WO2022215409A1 - 物体追跡装置 - Google Patents
物体追跡装置 Download PDFInfo
- Publication number
- WO2022215409A1 WO2022215409A1 PCT/JP2022/009748 JP2022009748W WO2022215409A1 WO 2022215409 A1 WO2022215409 A1 WO 2022215409A1 JP 2022009748 W JP2022009748 W JP 2022009748W WO 2022215409 A1 WO2022215409 A1 WO 2022215409A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- detection frame
- detection
- information
- frame
- unit
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/277—Analysis of motion involving stochastic approaches, e.g. using Kalman filters
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/70—Labelling scene content, e.g. deriving syntactic or semantic representations
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20076—Probabilistic image processing
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30241—Trajectory
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30248—Vehicle exterior or interior
- G06T2207/30252—Vehicle exterior; Vicinity of vehicle
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Definitions
- the present invention relates to an object tracking device.
- trajectory information viewed from a bird's-eye view can be easily visualized by projecting it on a two-dimensional map, and can also be used for work behavior management and abnormal motion detection.
- a method of generating the trajectory there is a method of calculating the distance from the camera to the detection target in each frame of the camera image, acquiring three-dimensional position information, and integrating the position information for each frame.
- distance calculation a method using circumscribed rectangle information of a target obtained by detecting the target from a camera image and camera parameters is generally used.
- the camera parameters can be estimated by photographing in an ideal environment in advance, but the circumscribing rectangle information must be estimated from the camera image obtained at the actual photographing site. Therefore, in order to generate a highly accurate trajectory, there is a need for a technique for accurately estimating the position of a circumscribed rectangle (hereinafter referred to as a detection frame) even in various shooting locations.
- a detection frame a technique for accurately estimating the position of a circumscribed rectangle
- a detection frame there is a method using dictionary information created by machine learning such as a convolutional neural network.
- Machine learning requires the creation of training data in advance, and while it would be nice if a wide variety of scenes could be learned, the number of scene variations would be limited given the realistic number of man-hours.
- Patent Document 1 proposes a method of correcting detection frame information by performing object tracking processing and object recognition processing in parallel.
- Patent Document 2 proposes a method of extracting image feature points and estimating a motion vector. proposed a method for correcting the position of the rectangle when the detection target is blocked by an obstacle.
- Patent Literature 1 a detection target is tracked by object tracking processing to generate a trajectory of the same target, and high-precision object recognition is performed at predetermined frame intervals to correct the ID information of a rectangle, thereby obtaining trajectories of multiple targets.
- a technique for calculating with high accuracy is described, it is not possible to correct the size and position information of the rectangle.
- the position of the rectangle can be corrected by predicting the position of the rectangle in the next and subsequent frames from the motion vector information obtained from the image feature within the circumscribing rectangle of the detection target, but the size of the rectangle cannot be corrected. can't.
- the present invention is an invention for solving the above problems, and is an object tracking device that generates a trajectory of an object within a measurement range by correcting the detection frame position according to the reliability calculated from the flow information. , to output a highly accurate trajectory of a detection target.
- an object tracking device that generates a trajectory of an object within a measurement range of a sensor, an object detection unit that detects an object in each of a plurality of frames acquired by the sensor; a detection frame reliability calculation unit that calculates the reliability of the detection frame based on flow information between frames of the detection frame in which the object is detected; a detection frame position correction unit that corrects detection frame information of a low-reliability detection frame with a lower reliability than the high-reliability detection frame using detection frame information of a high-reliability detection frame with a reliability higher than a threshold; , a trajectory generating unit that generates a trajectory of the object using the corrected detection frame information; characterized by having
- FIG. 1 is a functional block diagram of Embodiment 1 of the present invention
- FIG. 4 is a diagram for explaining an object detection unit 4
- FIG. FIG. 2 is a configuration diagram of an existence probability map determination unit 5
- 4 is a diagram for explaining an existence probability map creation unit 21
- FIG. An example of how to create an existence probability map is a diagram for explaining an existence probability map interpolation unit 22
- FIG. The figure explaining the flow which adds a detection frame. 4 is a diagram for explaining a flow calculation unit 7
- FIG. FIG. 5 is a diagram for explaining a detection frame reliability calculation unit 8
- FIG. 2 is a configuration diagram of a detection frame position correction unit 9
- FIG. 4 is a configuration diagram of a detection target moving direction prediction unit 91.
- FIG. The functional block diagram of Example 3 of this invention is a diagram showing Equation 1;
- FIG. 2 is a diagram showing Equation 2;
- FIG. 3 is a diagram showing Equation 3;
- FIG. 4 is a diagram showing Equation 4;
- FIG. 1 is a block configuration diagram of Embodiment 1 of the present invention.
- a camera is used as the measurement device
- the measurement device is not limited to this, and can be applied to other sensors such as a stereo camera and a distance sensor.
- the object tracking device 1 shown in FIG. 1 corrects the detection frame position based on presence probability map information generated from the object detection results of all frames captured by the camera 2 and detection frame reliability information calculated from flow information, etc. It is a device that generates a highly accurate trajectory.
- the camera 2 is attached, for example, to the vehicle body of an automobile, and images the road surface in front of the vehicle and other vehicles.
- the object tracking device 1 is a device that acquires an image captured by a camera 2 and analyzes the trajectory of an object in front of a vehicle that is a detection target by performing analysis processing.
- the object tracking device 1 is implemented in a computer PC prepared separately from the camera 2 .
- the computer PC has an arithmetic device, a main storage device, and an external storage device, and includes a frame collection unit 3, an object detection unit 4, an existence probability map determination unit 5, a detection frame adjustment determination unit 6, a flow calculation unit 7, detection
- the functions of a frame reliability calculation unit 8, a detection frame position correction unit 9, and a trajectory generation unit 10 are realized.
- the object tracking device 1 may be provided integrally inside the camera 2 .
- the existence probability map determining unit 5 generates or interpolates an existence probability map of a detection target from the object detection result of each frame; the detection frame adjustment determining unit 6 uses the existence probability map information to perform detection in each frame.
- the flow calculator 7 has a function of calculating flow information between frames, and the detection frame reliability calculator 8 calculates the final detection frame output from the detection frame addition/subtraction decision unit 6.
- the detection frame position correcting unit 9 corrects the detection frame position and size according to the reliability of the detection frame, and the trajectory generation unit 10 corrects the detection It has a function of generating a trajectory of a detection target from frame information. Details of the functions 3, 4, 5, 6, 7, 8, 9, and 10 will be described below.
- the frame collection unit 3 collects frames that are captured images captured by the camera 2 .
- the frames to be collected may be a plurality of frames in which detection targets exist, or may be frames selected automatically or manually from all the frames.
- a selection method there is a method of automatically extracting frames within the specified shooting time, a method of automatically extracting frames in which the detection target exists within the measurement range in the video specified by the user, a GUI (Graphical User Interface), etc.
- There is a method of manually selecting by the user through and the method is not particularly limited.
- the object detection unit 4 detects a detection target for each of the plurality of frames collected by the frame collection unit 3 .
- 11 is a captured image (frame)
- 12 and 13 are detection targets (a person and a car in this example)
- 14a and 14b are detection frames estimated by the object detection unit 4 to include the target object from the image.
- An example, 15 shows an example of detection frame information.
- the object detection unit 4 detects a detection target existing in the captured image by utilizing a phenomenon in which the object can be detected in the image generated in advance from learning data by machine learning or the like.
- Algorithms for generating a dictionary for detection may be general algorithms such as convolutional neural networks and AdaBoost, and are not particularly limited.
- the object detection unit 4 detects an object from the plurality of frames acquired by the frame collection unit 3 using the dictionary, and stores detection frame information as indicated by 15 in FIG.
- the detection frame information includes, for each detected object, class information indicating the type of the object, reliability information indicating the accuracy of the class information, starting points (X, Y) indicating the upper left image coordinates of the detection frames 14a and 14b, There is information such as the width and height of the detection frame. Note that any method other than this example may be used as long as it can estimate the position of the detection target in the image.
- FIG. 3 shows a block configuration of the existence probability map determination unit 5.
- the existence probability map determination unit 5 performs processing for generating an existence probability map indicating the existence probability of each detection target in a plurality of frames.
- the existence probability map determination unit 5 includes an existence probability map creation unit 21 that creates an existence probability map for each frame from the detection frame information output from the object detection unit 4, and an existence probability map creation unit 21 that creates an existence probability map for each frame from the detection frame information output from the object detection unit 4.
- An existence probability map interpolation unit 22 that interpolates map information and generates an interpolated existence probability map of a target frame, and compares the existence probability map and the interpolated existence probability map to select a final existence probability map for each frame.
- An existence probability map selection unit 23 is provided. 21, 22 and 23 will be described in detail below.
- the existence probability map creating unit 21 will be described with reference to FIG.
- the existence probability map creating unit 21 generates an existence probability map indicating the existence probability of each object in a plurality of frames.
- the existence probability map creating unit 21 calculates an existence probability map for each class to be detected from the detection frame information for each frame.
- the existence probability map creation unit 21 creates either a probability density function calculated from a normal distribution from the position information of the detection frame of the object detected by the object detection unit 4 and the reliability information, or an object detection result from a plurality of different dictionaries. The one or more pieces of information are used to generate a presence probability map.
- the captured image 11 is divided into a plurality of small regions such as 31a and 31b, and the degree of overlap between the detection frames 14a and 14b of each class and each small region is calculated by ratio. , 32a, 32b.
- the number of divided small regions is not particularly limited, and may be determined in consideration of the specifications of the computer PC that executes the processing. may be generated.
- the value of the existence probability map is not particularly limited as long as it indicates the ratio of the objects of each class in the small area. You may use the method of utilizing such a probability density function.
- FIG. 5 is a diagram illustrating an example of a method of creating an existence probability map.
- 11 is a captured image
- 41 is a detection frame
- 42 is the center coordinates (x c , y c ) of the detection frame
- 43 and 44 are the X direction of the image coordinate axis used for calculating the existence probability map
- the Y direction is shown.
- the probability density functions f(x) and f(y) for each direction are shown.
- the probability density function f(x) in the X direction the average value shown in Equation 1 in FIG.
- the probability density function f(y) in the Y direction is also calculated by Equation 2 in FIG.
- Equation 3 in FIG. (x, y) the product of f(x) and f(y) is obtained by Equation 3 in FIG. (x, y) can be generated.
- the value of variance ⁇ 2 is determined based on information on the width and height of the detection frame, or if the actual measurement information of the detection target is known in advance, information on the aspect ratio of the width and height, etc. You may use the method of determining based on. A method of adjusting the value of the existence probability of the detection object by multiplying the existence probability map f(x, y) by the reliability information of the detection frame may also be adopted.
- the presence probability map interpolation unit 22 will be described with reference to FIG.
- the presence-probability map interpolator 22 extracts the presence-probability map information of frames t-1 and t+1 before and after frame t, the presence-probability map information of a plurality of previous frames, or the presence-probability maps of a plurality of subsequent frames t+1 and t+2. Based on the information, a process of estimating the presence probability map of the relevant frame (target frame) is performed.
- the interpolation method by the existence probability map interpolation unit 22 is roughly classified into two patterns shown in 51 and 52 in FIG.
- the interpolated existence probability map corresponding to the t-th frame 54 is used to generate the existence probability maps of the frames before and after it, that is, the t-1th frame 53a and the t+1th frame 53b.
- Generate from As a generation method for example, there is a method of calculating the product or average value of the existence probabilities of the existence probability maps of the t ⁇ 1 th frame 53a and the t+1 th frame 53b.
- a vehicle existence probability map in the t ⁇ 1th frame 53a is generated based on the detection frame information of the detection frame 53a1, and the t+1th frame is generated based on the detection frame information of the detection frame 53b1.
- a car existence probability map is generated in the frame 53b of .
- the existence probability map interpolation unit 22 calculates the product or average value of the existence probabilities of the car existence probability map in the t ⁇ 1th frame 53a and the car existence probability map in the t+1th frame 53b. to generate an interpolated presence probability map for the tth frame 54 .
- the interpolated presence probability map corresponding to the tth frame 54 is changed from the presence probability map of the (t+2)th frame 53c to the presence probability map of the (t+1)th frame 53b. estimated based on For example, after estimating the approximate position of the detection frame in the tth frame 54 from the amount of change in the center position, vertical width, and horizontal width of the t+2 and t+1 detection frames 53c1 and 53b1, the t+1th frame The existence probability map corresponding to 53b is enlarged or reduced by linear interpolation or the like according to the amount of change in the size of the detection frame, thereby obtaining the size and position of the detection frame 54a1 in the tth frame.
- interpolated existence probability map There is a method such as calculating an interpolated existence probability map corresponding to .
- the method of calculating the interpolation existence probability map of the t-th frame from the t-th frame before and after or from the t+1-th and t+2-th frames was described.
- the interpolated existence probability map of the frame is calculated from the t-1th frame, which is a plurality of frames before the current frame, and the t-2nd frame. You can calculate.
- the interpolated existence probability map is generated for the first frame by the method shown in 52, and the interpolated existence probability map for the last frame is generated by the reverse flow of the method shown in 52.
- an interpolated existence probability map is generated by the method shown in 51 for the other frames.
- the existence probability map selection unit 23 selects the final existence probability map from the existence probability map and the interpolated existence probability map for each frame.
- a selection method for example, a method of calculating the difference between the existence probability map and the interpolated existence probability map and adopting the values of the interpolated existence probability map only for small regions where the difference is equal to or greater than a threshold, or a method of adopting the values of the existence probability map and the interpolated existence probability.
- the presence probability map determination unit 5 determines the presence probability map for each frame.
- the existence probability map and the interpolated existence probability map are generated for each frame, and then the existence probability map selection unit 23 determines the final value.
- Multiple maps can be generated. For example, except for the first frame and the last frame in the frame group, three types of interpolated existence probability maps are generated by the reverse flow method of 51, 52, and 52, and then the final value is similarly determined by the existence probability map selection unit 23. It is also possible to use a method of increasing the number of frames before and after the interpolation, and a method of increasing the patterns of the interpolation existence probability map.
- the interpolated presence probability map is created for all frames, but a method of generating the interpolated presence probability map only for some frames to correct the presence probability map for the relevant frame may be adopted.
- a method of correcting the presence probability map only for frames whose reliability of detection frame information is less than a threshold or a method of counting the number of detection frames to be detected in each frame and calculating the average value, and then using the number of detection frames other than the average.
- a method of correcting only a certain frame, a method of correcting only a frame selected by a user using a GUI, etc. are not particularly limited.
- a presence probability map generated by acquiring detection frame information of a detection target from a plurality of different dictionaries is treated as an interpolated presence probability map, and a method of correcting the presence probability map by the presence probability map selection unit 23 is used.
- the detection frame addition/subtraction determination unit 6 performs processing for determining addition or deletion of detection frames in each of a plurality of frames using the presence probability map.
- the detection frame addition/subtraction determination unit 6 uses the presence probability map information output from the presence probability map determination unit 5 to determine whether to add or reduce the detection frame in each frame. As a determination method, if there is no detection frame near an area with an existence probability equal to or higher than the threshold, the circumscribed rectangle of the area with an existence probability equal to or higher than the threshold is added as a detection bounding box, and if a detection bounding box exists in an area with an existence probability equal to or lower than the threshold. If so, there is a way to reduce the corresponding detection frame.
- Criteria for judging whether or not a detection frame exists in the existence probability map include the content rate of small areas within the detection frame that satisfy the threshold of the existence probability map, the center coordinates of the detection frame, and the threshold of the existence probability map. There is no particular limitation, such as a method in which a detection frame exists near the existence probability map as long as the Euclidean distance to the small area that satisfies is within a specified range.
- a margin specifying the size of the detection frame is added. You may add the process which enlarges by minutes.
- the existence probability map creation unit 21 divides the t ⁇ 1 and t+1 frames 53a and 53b into small areas 54a and 54b, and generates existence probability maps 55a and 55b from the overlapping ratios with the detection frames 54a1 and 54b1.
- the existence probability map 56 of the t-th frame is estimated by calculating the average of the existence probability maps 55a and 55b.
- the detection frame 57 is determined based on the predetermined threshold value and the value of the existence probability map. In the example shown in FIG.
- each small region of the existence probability map is classified into three classes of less than 0.5, 0.5 or more and less than 0.9, and 0.9 or more. Based on the conditions that the area is not included in the rectangle, all small areas of 0.9 or more are included in the rectangle, and small areas of 0.5 or more and less than 0.9 are included in the rectangle, but not more than half of them are included.
- a detection frame 57 is determined.
- FIG. 8 shows an example of calculating flow information between frames by analyzing the image feature amounts of the t-th frame 11a and the (t+1)-th frame 11b by the flow calculation unit 7 .
- 61 is the road surface
- 62 is the vehicle to be measured in the frame 11a
- 63 is the vehicle to be measured in the frame 11b
- 64a and 64b are examples of image feature points on the vehicle 62
- 65a and 65b are image features on the vehicle 63.
- An example of points, 66 indicates flow information between image feature points 64a and 65a
- 67 indicates flow information between image feature points 64b and 65b.
- Flow information is a movement vector of the same image feature point between frames, and indicates the direction and magnitude of the vector.
- the same image feature points between the previous and next frames can be obtained by using optical flow by the Lucas-Kanade method, or by tracking using image feature values such as SHIFT. and its flow information can be calculated.
- the flow information for frames other than the first frame is calculated from the corresponding frame (tth frame) and the previous frame (t-1th frame), and the flow information for the first frame is calculated from the corresponding frame and the next frame (t+1th frame). Obtained by inverting the sign of the number.
- the detection frame reliability calculation unit 8 will be described with reference to FIG.
- the detection frame reliability calculation unit 8 calculates the reliability of the detection frame based on the detected inter-frame flow information within the detection target detection frame.
- the detection frame reliability calculation unit 8 analyzes the flow information calculated by the flow calculation unit 7 in the final object detection detection frame information output from the detection frame adjustment determination unit 6, thereby determining the detection frame. Calculate confidence.
- Image feature points 72a and 72b are examples of image feature points with information
- image feature points 72a and 72b are examples of image feature points with flow information on the road surface.
- Optical flow generally calculates the flow information of moving bodies between frames. Therefore, when a camera is fixed like a surveillance camera, image feature points and flow information of a moving object are easily calculated, and image feature points and flow information are hardly calculated for a background such as a road surface.
- image feature points and flow information can be calculated in the same way for a moving object. tend to be small.
- an image feature point with flow information larger than a predetermined threshold is set as a moving body feature point (three-dimensional object feature point), and an image feature point with flow information smaller than the threshold is set as a plane feature point (road surface feature point).
- the reliability of the detection frame 70 is calculated from the content rate of the moving body feature points and the plane feature points within the detection frame 70 . That is, the detection frame reliability calculation unit 8 calculates the reliability of the detection frame based on the information of the three-dimensional object feature points and the road surface feature points within the detection frame. As a calculation method, as shown in (Formula 4) in FIG.
- the reliability may be calculated by considering the direction of the flow information in (Formula 4). For example, using the fact that image feature points of the same moving object have similar flow directions, the more three-dimensional object feature points that indicate similar flow directions exist within the detection frame, the higher the reliability of the detection frame. There are other ways to raise it.
- the flow information is used to calculate the reliability of the detection frame, but the reliability may be calculated using image information that can distinguish between the moving object (three-dimensional object) and the background. For example, after obtaining the inter-frame difference, the more difference areas are found within the detection frame, the higher the possibility that it is the area of a moving object. method, edge detection, etc., are performed within the detection frame to estimate the contour candidate to be detected, and a detection frame with a high overlap rate of the circumscribing rectangle of the contour and the detection frame is judged to be a highly reliable detection frame. method can be used.
- FIG. 10 shows a block configuration diagram of the detection frame position correction unit 9.
- a detection frame position correcting unit 9 uses detection frame information of a high-reliability detection frame having a reliability higher than a threshold to correct detection frame information of a low-reliability detection frame having a reliability lower than that of the high-reliability detection frame. Perform correction processing.
- the detection frame position correction unit 9 includes a detection target association unit 80 that associates detection targets that are considered to be the same from the detection results of all frames, and a high-reliability detection frame selection that selects a highly reliable detection frame for each identical detection target. and a low-reliability detection frame correction unit 82 that corrects the position information of the remaining low-reliability detection frames in the same detection target using the selected high-reliability detection frame information.
- a detection target association unit 80 that associates detection targets that are considered to be the same from the detection results of all frames
- a high-reliability detection frame selection that selects a highly reliable detection frame for each identical detection target.
- the detection target association unit 80 analyzes the detection frame information of all frames and gives the same ID information to the detection frames considered to be the same target.
- the Euclidean distance of the central coordinates between all the detection frames is calculated in the preceding and succeeding frames, and the closest detection frames are judged to be the same detection object.
- the high-reliability detection frame selection unit 81 collects information on detection frames to which the same ID information is assigned by the detection target associating unit 80, and selects a plurality of pieces of information on detection frames with high reliability (high-reliability detection frames). Select.
- the selection method is not particularly limited, and may be, for example, a method of selecting all detection frames whose reliability is equal to or higher than a predetermined threshold, or a method of selecting information on a specified number of detection frames in descending order of reliability.
- the information of the detection frame with high reliability may be selected by utilizing the image feature amount, GUI, or the like.
- the information of the detection frame of the corresponding image is not used, or the user manually selects a high-reliability detection frame using a GUI or the like. You can use a method such as
- the low-reliability detection frame correction unit 82 uses the detection frame information selected by the high-reliability detection frame selection unit 81 to correct the position information of unselected detection frames assigned the same ID information.
- the positional information of the detection frame is information on the position and size of the detection frame in the captured image as indicated by 15 in FIG.
- a correction method information on the center coordinate position, width, and height of the detection frame with high reliability is collected, and general interpolation methods such as spline interpolation are used to correct the center coordinate position, width, and height of the detection frame with low reliability.
- the method is capable of function approximation from a plurality of points, such as a method of calculating each vertical width.
- a method of using the reliability value as weight information in the interpolation method may be used.
- the trajectory generation unit 10 calculates the distance between the camera 2 and the detection target by using the detection frame information output from the detection frame position correction unit 9, and generates a trajectory.
- a general method utilizing camera parameters is used as a method for calculating the distance to the detection target. Specifically, among camera parameters, internal parameters including a focal length and a distortion correction coefficient are used to convert the image coordinates of the center of the lower end of the detection frame in the detection target into camera coordinates.
- the position of the detection target can be estimated, and the distance from the camera to the detection target can be calculated.
- the distance calculation method is not particularly limited as long as it can estimate the distance from the camera to the detection target based on the detection frame information in the image. It is possible to obtain a trajectory by connecting the three-dimensional world coordinates obtained by calculating the distance from the camera to the target using all the detection frame information of the same target and looking down.
- the detection frame is adjusted based on the presence probability map information generated from the object detection result of each frame.
- the object tracking device 1 of this embodiment detects a detection target for each of a plurality of frames acquired by the camera 2, and determines the reliability of the detection frame based on the flow information between the frames within the detection frame of the detected detection target. Calculate degrees. Then, the detection frame information of the high-reliability detection frame, which has a higher reliability than the threshold, is used to correct the detection frame information of the low-reliability detection frame, which has a lower reliability, and a trajectory is generated using the corrected detection frame information. do.
- the object tracking device 1 of the present embodiment it is possible to reduce detection omissions and erroneous detections of the detection target in the frame, reduce positional deviation of the detection frame, and generate a highly accurate trajectory of the same detection target. can be done. Therefore, for example, in order to analyze the ADAS quality and the situation when a traffic accident occurs, the trajectory of the preceding vehicle can be estimated from the in-vehicle image and compared with the CAN data. It is possible to provide services that are utilized for evaluation and the like.
- the object tracking device 1 includes the existence probability map determination unit 5 and the detection frame adjustment determination unit 6, but these may be omitted. Even if the existence probability map determining unit 5 and the detection frame adjustment determining unit 6 are omitted, the detection frame The detection frame information of the detection target can be corrected based on the reliability, and a highly accurate trajectory can be generated using the corrected detection frame information. In the present embodiment, the object tracking device 1 has the existence probability map determining unit 5 and the detection frame adjustment determining unit 6, so that the detection accuracy of the detection frame can be further improved.
- FIG. 11 is a block configuration diagram of Embodiment 2 of the present invention.
- the object tracking device 90 has a detection target moving direction prediction unit (object motion estimation unit) 91 that estimates the motion of the object.
- the object tracking device 90 shown in FIG. 11 predicts the moving direction of the detection target in advance, and utilizes the prediction result to generate an existence probability map, calculate the reliability of the detection frame, and correct the position. It is a device that generates a highly accurate trajectory of an object.
- the trajectory generator 10 have the same or substantially the same functions as those of the first embodiment.
- the detection target movement direction prediction unit 91 generates trajectory information generated by the trajectory generation unit 10, detection target movement area candidate information, object motion information around the detection target, and a medium such as a vehicle in which the camera 2, which is a measurement device, is installed.
- the motion of the object is estimated from one or more pieces of information.
- a detection target moving direction prediction unit 91 shown in FIG. It has a function to predict. A method of generating an existence probability map, calculating a detection frame reliability, and correcting a detection frame position based on the detection target movement direction prediction unit 91 and the acquired surrounding environment information will be described below.
- FIG. 12 is a block diagram of the detection target moving direction prediction unit 91.
- a moving region extracting unit 95 has a function of analyzing frame information acquired from the frame collecting unit 3 and extracting a moving region to be detected.
- a function to acquire control data such as the steering angle and running speed of the corresponding vehicle, and a detection target moving direction output unit 98 analyzes the information acquired by the moving area extraction unit 95, the surrounding object information analysis unit 96, and the control data acquisition unit 97. By doing so, the moving direction of the detection target is estimated and the result is output to the subsequent processing block.
- Functions 95, 96 and 98 are described below.
- the movement area extraction unit 95 performs image processing on the frame, detects information on passages in buildings, information on outdoor sidewalks, information on white lines, etc., and extracts movement areas to be detected.
- an extraction method there is a method of detecting an obstacle, a white line, etc. by object recognition, straight line detection, or the like, and extracting an area surrounded by the obstacle or the white line as a moving area to be detected.
- the surrounding object information analysis unit 96 analyzes the object detection results, analyzes the detection results of objects existing in the surroundings other than the detection target in all frames, and acquires information such as the movement state and movement direction of the object.
- the flow described in the first embodiment is applied to objects other than the object to be detected, and by generating the trajectory of the object, information such as whether the object is stationary or moving can be obtained. If there is, there is a method such as acquiring information on the direction of movement.
- a method of outputting the motion states of selected objects may be adopted, such as a method of using only the motion states of objects with many highly reliable detection frames.
- the detection target movement direction output unit 98 estimates and outputs the detection target movement direction using the acquired detection target movement region, the operating state of the surrounding object, and the control data.
- a method of estimating the direction of movement for example, a method of approximating a moving region of a detection target to a rectangular shape and assuming that a moving object moves back and forth in the long side direction is used to determine the direction of movement of the detection target in the two directions. Analyzing the object's operating state and control data, assuming that the object to be detected is moving in the same way as the surrounding objects, and if the surrounding objects are stationary, it is determined that the object to be detected is stationary. There is a method of setting the moving direction to 0 and determining the moving direction as the moving direction of the detection target when the object is moving.
- a method for correcting information such as the existence probability map, detection frame reliability, and detection frame position from the movement direction of the detection object output by the detection object movement direction prediction unit 91 will be described.
- a method of correcting the existence probability map when determining the position of the existence probability map of the relevant frame using the existence probability maps of the preceding and following frames in the existence probability map interpolation unit 22, Therefore, a method of finely correcting the position of the map is conceivable.
- the reliability of the detection frame the movement direction of the center coordinates of the detection frame of the detection target between frames is obtained, and if it is similar to the predicted movement direction of the detection target, the reliability of the detection frame of the corresponding frame is determined.
- a correction method is conceivable to increase the degree.
- the method of correcting the position of the detection frame when the high-reliability detection frame selection unit 81 selects a highly-reliable detection frame, A method of preferentially selecting frames having similar directions can be considered. In addition to the method described above, there is no particular limitation as long as it is a method that can be used for correction by comparing the predicted movement direction of the detection target with the movement direction of the detection target used in the flow shown in the first embodiment. .
- the direction of movement of a detection target is predicted in advance, and the information on the direction of movement is utilized.
- the presence probability map information generated from the object detection results of each frame is used to adjust the amount of the detection frame, and the position of the detection frame is corrected according to the reliability calculated from the flow information. can be output.
- FIG. 13 is a block configuration diagram of Embodiment 3 of the present invention.
- the object tracking apparatus 100 shown in FIG. 13 adjusts the amount of the detection frame based on the existence probability map information generated from the object detection result of each frame, and corrects the detection frame position according to the reliability calculated from the flow information.
- This is a device that outputs a highly accurate trajectory of a detection target by generating a plurality of trajectories of the detection target and selecting a highly accurate trajectory from among them.
- a frame collection unit 3 an object detection unit 4, an existence probability map determination unit 5, a detection frame adjustment determination unit 6, a flow calculation unit 7, a detection frame reliability calculation unit 8, a detection frame position correction unit 9, and a trajectory generation unit.
- the part 10 has the same function as the first embodiment.
- An existence probability map determination unit 5 generates a plurality of existence probability maps of a detection target, a detection frame reliability calculation unit 8 calculates a plurality of reliability levels of the detection target, and a detection frame position correction unit 9 calculates the detection target. Generate multiple detection frame positions for
- the trajectory generation unit 10 generates a plurality of trajectories using a plurality of existence probability maps, a plurality of detection frame reliability levels, and a plurality of detection frame position information.
- the trajectory storage unit 101 has a function of storing the generated trajectories
- the trajectory selection unit 102 has a function of manually selecting any one trajectory from the stored trajectories using a GUI or the like.
- the trajectory generation unit 10 prepares multiple patterns of detection target existence probability maps, detection frame reliability, and detection frame position correction methods in the same frame, and generates multiple trajectories of the same target. For example, in the presence probability map determination unit 5, a method of generating or interpolating the presence probability map with a different algorithm is used, or the pattern of the regions selected by the presence probability map selection unit 23 is changed. It can generate an existence probability map. Further, the detection frame reliability calculation unit 8 can have a plurality of reliability levels by changing the reliability calculation algorithm. A plurality of pieces of detection window information can be generated by changing the attachment method pattern and the high-reliability detection window selection pattern.
- trajectory selection unit 102 a set of images of all frames to which a detection frame after position correction has been added and a trajectory generated from the detection frame are displayed on a GUI screen for each trajectory, so that the user can choose the trajectory that it considers optimal.
- Object tracking device 1... Object tracking device, 2... Camera (sensor), 3... Frame collection unit, 4... Object detection unit, 5... Existence probability map determination unit, 6... Detection frame adjustment Determination unit 7 Flow calculation unit 8 Detection frame reliability calculation unit 9 Detection frame position correction unit 10 Trajectory generation unit 11 Captured image 12, 13 ... detection target, 14 ... detection frame, 15 ... detection frame information
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Image Analysis (AREA)
Abstract
Description
センサの計測範囲内の物体の軌跡を生成する物体追跡装置であって、
前記センサにより取得した複数のフレームに対してそれぞれ物体の検出を行う物体検出部と、
該物体が検知された検知枠のフレーム間のフロー情報に基づいて前記検知枠の信頼度を算出する検知枠信頼度算出部と、
閾値よりも信頼度の高い高信頼度検知枠の検知枠情報を用いて、前記高信頼度検知枠よりも信頼度の低い低信頼度検知枠の検知枠情報を補正する検知枠位置補正部と、
該補正後の検知枠情報を用いて前記物体の軌跡を生成する軌跡生成部と、
を有することを特徴とする。
図1は本発明の実施例1のブロック構成図である。
本実施例では、計測装置をカメラとした場合の実施形態について説明するが、計測装置はこれに限定されるものではなく、ステレオカメラや距離センサなどの他のセンサに応用可能である。図1に示す物体追跡装置1は、カメラ2で撮影した全フレームの物体検出結果から生成した存在確率マップ情報と、フロー情報などから算出した検知枠の信頼度情報から検知枠位置を補正し、高精度な軌跡を生成する装置である。
物体検出部4は、フレーム収集部3により収集した複数のフレームに対してそれぞれ検出対象の検出を行う。図2において、11は撮像画像(フレーム)、12、13は検出対象(本例では人物と車)、14a、14bは物体検出部4によって画像中から対象物を含む領域を推定した検知枠の一例、15は検知枠情報の一例を示す。物体検出部4では、予め学習データから機械学習などにより生成した画像中から対象物を検出可能な事象を活用することで、撮像画像中に存在する検出対象を検出する。検出用の辞書を生成する際のアルゴリズムについては、畳み込みニューラルネットワークやAdaBoostなど一般的なもので良く、特に限定しない。物体検出部4では、前記辞書によりフレーム収集部3によって取得した複数のフレームに対して対象物を検出し、図2の15に示すような検知枠情報を保存する。検知枠情報としては、検出した対象ごとに、対象の種類を表すクラス情報、クラス情報の正確さを示す信頼度情報、検知枠14a、14bの左上の画像座標を示す始点(X、Y)、検知枠の横幅と縦幅の情報などがある。なお、画像中における検出対象の位置を推定できる手段であれば本例以外の方法を使用しても良い。
存在確率マップ決定部5は、複数のフレームにおいてそれぞれ検出対象の存在確率を示す存在確率マップを生成する処理を行う。存在確率マップ決定部5は、物体検出部4から出力される検知枠情報から各フレームにおける存在確率マップを作成する存在確率マップ作成部21と、対象フレームの前後2枚以上のフレームが有する存在確率マップの情報を補間して、対象フレームの補間存在確率マップを生成する存在確率マップ補間部22と、存在確率マップと補間存在確率マップを比較して最終的な存在確率マップをフレームごとに選定する存在確率マップ選定部23を備える。以下、21、22、23について詳細に説明する。
存在確率マップ作成部21は、複数のフレームにおいてそれぞれ物体の存在確率を示す存在確率マップを生成する。存在確率マップ作成部21では、検知枠情報から検出対象のクラスごとの存在確率マップをフレームごとに算出する。
図5において、11は撮像画像、41は検知枠、42は検知枠の中心座標(xc、yc )、43、44は存在確率マップを計算するために使用する画像座標軸のX方向、Y方向ごとの確率密度関数f(x)、f(y)を示している。X方向の確率密度関数f(x)の作成方法としては、図14の数式1に示す平均値を画像中心のx座標xc、分散σ^2とした正規分布によって算出する。そして、同様にY方向の確率密度関数f(y)も図15の数式2によって算出し、図16の数式3により、f(x)とf(y)の積を求めることで存在確率マップf(x、y)を生成できる。なお、分散σ^2の値としては検知枠の横幅や縦幅の情報を元に決定する方法や、検出対象の実測情報が予め把握できている場合は横幅と縦幅のアスペクト比の情報などを元に決定する方法などを使用しても良い。また、検知枠の信頼度情報を存在確率マップf(x、y)に掛け合わせることで、検知対象物の存在確率の値を調整するなどの方法も採用しても良い。
存在確率マップ補間部22は、フレームtの前後のフレームt-1、t+1の存在確率マップ情報、または、複数の前フレームの存在確率マップ情報、または、複数の後フレームt+1、t+2の存在確率マップ情報から、該当フレーム(対象フレーム)の存在確率マップを推定する処理を行う。存在確率マップ補間部22による補間方法は、大きく分けて図6の51と52に示す2パターンの方法がある。
図8は、フロー算出部7によりt枚目のフレーム11aとt+1枚目のフレーム11bの画像特徴量を解析することで、フレーム間のフロー情報を算出する例を示している。図8において、61は路面、62はフレーム11aにおける計測対象の車両、63はフレーム11bにおける計測対象の車両、64a、64bは車両62における画像特徴点の一例、65a、65bは車両63における画像特徴点の一例、66は画像特徴点64aと65a間のフロー情報、67は画像特徴点64bと65b間のフロー情報を示している。フロー情報とはフレーム間における同一画像特徴点の移動ベクトルであり、ベクトルの向きと大きさを示す。画像特徴点とフロー情報を取得する方法としては、Lucas-Kanade法によるオプティカルフローを活用する方法や、SHIFTなどの画像特徴量を用いた追跡により求める方法など、前後フレーム間における同一の画像特徴点とそのフロー情報を算出可能な方法であれば特に限定しない。本例では先頭フレーム以外のフロー情報は該当フレーム(t枚目)と前フレーム(t-1枚目)から算出し、先頭フレームのフロー情報は該当フレームと次フレーム(t+1枚目)から算出した数値の符号を反転することで取得する。
検知枠信頼度算出部8は、検出された検出対象の検知枠内のフレーム間のフロー情報に基づいて検知枠の信頼度を算出する。検知枠信頼度算出部8では、検知枠加減決定部6から出力された最終的な物体検出の検知枠情報内における、フロー算出部7により算出されたフロー情報を解析することで、検知枠の信頼度を算出する。
検知枠位置補正部9は、閾値よりも信頼度の高い高信頼度検知枠の検知枠情報を用いて、前記高信頼度検知枠よりも信頼度の低い低信頼度検知枠の検知枠情報を補正する処理を行う。検知枠位置補正部9は、全フレームの検出結果から同一と考えられる検出対象を対応付ける検出対象対応付け部80と、同一検出対象ごとに高信頼度の検知枠を選定する高信頼度検知枠選定部81と、選定した高信頼度の検知枠情報を用いて、同一検出対象における残りの低信頼度の検知枠の位置情報を補正する低信頼度検知枠補正部82を備える。以下は各機能について説明する。
図11は本発明の実施例2のブロック構成図である。
本実施例において特徴的なことは、物体追跡装置90が、物体の動作を推定する検出対象移動方向予測部(物体動作推定部)91を有することである。図11に示す物体追跡装置90は、検出対象の移動方向を事前に予測し、その予測結果を活用して存在確率マップの生成や検知枠の信頼度算出や位置補正を実行することで、検出対象の高精度な軌跡を生成する装置である。図11において、フレーム収集部3、物体検出部4、存在確率マップ決定部5’、検知枠加減決定部6、フロー算出部7、検知枠信頼度算出部8’、検知枠位置補正部9’、および軌跡生成部10は、実施例1と同じまたはほぼ同等の機能である。
図12において、移動領域抽出部95は、フレーム収集部3から取得したフレーム情報を解析して検出対象の移動領域を抽出する機能、周辺物体情報解析部96は物体検出部4から取得した物体検出結果を用いて検出対象の周辺の物体の状態などを認識し周辺物体の移動方向などの情報を取得する機能、制御データ取得部97はカメラ(計測装置)2が車両などに設置されている場合に該当車両の操舵角や走行速度などの制御データを取得する機能、検出対象移動方向出力部98は移動領域抽出部95、周辺物体情報解析部96、制御データ取得部97により取得した情報を解析することで、検出対象の移動方向を推定しその結果を後段の処理ブロックに出力する機能である。以降は、機能95、96、98について説明する。
図13は本発明の実施例3のブロック構成図である。
図13に示す物体追跡装置100は、各フレームの物体検出結果から生成した存在確率マップ情報から検知枠の加減を調整し、フロー情報から算出した信頼度に応じて検知枠位置を補正することで検出対象の複数の軌跡を生成し、その中から高精度な軌跡を選択することで、検出対象の高精度な軌跡を出力する装置である。図13において、フレーム収集部3、物体検出部4、存在確率マップ決定部5、検知枠加減決定部6、フロー算出部7、検知枠信頼度算出部8、検知枠位置補正部9、軌跡生成部10は実施例1と同じ機能である。存在確率マップ決定部5は、検出対象の複数の存在確率マップを生成し、検知枠信頼度算出部8は、検出対象の複数の信頼度を算出し、検知枠位置補正部9は、検出対象の複数の検知枠位置を生成する。
Claims (10)
- センサの計測範囲内の物体の軌跡を生成する物体追跡装置であって、
前記センサにより取得した複数のフレームに対してそれぞれ物体の検出を行う物体検出部と、
該検出された前記物体の検知枠内のフレーム間のフロー情報に基づいて前記検知枠の信頼度を算出する検知枠信頼度算出部と、
閾値よりも信頼度の高い高信頼度検知枠の検知枠情報を用いて、前記高信頼度検知枠よりも信頼度の低い低信頼度検知枠の検知枠情報を補正する検知枠位置補正部と、
該補正後の検知枠情報を用いて前記軌跡を生成する軌跡生成部と、
を有することを特徴とする物体追跡装置。 - 前記検知枠信頼度算出部は、前記検知枠内の立体物特徴点と路面特徴点の情報に基づいて前記検知枠の信頼度を算出することを特徴とする請求項1に記載の物体追跡装置。
- 前記検知枠位置補正部は、複数の前記高信頼度検知枠の位置と縦幅と横幅の情報を用いて、前記低信頼度検知枠の位置と縦幅と横幅の情報を補正することを特徴とする請求項1又は2に記載の物体追跡装置。
- 前記複数のフレームにおいてそれぞれ前記物体の存在確率を示す存在確率マップを生成する存在確率マップ決定部と、
前記存在確率マップを用いて前記複数のフレームにおいてそれぞれ前記検知枠の追加または削除を決定する検知枠加減決定部と、
を備えることを特徴とする請求項1に記載の物体追跡装置。 - 前記存在確率マップ決定部は、
前記物体検出部により出力される検知枠情報から各フレームにおける存在確率マップを作成する存在確率マップ作成部と、
前記複数のフレームに含まれる対象フレームの前後2枚以上のフレームが有する存在確率マップの情報を補間して、前記対象フレームの補間存在確率マップを生成する存在確率マップ補間部と、
前記存在確率マップと前記補間存在確率マップを比較して最終の存在確率マップを選定する存在確率マップ選定部と、
を備えることを特徴とする請求項4に記載の物体追跡装置。 - 前記存在確率マップ作成部は、前記物体検出部により検知された物体の検知枠の位置情報と前記信頼度情報から正規分布によって算出した確率密度関数、および、複数の異なる辞書による物体検出結果のいずれか1つないし複数の情報を用いて、前記存在確率マップを作成することを特徴とする請求項5に記載の物体追跡装置。
- 前記物体の動作を推定する物体動作推定部を備えることを特徴とする請求項6に記載の物体追跡装置。
- 前記物体動作推定部は、前記軌跡生成部によって生成された軌跡情報、物体の移動領域候補情報、物体周辺の物体動作情報、および前記センサを設置した媒体情報のいずれか1つないし複数の情報から前記物体の動作を推定することを特徴とする請求項7に記載の物体追跡装置。
- 前記物体動作推定部により求めた前記物体の動作情報を用いて、前記存在確率マップ、前記検知枠の信頼度、および前記検知枠の位置を再度補正することを特徴とする請求項8に記載の物体追跡装置。
- 前記存在確率マップ決定部は、前記物体の複数の存在確率マップを生成し、
前記検知枠信頼度算出部は、前記物体の複数の信頼度を算出し、
前記検知枠位置補正部は、前記物体の複数の検知枠位置を補正し、
前記軌跡生成部は、前記複数の存在確率マップ、前記複数の信頼度、および前記複数の検知枠情報を用いて複数の軌跡を生成し、いずれか一つの軌跡を選定することを特徴とする請求項4に記載の物体追跡装置。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US18/549,399 US12651353B2 (en) | 2021-04-05 | 2022-03-07 | Object tracking device |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2021064048A JP7675339B2 (ja) | 2021-04-05 | 2021-04-05 | 物体追跡装置 |
| JP2021-064048 | 2021-04-05 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2022215409A1 true WO2022215409A1 (ja) | 2022-10-13 |
Family
ID=83546358
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/JP2022/009748 Ceased WO2022215409A1 (ja) | 2021-04-05 | 2022-03-07 | 物体追跡装置 |
Country Status (3)
| Country | Link |
|---|---|
| US (1) | US12651353B2 (ja) |
| JP (1) | JP7675339B2 (ja) |
| WO (1) | WO2022215409A1 (ja) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115877328A (zh) * | 2023-03-06 | 2023-03-31 | 成都鹰谷米特科技有限公司 | 一种阵列雷达的信号收发方法及阵列雷达 |
| US20240386724A1 (en) * | 2021-10-01 | 2024-11-21 | Sony Semiconductor Solutions Corporation | Recognition processing device, recognition processing method, and recognition processing system |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023017723A1 (ja) * | 2021-08-13 | 2023-02-16 | キヤノン株式会社 | 情報処理装置、情報処理システム、情報処理方法、及びプログラム |
| PE20242036A1 (es) * | 2022-03-03 | 2024-10-10 | Wi Tronix Llc | Metodo y sistema de deteccion de amenazas operativas |
| JP7826952B2 (ja) * | 2023-01-11 | 2026-03-10 | トヨタ自動車株式会社 | 移動物体探索装置 |
| JP7778105B2 (ja) * | 2023-03-28 | 2025-12-01 | 三菱電機株式会社 | 物体検出装置及び物体検出システム |
| JP7790403B2 (ja) * | 2023-05-19 | 2025-12-23 | トヨタ自動車株式会社 | 再同定システム |
| TW202526831A (zh) * | 2023-12-29 | 2025-07-01 | 財團法人工業技術研究院 | 在物件偵測模型中更新定界框或關鍵點的方法 |
| CN118279576B (zh) * | 2024-06-04 | 2024-09-06 | 浙江大华技术股份有限公司 | 一种目标检测方法、装置以及计算机存储介质 |
| CN119091471B (zh) * | 2024-11-08 | 2025-02-25 | 成都考拉悠然科技有限公司 | 用于铁塔场景的飞鸟入侵检测方法 |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2013092955A (ja) * | 2011-10-27 | 2013-05-16 | Hitachi Ltd | 映像解析装置及びシステム |
| WO2017094140A1 (ja) * | 2015-12-02 | 2017-06-08 | 三菱電機株式会社 | 物体検出装置及び物体検出方法 |
| JP2021022315A (ja) * | 2019-07-30 | 2021-02-18 | キヤノン株式会社 | 画像処理装置、画像処理方法、及びプログラム |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2006323437A (ja) | 2005-05-17 | 2006-11-30 | Sony Corp | 物体追跡方法、物体追跡方法のプログラム、物体追跡方法のプログラムを記録した記録媒体及び物体追跡装置 |
| US8971575B2 (en) * | 2012-11-23 | 2015-03-03 | Cyberlink Corp. | Systems and methods for tracking objects |
| CN107301651A (zh) * | 2016-04-13 | 2017-10-27 | 索尼公司 | 对象跟踪装置和方法 |
| US10628961B2 (en) * | 2017-10-13 | 2020-04-21 | Qualcomm Incorporated | Object tracking for neural network systems |
| CN110163033B (zh) * | 2018-02-13 | 2022-04-22 | 京东方科技集团股份有限公司 | 正样本获取方法、行人检测模型生成方法和行人检测方法 |
| US11494922B2 (en) | 2018-03-23 | 2022-11-08 | Nec Corporation | Object tracking device, object tracking method, and object tracking program |
| CN110472594B (zh) * | 2019-08-20 | 2022-12-06 | 腾讯科技(深圳)有限公司 | 目标跟踪方法、信息插入方法及设备 |
| JP6891984B1 (ja) * | 2020-01-22 | 2021-06-18 | 沖電気工業株式会社 | 物体検出装置、物体検出方法、プログラム及び記録媒体 |
| CN111553282B (zh) * | 2020-04-29 | 2024-03-29 | 北京百度网讯科技有限公司 | 用于检测车辆的方法和装置 |
-
2021
- 2021-04-05 JP JP2021064048A patent/JP7675339B2/ja active Active
-
2022
- 2022-03-07 US US18/549,399 patent/US12651353B2/en active Active
- 2022-03-07 WO PCT/JP2022/009748 patent/WO2022215409A1/ja not_active Ceased
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2013092955A (ja) * | 2011-10-27 | 2013-05-16 | Hitachi Ltd | 映像解析装置及びシステム |
| WO2017094140A1 (ja) * | 2015-12-02 | 2017-06-08 | 三菱電機株式会社 | 物体検出装置及び物体検出方法 |
| JP2021022315A (ja) * | 2019-07-30 | 2021-02-18 | キヤノン株式会社 | 画像処理装置、画像処理方法、及びプログラム |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20240386724A1 (en) * | 2021-10-01 | 2024-11-21 | Sony Semiconductor Solutions Corporation | Recognition processing device, recognition processing method, and recognition processing system |
| US12633133B2 (en) * | 2021-10-01 | 2026-05-19 | Sony Semiconductor Solutions Corporation | Recognition processing device, recognition processing method, and recognition processing system |
| CN115877328A (zh) * | 2023-03-06 | 2023-03-31 | 成都鹰谷米特科技有限公司 | 一种阵列雷达的信号收发方法及阵列雷达 |
Also Published As
| Publication number | Publication date |
|---|---|
| US20240153106A1 (en) | 2024-05-09 |
| JP2022159694A (ja) | 2022-10-18 |
| US12651353B2 (en) | 2026-06-09 |
| JP7675339B2 (ja) | 2025-05-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP2022159694A (ja) | 物体追跡装置 | |
| JP7428213B2 (ja) | 情報処理システム、情報処理方法及びプログラム | |
| US8447069B2 (en) | Apparatus and method for moving object detection | |
| CN113989756B (zh) | 使用稳定的坐标系的道路垂直轮廓检测 | |
| KR102099265B1 (ko) | 보행자 상태 기반의 보행자와 차량간 충돌 경고 시스템 및 그 방법 | |
| US10452931B2 (en) | Processing method for distinguishing a three dimensional object from a two dimensional object using a vehicular system | |
| KR102463720B1 (ko) | 차량의 경로 생성 시스템 및 방법 | |
| EP2858008B1 (en) | Target detecting method and system | |
| KR102153607B1 (ko) | 영상에서의 전경 검출 장치 및 방법 | |
| Held et al. | Precision tracking with sparse 3d and dense color 2d data | |
| JP5136504B2 (ja) | 物体識別装置 | |
| CN112947419A (zh) | 避障方法、装置及设备 | |
| KR20150032789A (ko) | 어떤 대상의 자기 동작 추정을 위한 방법 | |
| JPWO2012014430A1 (ja) | 移動体検出装置および移動体検出方法 | |
| CN102194239A (zh) | 用于处理图像数据的方法和系统 | |
| KR102295183B1 (ko) | Cctv 프로젝션 모델을 이용한 cctv 영상의 객체 추적 방법 | |
| US12576883B2 (en) | System and method for embedding uncertainty estimation into deep-neural-network-based autonomous driving perception frameworks | |
| EP3635684B1 (en) | Pyramidal optical flow tracker improvement | |
| KR20110023468A (ko) | 적응적 배경 기반의 객체 검출 및 추적 장치 및 방법 | |
| CN116348937A (zh) | 物体识别装置 | |
| KR20210142518A (ko) | 동적 물체 탐지 방법 및 장치 | |
| CN113168703A (zh) | 用于借助数字图像序列来确定相对运动的方法 | |
| CN113936039A (zh) | 一种对象跟踪方法、装置及系统 | |
| KR20060121503A (ko) | 무인 감시 로봇에서 중요 얼굴 추적 장치 및 방법 | |
| CN115222771B (zh) | 一种目标跟踪方法及装置 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 22784394 Country of ref document: EP Kind code of ref document: A1 |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 18549399 Country of ref document: US |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 22784394 Country of ref document: EP Kind code of ref document: A1 |
|
| WWG | Wipo information: grant in national office |
Ref document number: 18549399 Country of ref document: US |