• 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 a method and apparatus for processing a sequence of images to detect movement of an object in the sequence. In particular, the method comprises (a) supplying an image frame of a first sequence; (b) initializing a reference image that includes image information about a stationary object in the scene represented by the first sequence of images; (c) supplying a next image frame temporarily following the image frame of the first sequence; (d) comparing the reference image with the next image frame to produce a two-dimensional motion image representing motion information relating to movement of an object in the scene; (e) updating the reference image with information in the next image frame, wherein the information used to update the reference image merely represents a fixed object in the scene and ignores the moving object and the temporary fixed object in the scene. To; And (f) repeating steps (c), (d), and (e) for each subsequent image supplied. The method is executed by an image processing apparatus. A particular embodiment of the method and apparatus of the present invention is a traffic monitoring system that identifies a change in the illuminance amplitude of the vehicle and eliminates error identification of non-physical movement in the scene, such as shadows and headlight reflected light.
    公开号:KR19980701568A
    申请号:KR1019970704959
    申请日:1996-01-17
    公开日:1998-05-15
    发明作者:람버트 에너스트 윅슨；스티븐 찰스 에이치에스유
    申请人:윌리암 제이. 버크；데이비드 사르노프 리서치 센타, 인코포레이티드；
    IPC主号:
    专利说明:

    Method and apparatus for detecting movement of an object in an image sequence
    Various types of traffic monitoring systems using image processing are known and examples of such systems are disclosed in US Pat. Nos. 4,433,325, 4,847,772, 5,161,107 and 5,313,295. However, there is a need for a robust traffic monitoring system that can be calculated efficiently and is relatively inexpensive to implement.
    Furthermore, the present invention provides a hierarchical structure such as the pyramid technique disclosed in Andersen et al., US Pat. No. 4,692,806, issued September 8, and Bergen et al. The image flow technique described with the model-based movement estimation can be used. The two known techniques are incorporated herein by reference.
    The present invention relates to an image processing system, and in particular to a system for digitally processing pixels of a continuous image frame (image sequence) derived from a video camera that monitors a scene so that the system detects the movement of an object in the image sequence. will be. One particular embodiment of the invention is a traffic monitoring system.
    1A and 1B illustrate an optional real-time and non-real-time scheme for coupling a video camera to a traffic monitoring image processor.
    2A, 2B and 2C show image fields of a camera capable of viewing a multi-lane road.
    3 is a block diagram illustrating a preprocessing portion of a digital image processor of the present invention.
    4 illustrates a detection and tracking portion of the digital image processor of the present invention.
    5 and 5a show how image pixels of a 2D region drawn by lines are integrated into a 1D strip.
    6 is a flow chart of a process of updating a reference image.
    7 is a flowchart of a process for modifying a reference image.
    8 is a block diagram of a 2D to 1D converter.
    9 shows an alternate image filter.
    The present invention relates to an improved digital image processing technique applied to an automobile traffic monitoring system that includes a video camera with a predetermined field of view for recording successive image frames of road traffic in a field of view. The digital image processing means responds to the pixel information defined by each successive image frame.
    In particular, the digital image processing technique is adapted to derive a stored initial reference image that defines only fixed objects in the field of view and to update the stored initial reference image with a reference image derived from an image frame recorded later than the initial train. A first means responsive to an initial train, wherein the digital amplitude level of each pixel of the reference image is determined by illuminance conditions present when the initial train and the late recorded frame are recorded; Second means for modifying a digital amplitude level of one of the stored reference image and one of the current image frames stored so that corresponding pixels defining a fixed object are equal to each other; Third means responsive to a digital amplitude level difference between the corresponding pixels of each frame of the continuous image frame and a corresponding reference pixel of the stored reference image for deriving a continuous image defining only a moving object in the field of view; Fourth means for identifying between a moving object having a position fixed to each other with respect to each of the continuously occurring images and a moving object having a change in position with respect to each of the continuously occurring images; And each of the fixed moving objects that define non-physical moving objects such as shadows and reflective headlights cast by the physical moving objects from the fixed moving objects that define the physical moving objects. In order to distinguish and remove each of the moving objects defining the non-physical moving object, the moving objects having fixed positions to each other comprise fifth means for responding to changes in the digital amplitude level of each pixel.
    The invention includes at least one video camera for deriving a continuous image frame of road traffic and a traffic surveillance image processor for digitally processing pixels of the continuous image frame. As shown in FIG. 1A, the output of video camera 100 may be provided directly as an input of traffic monitoring image processor 102 to digitally process pixels of consecutive image frames in real time. Optionally, as shown in FIG. 1B, the output of video camera 100 is first recorded by video cassette recorder (VCR) 104 or other type of image recorder. The pixels of the continuous image frame can then be read by the VCR and provided as input to the traffic monitoring image processor 102 to digitally process the pixels of the continuous image frame.
    Video camera 100 may be a charge coupled device (CCD) camera, an infrared (IR) camera, or another sensor that generates an image sequence. In an embodiment of the traffic monitoring system of the present invention, the camera is mounted on the road at a predetermined height and has a predetermined field of view for a predetermined length segment of the road. As shown in FIGS. 2A and 2B, the video camera 100 may be installed on the road at a height of 30 bits, extending from 50 feet to 300 feet relative to the protruding position of the video camera 100 on the road. Have a 62 ° field of view sufficient to observe the 60 foot width (5 lanes) of the length segment. 2C shows that the video camera 100 derives a 640x480 pixel image of the road portion within the field of view. Automobile traffic present on the length segment of the road has been omitted from the image of FIG. 2C.
    In the designed car traffic monitoring system, video camera 100 is one of a group of four dividing cameras, each of which is operated at a frame rate of 7.5 frames per second.
    It is a principal object of the present invention to provide an efficiently calculated digital traffic monitoring image processor capable of more accurately detecting, counting and tracking a vehicle moving over a certain length segment of the roadway. This can, for example, detect and track errors or increase the efficiency of the calculation.
    1. Low contrast;
    The car should be detected based on its contrast to the background road surface. This contrast allows the car to have a reflected light intensity similar to the reflected light intensity of a road. In weak light conditions and cloudy days, detection errors may occur. At this time, the system may miss some cars, or if the threshold threshold for detection is low, the system can interpret some background patterns as road cars as well as cars.
    2. Shadow and headlight reflected light:
    At some time of the day, the car may be obscured by shadows and cause headlight reflected light to cross adjacent lanes. Such shadows or headlight reflected light will have greater contrast than the vehicle itself. Conventional traffic monitoring systems can interpret the shadow as an additional vehicle, causing an overcount of traffic flow. The shadow of a large car, such as a truck, may completely obscure a passenger car or a motorcycle, so that shady cars may not be counted. Shadows can be caused by objects that are not on the road, such as trees, buildings, and clouds. Shadows can be caused by cars traveling in different directions on different roads. Again, the shadows can be misinterpreted as additional cars.
    3. Camera Vibration:
    The camera installed on the utility pole can be moved by the vibration of the pole by the wind. Cameras mounted on bridges on the main street can be vibrated as the truck passes over them. In either case, camera movement can cause movement of the image, which can cause detection and tracking errors. For example, camera vibrations can confuse lanes in the detection process and cause problems that seem like a stationary car is moving.
    4. Efficiency of calculation
    Since vehicle movement is limited to lanes and the general direction of travel is one-dimensional along the length of the lane, using two-dimensional image processing when detecting and tracking traffic volume of the vehicle is computationally inefficient.
    The present invention relates to an image processor integrated in a traffic monitoring system comprising means for solving one or more of the above four problems.
    3 shows a basic block diagram of a preferred embodiment of the digital traffic monitoring image processor 102 of the preprocessor portion. 3 shows an analog-to-digital (A / D) converter 300, pyramid means 302, stabilization means 304, reference image derivation and update means 306, frame storage 308, reference image correction means ( 310 and subtractor 312 are shown.
    The analog video signal input to the camera 100 or the VCR 104 after being digitized by the A / D 300 is analyzed to a certain number of Gaussian pyramid levels by the pyramid means 302 which reduces pixel density and image resolution. Can be The pyramid means 302 is not essential because the motor vehicle traffic system can be operated at a resolution of the pixel density generated by the video camera 100 (eg 640x480). However, since the resolution is higher than required for the present vehicle traffic system, the use of pyramid means 302 increases the system computational efficiency. All pyramid levels should not be used for each calculation. Moreover, all levels of the pyramid need not be remembered between each calculation since the highest level can be calculated from the lowest level. However, it is assumed that all particular numbers of Gaussian pyramid levels are available for each downstream calculation described below.
    The first calculation of these downstream is carried out by the stabilization means 304. Stabilization means 304 utilizes electronic image stabilization to compensate for problems with camera vibrations that can be induced in wind or truck movement. By moving the camera, the pixels of the image can be moved. Conventional vehicle traffic systems that do not compensate for camera movement will cause false position detection when the camera moves so that an image of an automobile or surface marking in the adjacent lane overlaps the detection area. Stabilization means 304 compensates for the movement of the image from frame to frame due to camera rotation about an axis perpendicular to the direction of attention. Compensation is performed by shifting the current image with a certain number of columns and rows such that the image is guided by means 306 and aligned with the reference image stored in frame storage 308 despite camera progress. The required shift is determined by placing two known target features in each frame. This is done through a matching filter.
    The problem of low contrast is overcome by the cooperative operation of the reference image derivation and updating means 306, the frame storage 308 and the reference image correction means 310. The means 306 generates a raw reference image r 0 by blurring the first generated image frame i 0 input from the means 304 with a large Gaussian filter (so that the reference image r 0 can contain a high pyramid level), The reference image r 0 is stored in the frame memory 308. Following this, the image stored in the frame storage 308 is updated by means 306 during the first initialization phase. In particular, the means 306 performs a cyclical temporal filtering operation on each corresponding pixel of the first several image frames of the successively stabilized image frames input to the means 304, between the reference image and the current image. If the difference is too large, the reference image is not updated in pixels. Mathematically,
    Where t t represents the reference image after frame t and i t represents the t th frame of the input image frame sequence from means 304. The constant V determines the reliability of the construction process.
    6 shows a flowchart of an example process 600 for executing equation 1 in a real system. That is, FIG. 6 describes the operation of the means 306 shown in FIG. 3 as described above. In particular, the reference image and the next image in the sequence are input to the means 306 in step 602. The reference image is a previously generated reference image, or if this is an initial reference image, the reference image is an unsharp version of the first image frame in the image sequence. In step 604, the pixel is selected from the reference image r t-1 (x, y) and the pixel is selected from the next image i t (x, y). Next, in step 606, the pixel value of the reference image is subtracted from the pixel value of the next image generating a differential factor (DIFF = i t (x, y)). At this time, the process calculates the absolute value of the differential factor | DIFF | in step 608. In step 606, the absolute value of the differential factor is compared to the threshold D. The process asks whether the absolute value of the differential factor is less than the threshold. If the question is answered negatively, the process proceeds to step 614. In step 612, the process updates the selected reference image pixel value with the same pixel value as the selected reference image pixel value multiplied by the update factor U. The update factor is a differential factor multiplied by the constant γ. In step 614, the process queries whether all pixels of the image have been processed. If not, the process returns to step 604. If all the pixels have been processed, the process ends at 616.
    The reliability setting of γ is a stationary background object where the reference image stored in the frame memory 308 can be seen by the camera 100 at the end of some image frames input to the means 306 including the first initialization phase. It should be slow enough to maintain a reference image of a temporary object, such as a moving car or a car that can be temporarily stopped by traffic jams to include only the bay. Such a reliability setting of γ cannot adjust r t quickly enough to be able to add a change in illumination (due to the cloud or auto-iris of camera 100) to the reference image. This problem is solved at the end of the initialization phase by the cooperative update operation of the reference image correcting means 310 (including the illumination / AGC compensator), the means 306 and the frame memory 308. In particular, when the initialization phase (as shown in Fig. 6) is completed, the initialization phase is replaced by a general second operating phase operating according to the following equation 2 (rather than equation 1).
    k t and c t are the estimated gain and offset between the reference image r t and the current image i t calculated by means 310. The means 310 calculates the gain and offset by plotting the cloud as a point and plotting the cloud as a line in 2D space where the x axis represents the blurry level intensity of the reference image and the y axis represents the blurry level intensity of the current image. The cloud is a set of points (r t-1 (x, y), i t (x, y)) for all image positions x, y. This approach is implemented using a method for calculating gains and offsets indicative of changes in illumination. For example, the gain can be estimated by comparing the histograms of the current image and the reference image. In addition, certain update rules do not need to use an absolute threshold D as described above. Instead, the update can be weighted by any function of | i t (x, y) -r t-1 (x, y) |.
    7 shows a flowchart of an example process 700 for executing equation 2 in a real system. 7 illustrates the operation of the means 310 of FIG. 3. The reference image and the next image in the sequence are inputs at step 702. This is the first time the means 310 is used, the reference image is the final reference image generated by the means 306; On the other hand, it is the previous reference image generated by the means 310. In step 704, the pixel is selected from the reference image r t-1 (x, y) and from the next image i t (x, y). Subsequently, in step 706, the pixel value of the reference image is subtracted from the pixel value of the next image to calculate the differential factor DIFF = (i t (x, y))-(r t-1 (x, y)). The process then calculates the absolute value of the differential factor (DIFF) at step 708. At step 710, the absolute value of the differential factor is compared to the threshold value D. The process determines that the absolute value of the differential factor is the threshold value. If the answer is negative, the process proceeds to step 714. In step 714, the process multiplies the selected reference image pixel value by a gain k t and scales by an offset c t . Transform to a pixel value equal to the selected reference image pixel value. If the answer to the question is affirmative, the process proceeds to step 712. In step 712, the process selects the selected reference image pixel value multiplied by the modification factor (M). Same as the reference image pixel value The transform factor is a differential factor multiplied by a constant γ If the process modifies a pixel value, the process asks whether all the pixels in the image have been processed in step 716. Otherwise, the process returns to step 704 Once all the pixels have been processed, the process ends at step 716.
    This approach allows for fast illumination changes to be added to the reference image while preventing temporary objects from being added. This is done by providing the coordination means with flexibility to determine whether the new reference image pixel value should be calculated as a function of the pixel value in the current image or simply by applying gain and offset to the current reference image. By applying the gain and offset to the current reference image, the illuminance change can be simulated without the risk of temporary objects appearing in the reference image.
    The results of FIG. 3 indicate that the fixed background display pixels (including both fixed background display pixels and moving objects (eg, vehicle traffic)) of the illuminance-compensated current image appearing at the output of the means 310 are always present in the frame storage 308. It is the same as the fixed background display pixel (including only the fixed background display pixel) of the reference image shown in the output. Therefore, the subtractor 312, which calculates the difference between the amplitudes of the corresponding pixels applied as inputs from the means 310 and 304, consists of a significant value of pixels representing only moving objects alone in each one of the successive 2D image frames. Induce the output. The output of the subtractor 312 proceeds to the detection and traffic portion of the traffic-monitoring image processor 102 shown in FIG.
    4 shows a 2D / 1D converter 400, a vehicle fragment detector 402, an image flow evaluator 404, a single frame delay 406, a pixel-amplitude squared means 408, a vehicle hypothesis generator 410, and a shadow. And reflected headlight filter 412 is shown.
    The 2D / 1D converter 400 operates to convert the 2D image information received from FIG. 3 applied as the first input into the 1D image information according to the user information applied as the second input. In this regard, reference is made to FIGS. 5 and 5A. FIG. 5 shows an image frame 500 in which the cars 504-1 and 504-2 traveling on the second lane 506 from the left are guided by the camera 100 of a straight five-lane road 502. It is. Vehicles 504-1 and 504-2 are located in image zone 508 represented by user control information applied as a second input to converter 400. By integrating the amplitude of the pixels horizontally across the image zone and subsampling the integrated pixels in the vertical direction along the center of the zone 508, the 1D strip 510 is calculated by the converter 400. The road need not be straight. As shown in FIG. 5A, curved road lane 512 includes a zone 514 defined by a user-marked lane boundary 516 that allows calculation of the center strip 518 by converter 400. . In Figures 5 and 5A, a user may use lane definition stripes that may exist in the image as landmarks to help define user-marked lane boundaries.
    More specifically, the calculation by converter 400 involves using each pixel position (x, y) to define an integrated window. For example, such a window may be (a) all image pixels on column y within the indicated lane boundaries, (b) all image pixels on row x within the indicated lane boundaries, or (c) a center strip at position (x, y). It may be all pixels on the lane perpendicular to the tangent of. Other types of integration windows not shown here may be used. 8 shows a block diagram of a 2D / 1D converter 400 including a zone defining block serially connected to an integrator 804 serially connected to an image sub-sampler 806. User input defining a zone in the 2D image is applied to a zone defining block 802.
    In FIG. 4, the 1D output from converter 400 is applied as input to filter 412 through detector 402, evaluator 404 and single frame delay 406, and means 410. Since each detection, tracking and filtering function performed by these elements works on a 1D or 2D signal, 1D operation is preferred because it significantly reduces the computational requirements. Therefore, the presence of converter 400 is not essential to the performance of the detection, tracking, and filtering functions. Next, the existence of the converter 400 will be described.
    Detector 402 uses a multi-level pyramid to provide coarse-to-fine operation for detecting the presence and partial position of a vehicle fragment in a 1D strip of a continuous image frame received from FIG. 3. do. Fragments are defined as groups of significant values of pixels at certain pyramid levels connected to each other. Detector 402 allows each vehicle to maximize the chance of being generated in a single fragment. In practice, however, it is impossible to achieve that each vehicle generates multiple fragments (like separate fragments corresponding to the hood, loop, and headlights of the same vehicle). Moreover, more than one pixel can be connected in a single fragment.
    One of the object detection techniques at each pixel strip position is to calculate a histogram of the image density values in the integrated window centered at that pixel position. Based on the contribution of this histogram (i.e., the number or percentage of pixels across a certain threshold), the strip pixels are classified as detection or background. By performing this operation on each strip pixel, it is possible to construct a one-dimensional array including detection or background for each pixel position. By performing component analysis connected within this array, adjacent detection pixels are grouped into fragments.
    The image flow evaluator 404 cooperates with the delay 406 to keep the object tracked over time. In short, in this case, this involves calculating and storing the average value contained in the integrated window at each pixel location. By performing this operation in each strip pixel, a one-dimensional array of average luminance values is constructed. Given two corresponding arrays of images obtained at t-1 and t times, a one-dimensional image flow is calculated that maps pixels from one array to another. This can be calculated through one-dimensional least-squares minimization or one-dimensional patchwise correlation. This flow information can be used to track objects between successive image frame pairs.
    Each output of detector 402 and evaluator 404 is applied as an input to vehicle hypothesis generator 410. Near fragments are grouped together as part of the same object (ie, vehicle) if they move in a similar way or are very close together. If the positions of multiple fragments remain fixed relative to each other in each of the trains of a continuous frame, they are assumed to indicate only a single vehicle. However, if the position of the fragment changes from frame to frame, they are assumed to be holding a separate vehicle. Moreover, if a single fragment breaks up into multiple fragments in one frame or the shape is stretched out in lengthwise from one frame to another, they are inferred to indicate a separate vehicle.
    At night, the presence of a vehicle can only be indicated by its headlights. Headlights tend to generate headlight reflected light on the roadway. Lighting conditions on the road during night and day tend to cause vehicle shadows on the road. Such shadows and headlight reflected light on the road generate detected fragments that may appear in the generator 410 as additional vehicles, thereby producing a defect positive error in the output from the generator 410. The shadow and headlight reflected light filter 412 that identifies between the fragment generated by the appropriate vehicle and the fragment generated by the reflected headlight and the shadow eliminates this defect positive error.
    The output from the pixel-amplitude squared means 408 represents the relative energy at each pyramid level pixel of the strip output of each successive image frame from the converter 400. The filter 412 identifies between the fragments generated by the appropriate vehicle and the fragments generated by the shadows and reflected headlights based on the analysis of the relative amplitudes of the energy indication pixels from the means 408. The fact that the change in the energy pixel amplitude (pixel brightness) of the shadow and headlight reflected light fragments is significantly lower than the change in the energy pixel amplitude of the appropriate vehicle fragment can be used as identification.
    Another filtering method shown in FIG. 4 is to use a converter 400 to identify between objects and shadows using background-adjusted reference images. 9 shows a block diagram of another filter. At each pixel location, the following information is calculated via the integrated window:
    (a) the number of pixels (element 904) having a luminance value greater than any threshold p, for all image pixels in the integrated window;
    (b) a maximum absolute value (element 906) for all image pixels in the integrated window;
    (c) the absolute difference I (x 1, y 1) - the number of I (x 2, y 2) are adjacent pixels in the integration window to exceed a threshold value (x 1, y 1), (x 2, y 2) (Element 908).
    This information is used by filter 902 to identify between objects and shadows using background-adjusted reference images.
    The fragments extracted as described above may be classified as objects or shadows based on these or other properties. For example, if the value of the added measure (a) for all strip pixels in the fragment exceeds the threshold, the fragment may not be a shadow (the shadow may be in the image applied to the converter 400 from FIG. 4). Because it never has a positive luminance value). Similar addition using measure (c) provides another test to measure the amount of texture in the fragment that can be thresholded to determine whether the fragment is an object or a shadow. Since the input to the filter 412 defines all hypothesized vehicle positions, the output from it defines only validated vehicle positions. The output from the filter 412 is provided to use means (not shown) that can perform the function of calculating speed and length and counting the vehicle.
    The vehicle fragment detector 402, the image flow evaluator 402, and the vehicle hypothesis generator 410 may use predetermined camera calibration information in operation. Moreover, each of the various techniques of the present invention described above may be used to benefit other types of imaging systems from the vehicle traffic monitoring system described herein.
    权利要求:
    Claims (11)
    [1" claim-type="Currently amended] In the image processing method,
    (a) supplying an image frame of a first sequence;
    (b) initializing a reference image that includes image information about a stationary object in the scene represented by the first sequence of images;
    (c) supplying a next image frame temporarily following the image frame of the first sequence;
    (d) comparing the reference image with the next image frame to produce a two-dimensional motion image representing motion information relating to movement of an object in the scene;
    (e) updating the reference image with information in the next image frame, wherein the information used to update the reference image merely represents a stationary object in the scene and represents a moving object and a temporary stationary object in the scene. Ignores; And
    (f) repeating steps (c), (d), and (e) for each subsequent image supplied.
    [2" claim-type="Currently amended] The method of claim 1, wherein the updating step,
    Selecting a pixel value of the next image frame having the same pixel position as the pixel value of the reference image and the selected pixel value of the reference image;
    Calculating a differential factor equal to an absolute value of the difference between the selected pixel value of the next image frame and the pixel value of the reference image;
    Comparing the difference with a first threshold;
    If the difference is less than the first threshold, replacing a pixel value of the reference image with an updated reference pixel value equal to the selected reference image pixel value changed by an update factor; And
    If the difference is greater than or equal to the first threshold, the pixel values of the reference image remain the same.
    [3" claim-type="Currently amended] 3. The method of claim 2, wherein the update factor is a difference between the selected pixel value of the next image and the pixel value of the reference image multiplied by a response factor, wherein the response factor is such that the updated reference image merely identifies a fixed object in the scene. Limiting the amount of update to ensure that it contains information about the information in the next image frame that it represents.
    [4" claim-type="Currently amended] 4. The method of claim 3, further comprising transforming the reference image using image information in the next image frame regarding a change in illumination of the scene.
    [5" claim-type="Currently amended] The method of claim 4, wherein the modifying step,
    Selecting a pixel value of the next image frame having the same pixel position as the pixel value of the reference image and the selected pixel value of the reference image;
    Calculating a differential factor equal to an absolute value of the difference between the selected pixel value of the next image frame and the pixel value of the reference image;
    Comparing the difference with a second threshold;
    If the difference is less than the second threshold, replacing a pixel value of the reference image with an updated reference pixel value equal to the selected reference image pixel value changed by an update factor; And
    And if the difference is greater than or equal to the second threshold, replacing the selected reference image pixel value with a new reference image pixel value equal to the weighted and scaled selected reference image pixel value.
    [6" claim-type="Currently amended] 6. The method of claim 5, wherein the distortion factor is the difference between the selected pixel value of the new image frame and the selected pixel value of the reference image multiplied by the response factor, wherein the response factor is such that the modified reference image is only within the scene. Limiting the amount of deformation to ensure that it contains information about the information in the new image representing the stationary object, wherein the weighted and selected scaling reference image image represents a change in illuminance in the scene.
    [7" claim-type="Currently amended] In the image processing apparatus,
    Image means (100) for supplying a continuous sequence of image frames representing a scene;
    Reference image initialization means (306) connected to said image means for initializing a reference image including image information about a stationary object in said scene;
    Comparator means connected to said image means and said reference image initialization means for comparing said reference image with an image frame supplied by said image means to produce a two-dimensional motion image representing motion information relating to movement of an object in said scene ( 312); And
    Means 306 for updating the reference image with information in the image frame, wherein the information used to update the reference image represents only fixed objects in the scene and ignores moving objects and temporary fixed objects in the scene. Device.
    [8" claim-type="Currently amended] 8. The apparatus of claim 7, wherein the method further comprises means (310) for modifying the reference image using image information in the next image frame regarding a change in illumination of the scene.
    [9" claim-type="Currently amended] A vehicle traffic monitoring system comprising a video camera having a predetermined field of view for recording continuous image frames of road traffic in a field of view, and digital image processing means responsive to pixel information defined by each of the continuous image frames.
    A first response responsive to the initial train of the continuous image frame to derive a stored initial reference image defining only a fixed object within the field of view and to update the stored initial reference image with a reference image derived from an image frame recorded later than the initial train. Means 306, wherein the digital amplitude level of each pixel of the reference image is determined by illuminance conditions present when the initial train and the late recorded frame are recorded;
    Second means (310) for modifying the digital amplitude level of one of the stored reference imposition and the current image frame stored such that the corresponding pixels defining the fixed object are identical to each other;
    Third means (404) responsive to a digital amplitude level difference between the corresponding pixel of the stored reference image for deriving a continuous image defining each frame of the continuous image frame and only a moving object in the field of view;
    Fourth means (402, 410) for discriminating between moving objects that remain fixed in position with respect to each of the continuously occurring images and moving objects whose positions change with respect to each of the successively occurring images; And
    Distinguish between each of the fixed moving objects that define a physical moving object and each of the fixed moving objects that define non-physical moving objects such as shadows and headlight reflections cast by the physical moving object. And fifth means responsive to a change in the digital amplitude level of each pixel of said moving object that is fixed in position to remove each of said moving objects defining said non-physical moving object. Device.
    [10" claim-type="Currently amended] 10. The method of claim 9, wherein the first means and the second means comprise a first set of equations, i.e.

    Cooperate during the generation of the initial train of the continuous image frame according to an equation, and then a second set of equations, i.e.

    Cooperate according to the equation,
    Where (x, y) represents the coordinates of each pixel in the image frame, r t represents the reference image after frame t, i t represents the t-th frame of the continuous image frame, and D represents the angle between i t and r t represents a predetermined difference between the pair of corresponding pixels amplitude level of a, k and c t represents the gain and offset values estimated for a single pair according to the amplitude level for a pixel of i t and r t i t and r t The corresponding pixel of reduces the statistical error in each difference in amplitude levels between the corresponding pixels of i t and r t , and g is the first and second means during and after the occurrence of the initial train of continuous image frames. Apparatus characterized in that the constant to determine the response of the cooperation.
    [11" claim-type="Currently amended] 11. The method of claim 10, wherein k and c t are single in amplitude level for the i t and r t a of the pixel which minimizes the statistical error in the difference in amplitude levels between corresponding pixels of i t and r t, respectively Indicating an estimated gain and offset of the pair.
    类似技术:
    公开号 | 公开日 | 专利标题
    JP6301828B2|2018-03-28|Apparatus for measuring the speed and position of a vehicle moving along a guiding track, and corresponding method and computer program product
    US20200065594A1|2020-02-27|Path sensing using structured lighting
    US7209832B2|2007-04-24|Lane recognition image processing apparatus
    US6687386B1|2004-02-03|Object tracking method and object tracking apparatus
    US6445409B1|2002-09-03|Method of distinguishing a moving object and apparatus of tracking and monitoring a moving object
    JP4788798B2|2011-10-05|Object detection device
    US8890951B2|2014-11-18|Clear path detection with patch smoothing approach
    Zheng et al.2013|A novel vehicle detection method with high resolution highway aerial image
    US20190294893A9|2019-09-26|Monocular cued detection of three-dimensional structures from depth images
    US8452053B2|2013-05-28|Pixel-based texture-rich clear path detection
    US6546115B1|2003-04-08|Method of updating reference background image, method of detecting entering objects and system for detecting entering objects using the methods
    DE69736764T2|2007-01-25|Local positioning device and method therefor
    Gloyer et al.1995|Video-based freeway-monitoring system using recursive vehicle tracking
    EP0945319B1|2001-05-30|Process for determining dynamic properties of motor vehicles
    US7623700B2|2009-11-24|Stereoscopic image processing apparatus and the method of processing stereoscopic images
    JP4856612B2|2012-01-18|Object detection device
    US5311305A|1994-05-10|Technique for edge/corner detection/tracking in image frames
    CN102867414B|2014-12-10|Vehicle queue length measurement method based on PTZ | camera fast calibration
    DE112011103690T5|2013-09-05|Detection and tracking of moving objects
    US7612800B2|2009-11-03|Image processing apparatus and method
    US8045761B2|2011-10-25|Detection of environmental conditions in a sequence of images
    TWI420906B|2013-12-21|Tracking system and method for regions of interest and computer program product thereof
    JP4970516B2|2012-07-11|Surrounding confirmation support device
    EP2087459B1|2018-03-21|Device, method, and computer program for determining a position on the basis of a camera image
    KR100792283B1|2008-01-07|Device and method for auto tracking moving object
    同族专利:
    公开号 | 公开日
    DE69635980T2|2007-01-25|
    WO1996022588A1|1996-07-25|
    US5847755A|1998-12-08|
    EP0804779B1|2006-03-29|
    CA2211079A1|1996-07-25|
    KR100377067B1|2005-09-08|
    DE69635980D1|2006-05-18|
    ES2259180T3|2006-09-16|
    JPH10512694A|1998-12-02|
    EP0804779A1|1997-11-05|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    法律状态:
    1995-01-17|Priority to US37292495A
    1995-01-17|Priority to US08/372,924
    1995-01-17|Priority to US8/372,924
    1996-01-17|Application filed by 윌리암 제이. 버크, 데이비드 사르노프 리서치 센타, 인코포레이티드
    1998-05-15|Publication of KR19980701568A
    2005-09-08|Application granted
    2005-09-08|Publication of KR100377067B1
    优先权:
    申请号 | 申请日 | 专利标题
    US37292495A| true| 1995-01-17|1995-01-17|
    US08/372,924|1995-01-17|
    US8/372,924|1995-01-17|
    [返回顶部]