020-02-25|2021-06-25|青岛理工大学|Electromyographic signal-torque matching method based on multi-granularity parallel CNN model|
    CN112965372A|2021-02-01|2021-06-15|中国科学院自动化研究所|Micro-part precision assembly method, device and system based on reinforcement learning|
    法律状态:
    优先权:
    申请号 | 申请日 | 专利标题
    CN201910131395.1A|CN109816049B|2019-02-22|2019-02-22|Assembly monitoring method and device based on deep learning and readable storage medium|
    [返回顶部]