Cross-camera target tracking method and device, medium and equipment

Abstract

The invention relates to the technical field of image processing, in particular to a cross-camera target tracking method and device, a medium and equipment. The method comprises the following steps: constructing a virtual polygon prism according to at least three non-coplanar cameras; wherein the image visual angle of each non-coplanar camera corresponds to one side surface of the polygon prism; within a preset video frame range, performing target tracking on each target object within a visual angle range of each non-coplanar camera in the at least three non-coplanar cameras; and in each video frame in the preset video frame range, performing cross-camera target tracking on the target object according to the virtual polygon prism. According to the invention, while the calculation complexity is reduced, the accuracy and scene adaptability of cross-camera target tracking are improved, and dispersed visual information is converted into a quantifiable spatial relationship through a structured geometric model.

Description

Technical Field

[0001] This application relates to the fields of image processing technology and artificial intelligence, and in particular to a method, apparatus, medium and device for target tracking across cameras. Background Technology

[0002] Currently, many excellent algorithms have emerged in real-time multi-target tracking solutions across cameras based on pure vision. Multi-target tracking, as the name suggests, involves tracking multiple targets in a video frame and obtaining their motion trajectories. The core lies in target detection and data association. Target detection is performed in each frame, and the results are then used for target tracking. The latter step is generally called data association, which relies heavily on manual feature extraction (appearance features, motion features, shape features). Kalman filtering and Hungarian matching backend tracking optimization algorithms are representative multi-target tracking algorithms. However, they are almost ineffective for multi-camera tracking scenarios, as Kalman filtering and REID models cannot associate or maintain association for extended periods with the same target appearing in different cameras. Summary of the Invention

[0003] The technical problem this application aims to solve is: how to stably correlate the same target appearing under different cameras in order to achieve target tracking across cameras.

[0004] To address the aforementioned technical problems, according to a first aspect of this application, a cross-camera target tracking method is provided, the method comprising: A virtual polygonal prism is constructed based on at least three non-coplanar cameras; wherein, the image viewpoint of each non-coplanar camera corresponds to one side of the polygonal prism; the adjacent spatial relationship between the sides of the polygonal prism is the same as the adjacent spatial relationship between the corresponding non-coplanar cameras; the two-dimensional coordinate direction of the image viewpoint of the non-coplanar cameras is consistent with the orthogonal basis vector direction space of the three-dimensional plane where the corresponding side of the polygonal prism is located. Within a preset video frame range, target tracking is performed on each target object within the field of view of each of at least three non-coplanar cameras; wherein each target object has a unique identifier within each video frame of its respective non-coplanar camera; each non-coplanar camera has a corresponding target object list; the target object list includes each unique identifier corresponding to each video frame within the preset video frame range; and, In each video frame within a preset video frame range, the target object is tracked across cameras based on the virtual prism; wherein, the cross-camera target tracking of the target object is determined based on the Euclidean distance between the target objects corresponding to the non-coplanar cameras on each two adjacent sides of the virtual prism within the corresponding video frame image.

[0005] In one exemplary embodiment of this application, cross-camera target tracking based on a virtual prism includes: Get the list of target objects corresponding to each non-coplanar camera in the current video frame; Obtain the Euclidean distance between each target object in the target object list of each non-coplanar camera in the current video frame and each target object in the target object list of each adjacent non-coplanar camera; wherein, the Euclidean distance is calculated based on the coordinate position of the target box of each target object in the virtual prism side in the current video frame; If the Euclidean distance between any two target objects in any two adjacent non-coplanar cameras is less than a preset distance threshold, then the matching degree between the two target objects is obtained. If the matching degree between the two target objects is greater than the preset matching degree threshold, then the unique identifier of the non-critical target object is replaced with the unique identifier of the critical target object; wherein, the number of video frames in which the critical target object appears in the target object list of the corresponding non-coplanar camera is greater than the number of video frames in which the non-critical target object appears in the target object list of the corresponding non-coplanar camera.

[0006] In one exemplary embodiment of this application, obtaining the matching degree between the two target objects includes: Extract the object features of the bounding box of each of the two target objects based on the re-identification model; Obtain the matching degree between the object features of the two target objects.

[0007] In one exemplary embodiment of this application, after obtaining the Euclidean distance between each target object in the target object list of each non-coplanar camera in the current video frame and each target object in the target object list of each adjacent non-coplanar camera, the method further includes: If the Euclidean distance between any two target objects in any two adjacent non-coplanar cameras is equal to or greater than a preset distance threshold, then it is determined that there is currently no same target object appearing simultaneously within the shooting range of different non-coplanar cameras.

[0008] In one exemplary embodiment of this application, target tracking is performed on each target object within the field of view of each of at least three non-coplanar cameras within a preset video frame range, including: Obtain the coordinates of the bounding box of each target object and the initial tracking trajectory of each target object in the first frame of each of the several non-coplanar cameras at the start of tracking; wherein each initial tracking trajectory is in an unconfirmed state; The coordinates of the predicted target box corresponding to each initial tracking trajectory in the current frame are predicted based on the Kalman filter. Based on the coordinates of the predicted bounding box corresponding to each initial tracking trajectory in the current frame and the actual bounding box of each target object in the current frame, the IOU cost matrix corresponding to each initial tracking trajectory is obtained. The IOU cost matrix is ​​input into the Hungarian algorithm for matching to obtain the matching results; the matching results include trajectory matching failure, target box matching failure, and trajectory and target box matching success. The state of each initial tracking trajectory is updated according to the matching result corresponding to the current frame. If the matching result is that the trajectory and the target box match successfully, it is determined whether the corresponding initial tracking trajectory is in a confirmed state and the initial tracking trajectory is updated. If the matching result is that the trajectory match fails and the number of consecutive matching failures exceeds a preset failure number threshold, the corresponding initial tracking trajectory is deleted. If the matching result is that the target box match fails, an initial tracking trajectory corresponding to the target box is created and the initial tracking trajectory is set to an unconfirmed state. Then, the process jumps to the step of predicting the predicted target box coordinates of each initial tracking trajectory in the current frame according to the Kalman filter, until the end video frame of the preset video frame range is reached, at which point the target tracking ends.

[0009] In one exemplary embodiment of this application, an IOU cost matrix is ​​obtained based on the coordinates of the predicted bounding box corresponding to each initial tracking trajectory in the current frame and the actual bounding box of each target object in the current frame, including: If any initial tracking trajectory is confirmed as a confirmed trajectory in the previous frame of the current frame and the current frame matches successfully, then the historical actual features corresponding to the initial tracking trajectory in the previous preset number of historical video frames are obtained. Based on the current actual features corresponding to the actual target box of each target object in the current frame and the historical actual features corresponding to the initial tracking trajectory in a preset number of historical video frames before the current frame, the key matching degree of the initial tracking trajectory in the current frame is obtained; where the key matching degree is the maximum matching degree between all corresponding current actual features and all corresponding historical actual features. Based on the coordinates of the predicted bounding box and the coordinates of each actual bounding box in the current frame corresponding to the initial tracking trajectory, the intersection-over-union ratio (IoU) of the predicted bounding box and each actual bounding box in the current frame is obtained. The IOU cost matrix is ​​obtained based on the key matching degree of the current frame and the intersection-union ratio of the predicted target box and each actual target box in the current frame.

[0010] In one exemplary embodiment of this application, the IOU cost matrix is ​​obtained based on the coordinates of the predicted bounding box corresponding to each initial tracking trajectory in the current frame and the actual bounding box of each target object in the current frame, and further includes: If any initial tracking trajectory was not confirmed in the previous frame of the current frame, or if any initial tracking trajectory was confirmed in the previous frame of the current frame but failed to match the target box in the current frame, or if any initial tracking trajectory failed to match the target box in the current frame, then the intersection-over-union ratio of the predicted target box and each actual target box in the current frame is obtained based on the coordinates of the predicted target box corresponding to the initial tracking trajectory in the current frame and the coordinates of each actual target box. The IOU cost matrix is ​​obtained based on the intersection-union ratio of the predicted target box and each actual target box in the current frame.

[0011] According to a second aspect of this application, a target tracking device across cameras is provided, the device comprising: A building unit is used to construct a virtual polygonal prism based on at least three non-coplanar cameras; wherein, the image viewpoint of each non-coplanar camera corresponds to one side of the polygonal prism; the adjacent spatial relationship between the sides of the polygonal prism is the same as the adjacent spatial relationship between the corresponding non-coplanar cameras; the two-dimensional coordinate direction of the image viewpoint of the non-coplanar camera is consistent with the orthogonal basis vector direction space of the three-dimensional plane where the corresponding side of the polygonal prism is located; The tracking unit is configured to track each target object within the field of view of each of at least three non-coplanar cameras within a preset video frame range; wherein each target object has a unique identifier within each video frame image of its respective non-coplanar camera; each non-coplanar camera has a corresponding target object list; the target object list includes each unique identifier corresponding to each video frame within the preset video frame range; and, The tracking unit is used to perform cross-camera target tracking of the target object in each video frame within a preset video frame range, based on a virtual polygonal prism; wherein, the cross-camera target tracking of the target object is determined based on the Euclidean distance between the target objects corresponding to the non-coplanar cameras of each two adjacent sides of the virtual polygonal prism in the corresponding video frame images.

[0012] According to a third aspect of this application, a non-transitory computer-readable storage medium is provided, wherein at least one instruction or at least one program is stored in the storage medium, and the at least one instruction or at least one program is loaded and executed by a processor to implement the above-described cross-camera target tracking method.

[0013] According to a fourth aspect of this application, an electronic device is provided, including a processor and the aforementioned non-transitory computer-readable storage medium.

[0014] This application has at least the following beneficial effects: The cross-camera target tracking method provided in this application establishes spatial consistency through a virtual prism, reducing the complexity of cross-camera coordinate mapping. The image viewpoints of non-coplanar cameras correspond one-to-one with the sides of the prism, and the adjacency relationships of the sides replicate the spatial adjacency relationships of the cameras. This is equivalent to "binding" physically dispersed cameras into a unified three-dimensional geometric structure. Traditional cross-camera tracking requires handling complex transformations of different camera coordinate systems (such as 3D reconstruction through camera calibration matrices). In this scheme, the orthogonal basis vectors of the prism sides are consistent with the two-dimensional coordinate direction space of the corresponding camera images, directly associating the two-dimensional coordinates of the image with local directions in three-dimensional space, reducing the computational load and errors of coordinate system transformation.

[0015] Within a preset video frame, each camera has achieved target tracking within a single camera using a unique identifier. Adjacent sides of the polygonal prism correspond to spatially adjacent cameras (e.g., front / back, left / right cameras). When a target moves from the field of view of one camera to the field of view of another adjacent camera, its coordinates in the images of the two cameras can be transformed into spatial points on adjacent sides within the polygonal prism structure, and the Euclidean distance between the two points will exhibit continuity. Compared to traditional cross-camera matching based on appearance features (e.g., color, shape) (which is susceptible to lighting and angle effects), this scheme utilizes the continuity of spatial position (the rationality of Euclidean distance) for matching. Furthermore, when adding a new non-coplanar camera, simply adding a new side to the polygonal prism and defining its adjacency with existing sides expands the system's tracking range without redesigning the overall matching logic, thus improving the system's scalability. In summary, this technical solution, through geometric structure reuse (polyprism), spatial relationship mapping, and physical continuity matching, reduces computational complexity while improving the accuracy and scene adaptability of cross-camera target tracking, transforming scattered visual information into quantifiable spatial relationships through a structured geometric model. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a cross-camera target tracking method provided in one embodiment of this application; Figure 2 A schematic diagram of a virtual polygonal prism in a cross-camera target tracking method provided in one embodiment of this application; Figure 3This is a schematic diagram of video frames at the same moment in the actual application of the cross-camera target tracking method provided in one embodiment of this application; Figure 4 A flowchart of a cross-camera target tracking method provided in another embodiment of this application; Figure 5 This is a structural block diagram of a cross-camera target tracking device provided in one embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] like Figure 1 As shown, one embodiment of this application provides a target tracking method across cameras, the method comprising: Step S100: Construct a virtual polygonal prism based on at least three non-coplanar cameras; wherein, the image viewpoint of each non-coplanar camera corresponds to one side of the polygonal prism; the adjacent spatial relationship between the sides of the polygonal prism is the same as the adjacent spatial relationship between the corresponding non-coplanar cameras; the two-dimensional coordinate direction of the image viewpoint of the non-coplanar cameras is consistent with the orthogonal basis vector direction space of the three-dimensional plane where the corresponding side of the polygonal prism is located. The shooting ranges of two non-coplanar cameras with adjacent spatial relationships have overlapping parts, but are not completely overlapping. The two-dimensional coordinate direction of the image viewpoint of the non-coplanar cameras can be understood as the shooting direction / viewpoint direction of the non-coplanar cameras in real space. It can be understood that, in this embodiment, the non-coplanar nature of the non-coplanar cameras actually means that the corresponding sides of the polygonal prism are all different from each other.

[0020] Step S200: Within a preset video frame range, target tracking is performed on each target object within the field of view of each of at least three non-coplanar cameras; wherein each target object has a unique identifier within each video frame image of its respective non-coplanar camera; each non-coplanar camera has a corresponding target object list; the target object list includes each unique identifier corresponding to each video frame within the preset video frame range; and, Step S300: In each video frame within a preset video frame range, cross-camera target tracking is performed on the target object based on the virtual polygonal prism; wherein, the cross-camera target tracking of the target object is determined based on the Euclidean distance between the target objects corresponding to the non-coplanar cameras of each two adjacent sides of the virtual polygonal prism within the corresponding video frame image.

[0021] It is understood that in this embodiment, in the time dimension, each non-coplanar camera performs object tracking independently, and the tracking algorithm used for object tracking can be any object tracking algorithm. That is, in step S200, for multiple video frames corresponding to a certain time period (i.e., a preset video frame range), any object tracking algorithm can be used to track the target object. In this embodiment, the preset video frame range specifically refers to the set of all video frames corresponding to a certain time period, and the set of all video frames corresponding to different non-coplanar cameras is different. That is, the set of all video frames corresponding to each non-coplanar camera is the video frames it captured within the aforementioned time period. In the video frames captured by different non-coplanar cameras at the same time (i.e., in each video frame within the preset video frame range), the target object can be tracked across cameras by means of a virtual prism, that is, to determine whether the target objects appearing in different non-coplanar cameras at the same time are the same person or object. If they are the same person or object, the unique identifiers corresponding to the two will be set to be the same. In this way, cross-camera object tracking can be achieved.

[0022] The cross-camera target tracking method provided in this application establishes spatial consistency through a virtual prism, reducing the complexity of cross-camera coordinate mapping. The image viewpoints of non-coplanar cameras correspond one-to-one with the sides of the prism, and the adjacency relationships of the sides replicate the spatial adjacency relationships of the cameras. This is equivalent to "binding" physically dispersed cameras into a unified three-dimensional geometric structure. Traditional cross-camera tracking requires handling complex transformations of different camera coordinate systems (such as 3D reconstruction through camera calibration matrices). In this scheme, the orthogonal basis vectors of the prism sides are consistent with the two-dimensional coordinate direction space of the corresponding camera images, directly linking the two-dimensional coordinates of the image to local directions in three-dimensional space, reducing the computational load and errors of coordinate system transformation.

[0023] Within a preset video frame, each camera has achieved target tracking within a single camera using a unique identifier. Adjacent sides of the polygonal prism correspond to spatially adjacent cameras (e.g., front / back, left / right cameras). When a target moves from the field of view of one camera to the field of view of another adjacent camera, its coordinates in the images of the two cameras can be transformed into spatial points on adjacent sides within the polygonal prism structure, and the Euclidean distance between the two points will exhibit continuity. Compared to traditional cross-camera matching based on appearance features (e.g., color, shape) (which is susceptible to lighting and angle effects), this scheme utilizes the continuity of spatial position (the rationality of Euclidean distance) for matching. Furthermore, when adding a new non-coplanar camera, simply adding a new side to the polygonal prism and defining its adjacency with existing sides expands the system's tracking range without redesigning the overall matching logic, thus improving the system's scalability. In summary, this technical solution, through geometric structure reuse (polyprism), spatial relationship mapping, and physical continuity matching, reduces computational complexity while improving the accuracy and scene adaptability of cross-camera target tracking, transforming scattered visual information into quantifiable spatial relationships through a structured geometric model.

[0024] Specifically, the number of non-coplanar cameras is the same as the number of sides of the virtual polygonal prism, and they correspond one-to-one; in one exemplary embodiment, the virtual polygonal prism can be constructed by referring to Figure 2 The diagram shows a quadrangular prism ABCD-A'B'C'D'; the origin of the spatial coordinate system of the virtual polygonal prism can be the center point of the base ABCD; cameras 1 to 4 in the diagram capture images of the non-coplanar parts corresponding to each side. It is important to note that the labels for cameras 1 to 4 only indicate a correspondence, meaning that camera 1 corresponds to side A'ADD', not that camera 1 is located at A'.

[0025] In practical applications, a virtual prism is a virtual spatial substitute for a real-world object; for example, it can be used as a virtual representation of a real-world object. Figure 2 In the diagram, the quadrangular prism ABCD-A'B'C'D' is considered as a vehicle in physical space, and cameras 1 to 4 are cameras installed at the front, rear, left, and right of the vehicle. Specifically, for example, camera 2 can be a non-coplanar camera at the front of the vehicle (front end), and camera 3 can be a non-coplanar camera at the rear of the vehicle (rear end).

[0026] In practical applications, this method can be used in the vehicle's sentry mode. In sentry mode, this method detects and tracks targets around the vehicle, determining the overall trajectory of the target object in the surrounding environment through single-camera tracking and cross-camera target association. Based on the overall trajectory, it determines whether the target object intends to perform unauthorized actions (such as damaging the vehicle). If unauthorized intent is detected, a beeping warning is issued, and an alarm message is sent to associated electronic devices (such as the vehicle owner's mobile phone).

[0027] In one exemplary embodiment of this application, cross-camera target tracking based on a virtual prism includes: Step S310: Obtain the target object list corresponding to each non-coplanar camera in the current video frame; wherein, the target object list can be empty, that is, no target object is detected in the current video frame; specifically, the target object can be a pedestrian, a two-wheeled vehicle, a three-wheeled vehicle, a four-wheeled vehicle, etc. Each target has a corresponding bounding box, which can be understood as a detection box; for example, Figure 2 The detection box corresponding to target 1 in side A'ADD'.

[0028] Step S320: Obtain the Euclidean distance between each target object in the target object list of each non-coplanar camera in the current video frame and each target object in the target object list of each adjacent non-coplanar camera; wherein, the Euclidean distance is calculated based on the coordinate position of the target bounding box of each target object in the virtual polygonal prism side in the current video frame; specifically, the adjacency relationship is a vector in space between sides, for example, side A'ADD' is adjacent to side A'ABB', side A'ADD' is adjacent to side D'DCC', but side A'ABB' and side D'DCC' are not adjacent. It should be noted that the current video frame of each non-coplanar camera corresponds to the same time point in step S320. For example... Figure 3 The four sub-images shown are video frames captured by four non-coplanar cameras at the same time point.

[0029] Specifically, when calculating the Euclidean distance between target objects, the method involves mapping the current video frame of a non-coplanar camera onto the corresponding side face of a virtual polygonal prism. For example, [the method involves mapping the current video frame of a non-coplanar camera onto the corresponding side face of a virtual polygonal prism]. Figure 3 The top-left sub-image is mapped to side B'BCC' of the quadrangular prism ABCD-A'B'C'D' (i.e., the left side of the vehicle), the top-right sub-image is mapped to side A'ADD' (i.e., the right side of the vehicle), the bottom-left sub-image is mapped to side A'ABB' (i.e., the front of the vehicle), and the bottom-right sub-image is mapped to side D'DCC' (i.e., the rear of the vehicle). Then, the bounding box corresponding to each target object is obtained (e.g., ...). Figure 2The center point of the detection box of target 1 is the coordinate of the target object's bounding box in the side of the virtual polygonal prism.

[0030] Step S330: If the Euclidean distance between any two target objects in any two adjacent non-coplanar cameras is less than a preset distance threshold, then obtain the matching degree between the two target objects.

[0031] The process of obtaining the matching degree between the two target objects includes: Step S331: Extract the object features of the bounding box of each of the two target objects based on the re-identification model. Specifically, the re-identification model can be a ReID model.

[0032] Step S332: Obtain the matching degree between the object features of the two target objects. The matching degree can be determined based on the cosine distance between the object features. Specifically, the smaller the cosine distance, the greater the matching degree.

[0033] Step S340: If the matching degree between the two target objects is greater than a preset matching degree threshold, then the unique identifier of the non-critical target object is replaced with the unique identifier of the critical target object. Specifically, the number of video frames in which the critical target object appears in the target object list of the corresponding non-coplanar camera is greater than the number of video frames in which the non-critical target object appears in the target object list of the corresponding non-coplanar camera. That is, if the target objects appearing in different non-coplanar cameras at the same time are the same person or object, then their corresponding unique identifiers will be set to the same unique identifier. This enables cross-camera object tracking.

[0034] Step S350: If the Euclidean distance between any two target objects in any two adjacent non-coplanar cameras is equal to or greater than a preset distance threshold, then it is determined that there is currently no same target object appearing simultaneously within the shooting range of different non-coplanar cameras.

[0035] Please refer to Figure 2 and Figure 3 In real-world scenarios, if the image appears to be positioned to the right front of the vehicle, it will simultaneously be captured by the non-coplanar camera at the front of the vehicle (i.e., ...). Figure 2 Camera 2, which corresponds to Figure 3 The image in the lower left corner) and the non-coplanar camera on the right side of the vehicle (i.e. Figure 2 Camera 1, which corresponds to Figure 3 The image in the upper right corner of the image was captured simultaneously. At this moment, the non-coplanar camera in front of the vehicle was able to detect a target object within the current video frame and mark the target bounding box (i.e., Figure 2 (Target 2) Within the current video frame of the non-coplanar camera on the right side of the vehicle, a target object can be detected and its bounding box marked (i.e., ...). Figure 2 Target 1). Reference Figure 2 By mapping the current video frames from the non-coplanar camera in front of the vehicle and the non-coplanar camera on the right side of the vehicle onto a virtual polygonal prism (i.e., a quadrangular prism ABCD-A'B'C'D'), the spatial distance between the detection boxes of target 1 and target 2 within the virtual polygonal prism is small. Therefore, it can be determined that the target objects corresponding to target 1 and target 2 are actually the same. Thus, in this embodiment, the Euclidean distance between any two target objects from adjacent non-coplanar cameras can be used to determine whether the target objects detected by two adjacent non-coplanar cameras at the same time are the same person or object. This enables cross-camera recognition of the same target. Furthermore, since it does not require building a 3D scene based on the current video frame as in other existing technologies, but only maps the current video frame onto a two-dimensional plane in three-dimensional space, the three-dimensional distance between target objects on different two-dimensional planes at different times can be used to determine whether they are the same person or object, greatly reducing the computational load.

[0036] In one exemplary embodiment of this application, target tracking is performed on each target object within the field of view of each of at least three non-coplanar cameras within a preset video frame range, including: Step S210: Obtain the coordinates of the target box of each target object and the initial tracking trajectory of each target object in the first frame of the first tracking frame of each of the several non-coplanar cameras; wherein each initial tracking trajectory is in an unconfirmed state. Step S220: Predict the coordinates of the predicted target box corresponding to each initial tracking trajectory in the current frame based on the Kalman filter.

[0037] Step S230: Based on the coordinates of the predicted target box corresponding to each initial tracking trajectory in the current frame and the actual target box of each target object in the current frame, obtain the IOU cost matrix corresponding to each initial tracking trajectory.

[0038] Step S230 specifically includes: Step S231: If any initial tracking trajectory is confirmed as a confirmed trajectory in the previous frame of the current frame and the current frame matches successfully, then obtain the historical actual features corresponding to a preset number of historical video frames before the current frame for the initial tracking trajectory.

[0039] Step S232: Based on the current actual features corresponding to the actual target box of each target object in the current frame and the historical actual features corresponding to the initial tracking trajectory in a preset number of historical video frames before the current frame, obtain the key matching degree of the initial tracking trajectory in the current frame; wherein, the key matching degree is the maximum matching degree between all corresponding current actual features and all corresponding historical actual features.

[0040] Step S233: Based on the coordinates of the predicted target box and the coordinates of each actual target box in the current frame corresponding to the initial tracking trajectory, obtain the intersection-union ratio of the predicted target box and each actual target box in the current frame.

[0041] Step S234: Based on the key matching degree of the current frame and the intersection-union ratio of the predicted target box and each actual target box in the current frame, obtain the IOU cost matrix.

[0042] Step S235: If any initial tracking trajectory was not confirmed in the previous frame of the current frame, or if any initial tracking trajectory was confirmed in the previous frame of the current frame but failed to match the target box in the current frame, or if any initial tracking trajectory failed to match the target box in the current frame, then the intersection-union ratio of the predicted target box and each actual target box in the current frame is obtained based on the coordinates of the predicted target box corresponding to the initial tracking trajectory in the current frame and the coordinates of each actual target box. Step S236: Obtain the IOU cost matrix based on the intersection-union ratio of the predicted target box and each actual target box in the current frame.

[0043] Step S240: Input the IOU cost matrix into the Hungarian algorithm for matching to obtain the matching results; wherein, the matching results include trajectory matching failure, target box matching failure, and trajectory and target box matching success.

[0044] Step S250: Update the state of each initial tracking trajectory according to the matching result corresponding to the current frame; wherein, if the matching result is that the trajectory and the target box are successfully matched, determine whether the corresponding initial tracking trajectory is in the confirmed state and update the initial tracking trajectory; if the matching result is that the trajectory matching fails and the number of consecutive matching failures is greater than a preset failure number threshold, delete the corresponding initial tracking trajectory; if the matching result is that the target box matching fails, create the initial tracking trajectory corresponding to the target box and determine the initial tracking trajectory as unconfirmed; and jump to the step of predicting the predicted target box coordinates of each initial tracking trajectory in the current frame according to the Kalman filter, until the end video frame of the preset video frame range is reached, then the target tracking ends.

[0045] In this embodiment, in the time dimension, each non-coplanar camera performs object tracking independently, and the tracking algorithm used for object tracking can be any object tracking algorithm. That is, in step S200, for multiple video frames corresponding to a certain time period (i.e., a preset video frame range), any object tracking algorithm can be used to track the target object. In this embodiment, the preset video frame range specifically refers to the set of all video frames corresponding to a certain time period, and the set of all video frames corresponding to different non-coplanar cameras is different. That is, the set of all video frames corresponding to each non-coplanar camera is the video frames it captured within the aforementioned time period. In the video frames captured by different non-coplanar cameras at the same time (i.e., in each video frame within the preset video frame range), the target object can be tracked across cameras using a virtual prism, that is, to determine whether the target objects appearing in different non-coplanar cameras at the same time are the same person or object. If they are the same person or object, the unique identifiers corresponding to the two will be set to be the same. In this way, cross-camera object tracking can be realized. Thus, cross-camera target tracking with multi-camera linkage is realized, and the computational requirements are greatly reduced.

[0046] like Figure 4 As shown, a cross-camera target tracking flowchart provided in one embodiment of this application illustrates the flow logic of the actual implementation of this solution. The specific algorithm steps are as follows: (1) Create the corresponding Tracks from the results detected in the first frame. Initialize the motion variables of the Kalman filter and predict the corresponding bounding boxes using the Kalman filter. At this point, the Tracks will definitely be unconfirmed.

[0047] (2) Perform IOU matching between the target detection bounding box in this frame and the bounding box predicted by Tracks in the previous frame, and then calculate the cost matrix (calculated as 1-IOU) based on the IOU matching result.

[0048] (3) Use all the cost matrices obtained in (2) as input to the Hungarian algorithm to obtain the linear matching result. There are three types of results. The first type is Tracks mismatch (Unmatched Tracks), which are directly deleted (because the Tracks are in an uncertain state. If they were in a deterministic state, they would need to be continuously deleted a certain number of times (30 times by default)). The second type is Detections mismatch (Unmatched Detections), which are initialized as new Tracks. The third type is that the detection box and the predicted box are successfully matched, which means that the previous frame and the next frame are successfully tracked. The corresponding Detections are updated with their corresponding Tracks variables through Kalman filtering.

[0049] (4) For successfully matched detection boxes, use the REID model to extract features and determine whether the target ID is in the tracker history queue. If it is, determine whether there is a spatially consistent association based on the target's 3D Euclidean distance. Otherwise, do not update the tracker and proceed to the next frame loop.

[0050] (5) If the 3D Euclidean distance is less than the set distance threshold, the feature similarity is calculated. If the similarity is less than the threshold, the tracker is not updated and the loop continues to the next frame. If the similarity is greater than the similarity threshold, the two targets are determined to be the same, and the target is assigned the corresponding ID in the historical tracker column and the loop continues to the next frame.

[0051] (6) Repeat steps (2)-(5) until a confirmed Tracks appears or the video frame ends.

[0052] (7) Predict the bounding boxes corresponding to the confirmed Tracks and the unconfirmed Tracks using Kalman filtering. Concatenate the bounding boxes of the confirmed Tracks and the Detections (previously, whenever a Track matched, the appearance features and motion information of the Detections were saved. The first 100 frames were saved by default. Concatenated matching of appearance features and motion information with the Detections was performed because confirmed Tracks and Detections are more likely to match).

[0053] (8) There are three possible results after cascaded matching. The first is Tracks matching, where Tracks are updated with their corresponding Tracks variables through Kalman filtering. The second and third are Detections and Tracks mismatch. In this case, the previously unconfirmed Tracks and the mismatched Tracks are matched with the Unmatched Detections one by one using IOU, and the cost matrix (calculated as 1-IOU) is then calculated based on the IOU matching results.

[0054] (9) Use all the cost matrices obtained in (8) as input to the Hungarian algorithm to obtain the linear matching result. There are three types of results. The first type is Tracks mismatch (Unmatched Tracks), which are directly deleted (because the Tracks are in an uncertain state. If they were in a deterministic state, they would need to be deleted after a certain number of times (30 times by default). The second type is Detections mismatch (Unmatched Detections), which are initialized as new Tracks. The third type is that the detection box and the predicted box are successfully matched, which means that the previous frame and the next frame are successfully tracked. The corresponding Detections are updated with their corresponding Tracks variables through Kalman filtering.

[0055] (10) Repeat steps (7)-(9) until the video frame ends.

[0056] Furthermore, although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0057] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, mobile terminal, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0058] Please refer to Figure 5As shown, an embodiment of this application provides a target tracking device 100 across cameras, the device comprising: Construction unit 110 is used to construct a virtual polygonal prism based on at least three non-coplanar cameras; wherein, the image view of each non-coplanar camera corresponds to one side of the polygonal prism; the adjacent spatial relationship between the sides of the polygonal prism is the same as the adjacent spatial relationship between the corresponding non-coplanar cameras; the two-dimensional coordinate direction of the image view of the non-coplanar cameras is consistent with the orthogonal basis vector direction space of the three-dimensional plane where the corresponding side of the polygonal prism is located. Tracking unit 120 is configured to track each target object within the field of view of each of at least three non-coplanar cameras within a preset video frame range; wherein each target object has a unique identifier within each video frame of the non-coplanar camera; each non-coplanar camera has a corresponding target object list; the target object list includes each unique identifier corresponding to each video frame within the preset video frame range; and, The tracking unit 130 is used to perform cross-camera target tracking of the target object in each video frame within a preset video frame range, based on a virtual polygonal prism; wherein, the cross-camera target tracking of the target object is determined based on the Euclidean distance between the target objects corresponding to the non-coplanar cameras of each two adjacent sides of the virtual polygonal prism in the corresponding video frame images.

[0059] In one exemplary embodiment of this application, the tracking unit 130 includes: The first acquisition subunit is used to acquire the list of target objects corresponding to each non-coplanar camera in the current video frame; The second acquisition subunit is used to acquire the Euclidean distance between each target object in the target object list of each non-coplanar camera in the current video frame and each target object in the target object list of each adjacent non-coplanar camera; wherein, the Euclidean distance is calculated based on the coordinate position of the target box of each target object in the virtual polygonal prism side in the current video frame; The matching subunit is used to obtain the matching degree between two target objects if the Euclidean distance between any two target objects in any two adjacent non-coplanar cameras is less than a preset distance threshold. The replacement subunit is used to replace the unique identifier of the non-critical target object with the unique identifier of the critical target object if the matching degree between the two target objects is greater than a preset matching degree threshold; wherein, the number of video frames in which the critical target object appears in the target object list of the corresponding non-coplanar camera is greater than the number of video frames in which the non-critical target object appears in the target object list of the corresponding non-coplanar camera.

[0060] In one exemplary embodiment of this application, the matching subunit includes: The feature acquisition component is used to extract the object features of the bounding box of each of the two target objects according to the re-identification model if the Euclidean distance between any two target objects in any two adjacent non-coplanar cameras is less than a preset distance threshold.

[0061] The matching component is used to obtain the degree of matching between the object features of the two target objects.

[0062] In one exemplary embodiment of this application, after the second acquisition subunit, the apparatus further includes: The determination subunit is used to determine that if the Euclidean distance between any two target objects in any two adjacent non-coplanar cameras is equal to or greater than a preset distance threshold, then the same target object is not currently appearing simultaneously within the shooting range of different non-coplanar cameras.

[0063] In one exemplary embodiment of this application, the tracking unit 120 includes: The trajectory acquisition subunit is used to acquire the coordinates of the target box of each target object and the initial tracking trajectory of each target object in the first frame of the first tracking frame of each of the several non-coplanar cameras; wherein each initial tracking trajectory is in an unconfirmed state. The coordinate acquisition subunit is used to predict the coordinates of the predicted target box corresponding to each initial tracking trajectory in the current frame based on the Kalman filter; The matrix acquisition sub-unit is used to obtain the IOU cost matrix corresponding to each initial tracking trajectory based on the coordinates of the predicted target box corresponding to each initial tracking trajectory in the current frame and the actual target box of each target object in the current frame. The result acquisition sub-unit is used to input the IOU cost matrix into the Hungarian algorithm for matching to obtain the matching results; the matching results include trajectory matching failure, target box matching failure, and trajectory and target box matching success. The update subunit is used to update the state of each initial tracking trajectory according to the matching result corresponding to the current frame. If the matching result is that the trajectory and the target box match successfully, it is determined whether the corresponding initial tracking trajectory is in the confirmed state and the initial tracking trajectory is updated. If the matching result is that the trajectory matching fails and the number of consecutive matching failures is greater than a preset failure number threshold, the corresponding initial tracking trajectory is deleted. If the matching result is that the target box matching fails, an initial tracking trajectory corresponding to the target box is created and the initial tracking trajectory is determined to be in the unconfirmed state. Then, the process jumps to the step of predicting the predicted target box coordinates of each initial tracking trajectory in the current frame according to the Kalman filter, until the target tracking ends at the end video frame of the preset video frame range.

[0064] In one exemplary embodiment of this application, the matrix acquisition subunit includes: The historical feature acquisition component is used to acquire the historical actual features corresponding to a preset number of historical video frames before the current frame if any initial tracking trajectory is confirmed as a confirmed trajectory in the previous frame and the current frame is successfully matched. The key matching component is used to obtain the key matching degree of the initial tracking trajectory in the current frame based on the current actual features corresponding to the actual bounding box of each target object in the current frame and the historical actual features corresponding to the initial tracking trajectory in a preset number of historical video frames before the current frame; wherein, the key matching degree is the maximum matching degree between all corresponding current actual features and all corresponding historical actual features. The first intersection-over-union (IoU) determination component is used to obtain the IoU of the predicted target box and each actual target box in the current frame based on the coordinates of the predicted target box and the coordinates of each actual target box in the current frame corresponding to the initial tracking trajectory. The first matrix acquisition component is used to obtain the IOU cost matrix based on the key matching degree of the current frame and the intersection-union ratio of the predicted target box and each actual target box in the current frame.

[0065] In one exemplary embodiment of this application, the matrix acquisition subunit further includes: The second intersection-over-union (IoU) determination component is used to determine the IoU ratio of the predicted target box and each actual target box in the current frame if any initial tracking trajectory was not confirmed in the previous frame of the current frame, or if any initial tracking trajectory was confirmed in the previous frame of the current frame but failed to match the target box in the current frame, or if any initial tracking trajectory failed to match the target box in the current frame. The second matrix acquisition component obtains the IOU cost matrix based on the intersection-union ratio of the predicted target box and each actual target box in the current frame.

[0066] In an exemplary embodiment of this disclosure, an electronic device capable of implementing the above-described method is also provided.

[0067] Those skilled in the art will understand that various aspects of this application can be implemented as a system, method, or program product. Therefore, various aspects of this application can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, collectively referred to herein as a "circuit," "module," or "system."

[0068] An electronic device according to this embodiment of the present application. The electronic device is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of this application.

[0069] Electronic devices are manifested in the form of general-purpose computing devices. Components of an electronic device may include, but are not limited to: at least one processor, at least one memory, and buses connecting different system components (including memory and processor).

[0070] The aforementioned storage device stores program code that can be executed by the aforementioned processor, causing the processor to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this application.

[0071] The storage may include readable media in the form of volatile storage, such as random access memory (RAM) and / or cache memory, and may further include read-only memory (ROM).

[0072] The storage may also include programs / utilities having a set (at least one) of program modules, including but not limited to: an operating system, one or more applications, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0073] A bus can represent one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus that uses any of the various bus architectures.

[0074] The electronic device can also communicate with one or more external devices (e.g., keyboards, pointing devices, Bluetooth devices, etc.), one or more devices that enable a user to interact with the electronic device, and / or any device that enables the electronic device to communicate with one or more other computing devices (e.g., routers, modems, etc.). This communication can be performed via input / output (I / O) interfaces. Furthermore, the electronic device can communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter. The network adapter communicates with other modules of the electronic device via a bus. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with the electronic device, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0075] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0076] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the methods described above is stored. In some possible implementations, various aspects of this application may also be implemented as a program product including program code, which, when the program product is run on a terminal device, causes the terminal device to perform the steps of the various exemplary embodiments of this application described in the "Exemplary Methods" section of this specification.

[0077] The above-described program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0078] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, apparatus, or device.

[0079] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0080] Program code for performing the operations of this application can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0081] Furthermore, the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of this application, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0082] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0083] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for cross-camera target tracking, characterized in that, The method comprises: constructing a virtual polygonal prism according to the at least three non-coplanar cameras; wherein the image view angle of each non-coplanar camera corresponds to a side of the polygonal prism; the adjacent spatial relationship between the sides of the polygonal prism is the same as the adjacent spatial relationship between the corresponding non-coplanar cameras; the two-dimensional coordinate direction of the image view angle of the non-coplanar camera is consistent with the orthogonal basis vector direction of the three-dimensional plane in which the corresponding side of the polygonal prism is located; tracking each target object in the view angle range of each non-coplanar camera in a preset video frame range; wherein each target object has a corresponding unique identifier in each video frame image of the non-coplanar camera in which the target object is located; each non-coplanar camera has a corresponding target object list; the target object list includes each unique identifier corresponding to each video frame in the preset video frame range; and in each video frame in the preset video frame range, performing cross-camera target tracking on the target objects according to the virtual polygonal prism; wherein the cross-camera target tracking of the target objects is determined according to the Euclidean distance between the corresponding target objects of the non-coplanar cameras corresponding to each two adjacent sides of the virtual polygonal prism in the corresponding video frame image.

2. The cross-camera target tracking method according to claim 1, wherein, The cross-camera target tracking of the target objects according to the virtual polygonal prism comprises: obtaining the target object list of each non-coplanar camera in the current video frame; obtaining the Euclidean distance between each target object in the target object list of each non-coplanar camera in the current video frame and each target object in the target object list of each adjacent non-coplanar camera; wherein the Euclidean distance is calculated according to the coordinate position of the target box of each target object in the current video frame in the side of the virtual polygonal prism; if the Euclidean distance between any two target objects in any two adjacent non-coplanar cameras is less than a preset distance threshold, obtaining the matching degree between the two target objects; if the matching degree between the two target objects is greater than a preset matching degree threshold, replacing the unique identifier of the non-key target object with the unique identifier of the key target object in the two target objects; wherein the number of video frames in which the key target object appears in the target object list of the corresponding non-coplanar camera is greater than the number of video frames in which the non-key target object appears in the target object list of the corresponding non-coplanar camera. 3.The cross-camera target tracking method of claim 2, wherein, The method further comprises: extracting the object features of the target box of each target object in the two target objects according to a re-identification model; obtaining the matching degree between the object features of the two target objects. 4.The cross-camera target tracking method according to claim 2 or 3, characterized in that, After obtaining the Euclidean distance between each target object in the target object list of each non-coplanar camera in the current video frame and each target object in the target object list of each adjacent non-coplanar camera, the method further comprises: if the Euclidean distance between any two target objects in any two adjacent non-coplanar cameras is equal to or greater than a preset distance threshold, determining that there is no same target object currently appearing simultaneously in the shooting range of different non-coplanar cameras.

5. The cross-camera target tracking method according to claim 1, wherein, In a preset video frame range, target tracking is performed on each target object in a view angle range of each non-coplanar camera in at least three non-coplanar cameras, including: obtaining the coordinates of the target frame of each target object and the initial tracking trajectory of each target object corresponding to each non-coplanar camera in a plurality of non-coplanar cameras in a first frame of starting tracking; wherein each initial tracking trajectory is in an unconfirmed state; predicting the coordinates of the predicted target frame corresponding to each initial tracking trajectory in the current frame according to the Kalman filter; obtaining the IOU cost matrix corresponding to each initial tracking trajectory according to the coordinates of the predicted target frame corresponding to each initial tracking trajectory in the current frame and the actual target frame of each target object in the current frame; inputting the IOU cost matrix into the Hungarian algorithm for matching to obtain a matching result; wherein the matching result includes track matching failure, target frame matching failure, and track and target frame matching success; updating the state of each initial tracking trajectory according to the matching result corresponding to the current frame; wherein if the matching result is track and target frame matching success, it is determined whether the corresponding initial tracking trajectory is in a confirmed state, and the initial tracking trajectory is updated; if the matching result is track matching failure, and the number of consecutive matching failures is greater than a preset failure number threshold, the corresponding initial tracking trajectory is deleted; if the matching result is target frame matching failure, an initial tracking trajectory corresponding to the target frame is created, and the initial tracking trajectory is determined to be in an unconfirmed state; and jumping to the step of predicting the coordinates of the predicted target frame corresponding to each initial tracking trajectory in the current frame according to the Kalman filter until the end video frame of the preset video frame range is reached, and then the target tracking ends.

6. The cross-camera target tracking method according to claim 5, characterized in that, According to the coordinates of the predicted target frame corresponding to each initial tracking trajectory in the current frame and the actual target frame of each target object in the current frame, the IOU cost matrix is obtained, including: if any initial tracking trajectory is confirmed as a confirmed state trajectory in the previous frame of the current frame and the current frame matching is successful, the historical actual features corresponding to a preset number of historical video frames before the current frame of the initial tracking trajectory are obtained; obtaining the key matching degree of the initial tracking trajectory in the current frame according to the current actual features corresponding to each actual target frame in the current frame and the historical actual features corresponding to a preset number of historical video frames before the current frame of the initial tracking trajectory; wherein the key matching degree is the maximum matching degree between all corresponding current actual features and all corresponding historical actual features; obtaining the intersection over union corresponding to the predicted target frame and each actual target frame in the current frame according to the coordinates of the predicted target frame corresponding to the initial tracking trajectory in the current frame and the coordinates of each actual target frame; obtaining the IOU cost matrix according to the key matching degree of the current frame and the intersection over union corresponding to the predicted target frame and each actual target frame in the current frame.

7. The cross-camera target tracking method according to claim 6, wherein, According to the coordinates of the predicted target frame corresponding to each initial tracking trajectory in the current frame and the actual target frame of each target object in the current frame, the IOU cost matrix is obtained, further including: If any initial tracking trajectory is not confirmed as a confirmed state in the previous frame of the current frame, or any initial tracking trajectory is confirmed as a confirmed state trajectory in the previous frame of the current frame but fails to match the target frame in the current frame, or any initial tracking trajectory fails to match the target frame in the current frame, then according to the coordinates of the predicted target frame corresponding to the initial tracking trajectory in the current frame and the coordinates of each actual target frame, the intersection over union of the predicted target frame and each actual target frame corresponding to the current frame is obtained. According to the intersection over union of the predicted target frame and each actual target frame corresponding to the current frame, an IOU cost matrix is obtained.

8. A cross-camera target tracking apparatus, characterized by, The device comprises: The construction unit is configured to construct a virtual polygonal prism according to the at least three non-coplanar cameras; wherein the image perspective of each non-coplanar camera corresponds to a side of the polygonal prism; the adjacent spatial relationship between the sides of the polygonal prism is the same as the adjacent spatial relationship between the corresponding non-coplanar cameras; the two-dimensional coordinate direction of the image perspective of the non-coplanar camera is spatially consistent with the direction of the orthogonal basis vector of the three-dimensional plane corresponding to the side of the polygonal prism; The tracking unit is configured to perform target tracking on each target object in the perspective range of each non-coplanar camera in a preset video frame range; wherein each target object has a corresponding unique identifier in each video frame image of the non-coplanar camera; each non-coplanar camera has a corresponding target object list; the target object list includes each unique identifier corresponding to each video frame in the preset video frame range; and The tracking unit is configured to perform cross-camera target tracking on the target object according to the virtual polygonal prism in each video frame in the preset video frame range; wherein the cross-camera target tracking of the target object is determined according to the Euclidean distance between the corresponding target objects of the non-coplanar cameras corresponding to each two adjacent sides of the virtual polygonal prism in the corresponding video frame image. 9.A non-transitory computer-readable storage medium having stored therein at least one instruction or at least one piece of program, characterized in that, The at least one instruction or the at least one program is loaded and executed by the processor to implement the method of any one of claims 1-7.

10. An electronic device, comprising: The device comprises a processor and a non-transitory computer readable storage medium as claimed in claim 9. The device comprises a processor and a non-transitory computer readable storage medium as claimed in claim 9.

Images