Cross-camera tracking methods, devices, electronic equipment, and computer program products

Abstract

This application relates to the field of security technology, providing a cross-camera tracking method, apparatus, electronic device, and computer program product. The method includes: determining path information between cameras; determining a first camera to be detected based on a target user's first movement direction and path information, and acquiring an image of the target user based on the first camera to be detected; determining the credibility of the target user image based on a first image feature of the target user image and a second image feature of a template image; storing the target user image in a memory sequence based on the credibility; determining a second camera to be detected based on the memory sequence, path information, and a second movement direction of the target user, and tracking the target user based on the second camera to be detected. This application improves the autonomous capability of cross-camera tracking by calling the detection camera based on the target user's movement direction and path information, and improves the accuracy of cross-camera tracking by adding a memory storage strategy to the target detection algorithm.

Description

Technical Field

[0001] This application relates to the field of security technology, specifically to a cross-camera tracking method, apparatus, electronic device, and computer program product. Background Technology

[0002] Cross-camera tracking involves object detection and object tracking technologies. As two important tasks in computer vision, object detection and object tracking work by learning target features and building algorithm models through traditional machine learning or deep learning methods to autonomously discover the target location in an image. Object tracking requires additional continuous searching for the target in consecutive video frames.

[0003] Currently, camera network construction only considers distance and doesn't adapt to specific scenarios, leading to resource waste. Furthermore, blind spots between adjacent cameras can cause target loss if the user changes route before the next camera. On the other hand, existing deep learning-based object detection algorithms (such as R-CNN, YOLO, and SSD) haven't been modified for the unique challenges of cross-camera tracking, resulting in inadequate stability and accuracy for cross-camera tracking tasks. Therefore, existing cross-camera tracking methods suffer from low tracking accuracy. Summary of the Invention

[0004] This application provides a cross-camera tracking method, apparatus, electronic device, and computer program product to solve the technical problem of low accuracy in cross-camera tracking.

[0005] In a first aspect, embodiments of this application provide a cross-camera tracking method, including:

[0006] Determine the path information between each camera;

[0007] The first camera to be detected is determined based on the first direction of movement of the target user and the path information, and the image of the target user is acquired based on the first camera to be detected.

[0008] The credibility of the target user image is determined based on the first image features of the target user image and the second image features of the template image.

[0009] The target user image is stored in a memory sequence based on the credibility level.

[0010] The second camera to be detected is determined based on the memory sequence, the path information, and the second movement direction of the target user, and the target user is tracked based on the second camera to be detected.

[0011] In one embodiment, determining the credibility of the target user image based on a first image feature of the target user image and a second image feature of the template image includes:

[0012] Determine the third image feature of the initial template image, wherein the template image includes the initial template image;

[0013] Determine the maximum response value between the first image feature and the third image feature;

[0014] The credibility of the target user image is determined based on the maximum response value, the second image feature, and the number of template images.

[0015] In one embodiment, the method further includes:

[0016] If the number of template images in the memory sequence is greater than or equal to a set number, then the memory degree value of each template image is determined according to preset parameters and the credibility of each template image.

[0017] The template image to be deleted is determined based on the memory level value;

[0018] Delete the template image to be deleted.

[0019] In one embodiment, determining the memorability value of each template image based on preset parameters and the credibility of each template image includes:

[0020] If the credibility of the template image is greater than the first preset value, then the memory level of the template image is determined based on the initial memory level value of the template image and the first reinforcement parameter.

[0021] If the credibility of the template image is less than or equal to the first preset value, then the memory level of the template image is determined based on the initial memory level value, the first reinforcement parameter, the temporal value, and the second reinforcement parameter.

[0022] The preset parameters include the initial memory level value, the first reinforcement parameter, the time sequence value, and the second reinforcement parameter.

[0023] In one embodiment, determining the path information between the cameras includes:

[0024] Determine the application scenario of the target area, as well as the location information and field of view of each camera in the target area;

[0025] Based on the location information of each camera, its field of view, and the application scenario, the path information between the cameras is determined.

[0026] In one embodiment, after determining the path information between the cameras, the process includes:

[0027] The identification information of each camera is determined based on the path information;

[0028] Associate the identification information and the path information.

[0029] In one embodiment, the method further includes:

[0030] Determine the first weight value for face recognition and the second weight value for appearance detection;

[0031] The target detection algorithm is determined based on the first weight value, the second weight value, the memory sequence, the channel attention module, the spatial attention module, and the convolutional network management model.

[0032] The target user is detected according to the target detection algorithm.

[0033] Secondly, embodiments of this application provide a cross-camera tracking device, comprising:

[0034] The first determining module is used to determine the path information between each camera;

[0035] The second determining module is used to determine the first camera to be detected based on the first movement direction of the target user and the path information, and to acquire the target user image based on the first camera to be detected.

[0036] The third determining module is used to determine the credibility of the target user image based on the first image features of the target user image and the second image features of the template image;

[0037] A storage module is used to store the target user image into a memory sequence according to the credibility level;

[0038] The tracking module is used to determine the second camera to be detected based on the memory sequence, the path information, and the second movement direction of the target user, and to track the target user based on the second camera to be detected.

[0039] Thirdly, embodiments of this application provide an electronic device, including a processor and a memory storing a computer program, wherein the processor executes the program to implement the steps of the cross-camera tracking method described in the first aspect.

[0040] Fourthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the steps of the cross-camera tracking method described in the first aspect.

[0041] The cross-camera tracking method provided in this application determines the path information between cameras; determines the first camera to be detected based on the first movement direction and path information of the target user, and acquires the target user image based on the first camera to be detected; determines the credibility of the target user image based on the first image features of the target user image and the second image features of the template image; stores the target user image in a memory sequence based on the credibility; determines the second camera to be detected based on the memory sequence, path information, and the second movement direction of the target user, and tracks the target user based on the second camera to be detected. In this way, the next detection camera is called based on the movement direction and path information of the target user, which improves the autonomous capability of cross-camera tracking. At the same time, the addition of a memory storage strategy to the target detection algorithm improves the accuracy of cross-camera tracking. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is one of the flowcharts illustrating the cross-camera tracking method provided in this application embodiment;

[0044] Figure 2 This is a second schematic flowchart of the cross-camera tracking method provided in the embodiments of this application;

[0045] Figure 3 This is the third flowchart illustrating the cross-camera tracking method provided in this application embodiment;

[0046] Figure 4 This is the fourth flowchart illustrating the cross-camera tracking method provided in this application embodiment;

[0047] Figure 5 This is the fifth flowchart illustrating the cross-camera tracking method provided in this application embodiment;

[0048] Figure 6 This is a flowchart illustrating the principle of cross-camera tracking provided in an embodiment of this application;

[0049] Figure 7 This is a schematic diagram of camera selection decision provided in an embodiment of this application;

[0050] Figure 8 This is a schematic diagram of a cross-camera tracking process based on path planning provided in an embodiment of this application;

[0051] Figure 9This is a schematic diagram of the cross-mirror tracking device provided in the embodiments of this application;

[0052] Figure 10 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, 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.

[0054] Figure 1 This is one of the flowcharts illustrating the cross-camera tracking method provided in this application. (Refer to...) Figure 1 This application provides a cross-camera tracking method, which may include:

[0055] Step S10: Determine the path information between each camera;

[0056] In this embodiment of the application, path planning is performed in advance based on the application scenario of the target area, the location information of the camera, and the field of view to obtain the path information between each camera, as described in steps S11 to S12.

[0057] Step S20: Determine the first camera to be detected based on the first movement direction of the target user and the path information, and acquire the target user image based on the first camera to be detected;

[0058] In this embodiment, during target tracking, the target user's movement direction is detected in real time. Then, based on the target user's first movement direction and path information, the first camera to be detected can be determined. It is understood that the first camera to be detected refers to the next camera to track the target user. Further, the first camera to be detected is invoked to capture images of the target user. For example, refer to... Figure 7 Numbers 1 to 8 represent camera IDs, and there is path information between each camera. The arrow indicates the direction of movement of the target user. When the target user moves from the field of view of camera 1 in the direction indicated by the arrow, camera 2 can be identified as the camera to be detected by combining the path information. At this time, camera 2 is called to track the target user, while camera 1 continues to work for a period of time to prevent the target user from turning back. When the target user enters the field of view of camera 2, camera 2 automatically captures the image of the target user.

[0059] Step S30: Determine the credibility of the target user image based on the first image features of the target user image and the second image features of the template image;

[0060] To improve the accuracy of cross-camera tracking, the target detection algorithm in this application incorporates a memory storage strategy. This memory storage strategy is used to store target user images with relatively high credibility. Therefore, it is necessary to determine the credibility of the target user images captured by each camera.

[0061] Specifically, the first image features of the target user image and the second image features of the template image in the memory sequence are determined. It can be understood that the first image features include the facial features and appearance features of the target user image, and the second image features include the facial features and appearance features of the template image. The template image refers to the target user image stored in the memory sequence. Then, the credibility of the target user image is determined based on the first image features and the second image features, as described in steps S21 to S23.

[0062] Step S40: Store the target user image into a memory sequence based on the confidence level;

[0063] In this embodiment of the application, after determining the credibility of the target user image, the credibility is compared with a preset value τ1. If the credibility of the target user image is greater than τ1, it indicates that the prediction result of the target user image is reliable, and the target user image is stored in the memory sequence.

[0064] Step S50: Determine the second camera to be detected based on the memory sequence, the path information, and the second movement direction of the target user, and track the target user based on the second camera to be detected.

[0065] In this embodiment, a second camera to be detected is determined based on a memory sequence, path information, and the second movement direction of the target user. The target user is then tracked using the second camera to be detected. For example, a template image in the memory sequence is used as a detection template to detect the target user image captured by the current camera. If the target user image is determined to be reliable based on the detection template, the target user image is stored in the memory sequence. At the same time, the second movement direction of the target user is obtained. The second camera to be detected is determined based on the second movement direction of the target user and the corresponding path information between each camera. Then, the target user is tracked using the second camera to be detected.

[0066] This application embodiment determines the path information between cameras; determines the first camera to be detected based on the first movement direction and path information of the target user, and acquires the target user image based on the first camera to be detected; determines the credibility of the target user image based on the first image features of the target user image and the second image features of the template image; stores the target user image in a memory sequence based on the credibility; determines the second camera to be detected based on the memory sequence, path information, and the second movement direction of the target user, and tracks the target user based on the second camera to be detected. In this way, the next detection camera is called based on the movement direction and path information of the target user, which improves the autonomous capability of cross-camera tracking. At the same time, the addition of a memory storage strategy to the target detection algorithm improves the accuracy of cross-camera tracking.

[0067] Further, refer to Figure 2 , Figure 2 This is a second schematic flowchart of the cross-camera tracking method provided in this application embodiment, wherein step S30 includes:

[0068] Step S31: Determine the third image feature of the initial template image, wherein the template image includes the initial template image;

[0069] Step S32: Determine the maximum response value between the first image feature and the third image feature;

[0070] Step S33: Determine the credibility of the target user image based on the maximum response value, the second image feature, and the number of template images.

[0071] When the first camera begins tracking the target user, if the target user appears in the camera's field of view, the facial and appearance features of the target user are extracted using a target detection algorithm for subsequent cross-camera detection tasks. When the tracking task begins, a true target feature template is obtained for tracking with the initial camera. After the initial camera tracking ends, if only the initial template image is used as the detection template, the target morphology understood by the target detection algorithm will be too simplistic. When the appearance of the target user changes significantly due to angle and position changes during cross-camera tracking, the difficulty for the target detection algorithm to correctly locate the target user again will increase. Based on this, this application's embodiments establish a memory storage strategy. From the start of the tracking task, the tracking result of each frame is compared with the template image to obtain a similarity value, and the initial template image is used as the first feature map in the memory sequence, forming a "memory sequence".

[0072] In this embodiment, it is necessary to determine the credibility of the target user image captured by the camera, and then store the target user image whose credibility meets the set conditions in a memory sequence. Specifically, the third image feature of the initial template image is determined. It can be understood that the third image feature includes the facial features and appearance features of the initial template image. The template image includes the initial template image, which refers to the target user image captured by the first camera. Then, the maximum response value of the first image feature and the third image feature is determined. Finally, based on the maximum response value, the second image feature, and the number of template images, the credibility of the target user image is determined. The formula for the memory storage rule is as follows:

[0073]

[0074] Where x represents the first image feature of the current target user image, z represents the third image feature of the initial template image, max(h(x,z)) represents the maximum response value between the first and second image features. This means that the maximum response value is calculated as the peak value between the first and second image features, and this peak value is taken as the maximum response value. N represents the memory length, i.e., the number of template images, and f(z) i ) represents the second image feature of each template image in the memory sequence. The ratio of the numerator to the denominator represents the average value of the target response in the memory sequence. The ratio of the numerator to the denominator represents the confidence level of the current target user image. When the ratio is greater than τ1, it indicates that the prediction result of the target user image is reliable, and the target user image is stored in the memory sequence.

[0075] This application embodiment improves the accuracy of cross-camera tracking by formulating a memory storage strategy to determine the credibility of the target user image.

[0076] Further, refer to Figure 3 , Figure 3 This is a third schematic flowchart of the cross-camera tracking method provided in this application embodiment. The method further includes:

[0077] Step S60: If the number of template images in the memory sequence is greater than or equal to a set number, then the memory degree value of each template image is determined according to the preset parameters and the credibility of each template image.

[0078] Step S70: Determine the template image to be deleted based on the memory level value;

[0079] Step S80: Delete the template image to be deleted.

[0080] Understandably, the memory sequence stores each template image and its image features, including appearance features and facial features, with appearance features being the primary stored image features. The memory sequence also has a maximum storage capacity. When the stored image features reach this maximum capacity, a comparative analysis is performed on the stored image features, discarding the lower-quality features and adding new ones.

[0081] In this embodiment, if the number of template images in the memory sequence is greater than or equal to a set number, the memory level value of each template image is determined according to preset parameters and the credibility of each template image. Specifically, if the credibility of a template image is greater than a first preset value, the memory level value of the template image is determined according to the initial memory level value and a first reinforcement parameter; if the credibility of a template image is less than or equal to the first preset value, the memory level value of the template image is determined according to the initial memory level value, the first reinforcement parameter, the temporal value, and a second reinforcement parameter. The preset parameters include the initial memory level value, the first reinforcement parameter, the temporal value, and the second reinforcement parameter. Then, template images to be deleted are determined based on the memory level values, and these template images are deleted. The initial template image, as the true value, is permanently retained. The memory update strategy mimics the forgetting curve of human memory and is modeled as follows:

[0082]

[0083] Among them, c i This represents the degree of memory for the i-th image feature in the memory sequence. It is initialized to 1 at the start of tracking, indicating reliable memory. k (>0) represents the reinforcement parameter, i.e., the first reinforcement parameter. Meeting certain conditions proves that the image feature quality is good. When f... <h(g(z i If h(g(z))>>τ2, it indicates that the credibility of the template image is greater than the first preset value. At this time, according to min(1,kc i Determine the memory level of the template image.

[0084] b i It is the base value for memory, initially set to 1, and subsequently used in process b. i =min(1,kc) i ), a i This represents the timing value; when the first condition of the formula is met, a is... i Initialize to 0, otherwise a i =a i +1, h represents the memory strength parameter, i.e., the second reinforcement parameter, when f <h(g(z i If h(g(z))>≤τ2, it means that the credibility of the template image is less than or equal to the first preset value. At this time, according to b iexp(-a i / h) Determine the memory level of the template image.

[0085] Through modeling, when a new image feature is similar to existing memory, it is considered to be of high quality, its corresponding memory level is increased, and its temporal value is initialized to 0. When they are dissimilar, the temporal value increases, and the memory level decreases as the temporal value increases. During the memory update phase, the image feature with the lowest memory level is deleted. For example, after determining the memory level values ​​of each template image, the memory level values ​​are sorted. Based on the sorting results, the template image with the lowest memory level value is determined and designated as the template image to be deleted. This template image is then further deleted.

[0086] This application embodiment determines the memory level value of each template image, then determines the template image with the lowest memory level based on the memory level value, and deletes the template image, thereby freeing up the storage space of the memory sequence. At the same time, by deleting the template image with the lowest memory level, the accuracy of cross-camera tracking can be improved.

[0087] Further, refer to Figure 4 , Figure 4 This is a fourth flowchart illustrating the cross-camera tracking method provided in this application embodiment. Step S10 includes:

[0088] Step S11: Determine the application scenario of the target area, as well as the location information and field of view of each camera in the target area;

[0089] Step S12: Determine the path information between the cameras based on the location information of each camera, the field of view, and the application scenario.

[0090] It's important to note that when constructing the camera network, users often only consider the distance between cameras without adapting to specific scenarios. This leads to wasted resources. For example, in a shopping mall, adjacent cameras may be close together but isolated by walls or their coverage areas may be unconnected. If the wrong adjacent cameras are used based on user movement, it not only wastes resources but may also cause tracking failure. Furthermore, blind spots between adjacent cameras can lead to the target being lost if the user changes course before the next camera. Therefore, these factors result in poor autonomy in cross-camera tracking, ultimately causing the tracking task to fail.

[0091] In this embodiment, the application scenario of the target area is determined, as well as the location information and field of view of each camera in the target area. Then, path information is determined based on the location information, field of view of each camera, and application scenario. Next, the identification information of each camera is determined based on the path information, and the identification information is associated with the path information. Specifically, detailed path planning is performed for the specific scenario, as well as the location information and field of view of each camera. The arrangement of cameras is mapped to a two-dimensional space based on the planned path information to determine the hierarchical relationship between cameras and to encode the cameras. In cross-camera tracking, path planning plays a crucial role as a prerequisite for the overall solution execution. The successful implementation of the overall solution depends on the completeness and rigor of the pre-planned path. The specific principle is as follows: First, the roads with cameras in the area are organized and drawn, and the distribution of cameras is marked on the map. Starting from the entrance, a main road and several branch roads are drawn. Starting from the starting point, the cameras on the main road are assigned IDs starting from 1. Then, the same operation is performed on the branches in sequence, that is, the relative main paths are found within the branches, and IDs are assigned in sequence, ultimately resulting in... Figure 7 The marking results are shown below. Each camera within the area is a recording point. These points are connected to form a tree structure, ultimately resulting in an overall route planning map. All cameras and their IDs are encoded and input into the algorithm. For example, the first camera at the area entrance is designated as camera 1. Starting from this point, the second camera on the reachable path is designated as camera 2. This rule is followed by continuous encoding along the path. When a fork in the road is reached, the first camera entering each fork is further encoded. After encoding, the encoded ID of each camera and the encoded IDs of cameras connectable along the path are combined into structured data and sent to the server for storage. This implements the route planning strategy while also ignoring cases where adjacent cameras are not actually connected, such as those behind walls.

[0092] For example, refer to Figure 8 , Figure 8 This is a schematic diagram of a cross-camera tracking process based on path planning provided in this application embodiment. In this application embodiment, when the tracking task begins, the camera is invoked to track the target user. If the target user disappears from the current camera's field of view, the target user's movement direction is determined based on the target user's coordinate information. Furthermore, the camera corresponding to that direction is determined based on the movement direction and path information, and then that camera is invoked to detect the target user. When the target user appears within the camera's field of view, it is confirmed that the target user has been detected, and tracking of the target user continues. This improves the autonomous capability of cross-camera tracking.

[0093] In one specific embodiment, taking a shopping mall as an example, the following analysis illustrates the application scenario. In a shopping mall, customers' walking routes are relatively fixed, mostly consisting of single roads or circular routes. Cameras can be considered linearly connected, eliminating the need to consider adjacent cameras. Forks in the road may occur at intersections or stores, requiring branching. Camera IDs are assigned based on the defined routes, effectively preventing resource waste. For instance, if a target user is walking normally on a floor of the mall and needs to cross camera lines, only the next or previous camera on that route needs to be considered (considering backtracking). Therefore, when calling up cameras, a small number of cameras can be called according to the pre-planned path to complete the camera selection problem when crossing camera lines, thereby improving the autonomous capability of cross-camera tracking.

[0094] This application embodiment combines the actual roads, camera locations, and field of view within the tracking area, assigns ID codes to cameras according to main and branch roads, plans all paths that pedestrians can travel, selects paths based on the pedestrian's movement direction, and predicts which subsequent cameras will be called. In this way, the autonomous capability of cross-camera tracking can be improved.

[0095] Further, refer to Figure 5 , Figure 5 This is the fifth flowchart illustrating the cross-camera tracking method provided in this application embodiment. The method further includes:

[0096] Step S90: Determine the weight values ​​for face recognition and appearance detection;

[0097] Step S100: Determine the target detection algorithm based on the weight values, the memory sequence, the channel attention module, the spatial attention module, and the convolutional network management model;

[0098] Step S110: Detect the target user according to the target detection algorithm.

[0099] It should be noted that when performing target detection across cameras, face detection and appearance feature detection are combined to extract the target user's features from multiple levels and achieve better detection results. However, considering the actual installation position, height, and angle of the surveillance camera, the detection effect achieved by face detection and face recognition technology is relatively poor. In addition, there are situations where the user's back is to the camera, which greatly inhibits the effectiveness of the face detection algorithm. On the other hand, the target detection algorithm uses publicly available deep learning-based target detection algorithms (RCNN series, YOLO series, SSD, etc.), without corresponding improvements for the specificities of cross-camera tracking. As a result, the stability and accuracy of the detection algorithm cannot be fully applied to cross-camera tracking tasks. Based on this, the embodiments of this application optimize the target detection algorithm, including adjusting the weights of face recognition and appearance detection, adding an attention mechanism, and a memory storage strategy. Among them, the memory storage strategy is the primary approach, while the weight adjustment and attention mechanism are secondary approaches. Based on the combined effect of these three aspects, accurate capture of cross-camera targets is achieved.

[0100] In this embodiment, a first weight value for face recognition and a second weight value for appearance detection are determined. Then, based on the first weight value, the second weight value, the memory sequence, the channel attention module, the spatial attention module, and the convolutional network management model, a target detection algorithm is determined, and the target user is detected according to the target detection algorithm. Specifically, the weight of face recognition is reduced while the weight of appearance detection is increased, i.e., the first weight value is set to be less than the second weight value. Furthermore, on the main framework of the original detection algorithm Yolov5, after the backbone network extracts features from the input image, parallel channel attention modules and spatial attention modules are added. Channel attention enables the algorithm to focus on feature channels containing more effective information about the target, while spatial attention enables the algorithm to focus on more target information on each feature map. After the two attention modules are connected in parallel, convolution operations are performed to obtain a feature map with the same size and number of channels as before. Thus, without changing the attributes of the feature map, it simultaneously possesses autonomous perception capabilities in both space and channels, improving the target detection capability. Finally, a memory sequence is added to the target detection algorithm, i.e., a memory storage strategy is added. If multiple similar targets exist in a single frame, facial recognition can be used to distinguish between the real target and the interfering target.

[0101] Furthermore, target users are detected using object detection algorithms. Specifically, target regions are determined based on object detection algorithms and target appearance features, and then target users are detected based on these target regions. Target appearance features refer to the appearance features of the target user extracted from the target user's image. For example, using object detection algorithms and previously extracted target appearance features, the target user is searched within the field of view of the camera to be detected. The previous camera continues operating for a period to prevent the target from returning. This can be understood as follows: during the cross-camera target detection phase, the feature maps in the memory sequence are used as templates, and the region with the highest average similarity to the template group is identified as the target region. Then, the target user is detected within the target region.

[0102] This application's embodiments reduce the weight of face recognition and increase the weight of appearance recognition. This allows for less reliance on face detection under normal circumstances, while enabling the identification of real and fake targets using face detection when similar users are detected. Furthermore, the target detection network algorithm is designed and optimized for cross-camera tracking tasks. This involves incorporating an attention mechanism and a memory storage strategy into the target detection algorithm, enabling it to focus more on accurately identifying the same user and effectively distinguishing between different users, thus improving the algorithm's perceptual capabilities.

[0103] Further, refer to Figure 6 , Figure 6 This is a flowchart illustrating the principle of cross-camera tracking provided in the embodiments of this application.

[0104] In this embodiment, cross-camera tracking involves improvements in three aspects: hardware devices, tracking strategies, and algorithm principles. Regarding hardware devices, this refers to the work of the camera end, including pre-planning possible paths and encoding the devices according to the planned paths, assigning IDs for subsequent camera selection during tracking. Regarding the tracking strategy, it is responsible for selecting the camera during cross-camera tracking, considering all possible actions the target might take, such as stopping or turning back. Regarding the algorithm principles, it mainly optimizes the target detection algorithm, including adjusting the weights of face recognition and appearance detection, adding an attention mechanism, and a memory storage strategy. The memory storage strategy is primary, while weight adjustment and the attention mechanism are secondary. Based on the combined effect of these three aspects, accurate capture of cross-camera targets is achieved.

[0105] Specifically, path information is pre-planned based on the application scenario of the target area, the camera's location information, and its field of view. When the target user enters the camera's field of view, the target user's location information is detected. Simultaneously, feature extraction is performed on the target user image captured by the camera to obtain the first image feature. Based on the first image feature and the second image feature of the template image, the credibility of the target user image is determined. When the credibility is greater than a set value, the target user image is stored in a memory sequence. Further, the target user's movement direction is determined based on the location information, and the next camera corresponding to that direction is determined based on the movement direction and path information. The tracking algorithm is then activated, and that camera is used to track the target user. If the target user moves out of the current camera, the location of the next camera is determined, and the next camera is used to continue tracking the target user until tracking is complete.

[0106] This application embodiment constructs a route network that conforms to actual movement conditions through a path planning strategy. Considering the possibility that pedestrians cannot pass between two adjacent cameras, this application embodiment excludes some impractical routes, effectively avoiding resource waste compared to the existing technology that relies on networks composed of the distance between cameras. Furthermore, at the algorithm level, this application embodiment incorporates an attention module and a memory storage strategy to achieve autonomous perception capabilities for the target detection algorithm. It reduces the weight of face detection, which has a smaller impact, and increases the weight of appearance feature detection. However, even in cases of similar appearances, face detection can still distinguish the target person from interference, thus improving the accuracy of cross-camera tracking.

[0107] The cross-camera tracking device provided in the embodiments of this application is described below. The cross-camera tracking device described below can be referred to in correspondence with the cross-camera tracking method described above.

[0108] refer to Figure 9 , Figure 9 This is a schematic diagram of the cross-camera tracking device provided in the embodiments of this application. The cross-camera tracking device of this application includes a first determining module 901, a second determining module 902, a third determining module 903, a storage module 904, and a tracking module 905.

[0109] The first determining module 901 is used to determine the path information between each camera;

[0110] The second determining module 902 is used to determine the first camera to be detected based on the first movement direction of the target user and the path information, and to acquire the target user image based on the first camera to be detected;

[0111] The third determining module 903 is used to determine the credibility of the target user image based on the first image features of the target user image and the second image features of the template image;

[0112] The storage module 904 is used to store the target user image into a memory sequence according to the credibility.

[0113] The tracking module 905 is used to determine the second camera to be detected based on the memory sequence, the path information and the second movement direction of the target user, and to track the target user based on the second camera to be detected.

[0114] This application embodiment determines the path information between cameras; determines the first camera to be detected based on the first movement direction and path information of the target user, and acquires the target user image based on the first camera to be detected; determines the credibility of the target user image based on the first image features of the target user image and the second image features of the template image; stores the target user image in a memory sequence based on the credibility; determines the second camera to be detected based on the memory sequence, path information, and the second movement direction of the target user, and tracks the target user based on the second camera to be detected. In this way, the next detection camera is called based on the movement direction and path information of the target user, which improves the autonomous capability of cross-camera tracking. At the same time, the addition of a memory storage strategy to the target detection algorithm improves the accuracy of cross-camera tracking.

[0115] In one embodiment, the second determining module 903 is specifically used for:

[0116] Determine the third image feature of the initial template image, wherein the template image includes the initial template image;

[0117] Determine the maximum response value between the first image feature and the third image feature;

[0118] The credibility of the target user image is determined based on the maximum response value, the second image feature, and the number of template images.

[0119] In one embodiment, the method further includes:

[0120] If the number of template images in the memory sequence is greater than or equal to a set number, then the memory degree value of each template image is determined according to preset parameters and the credibility of each template image.

[0121] The template image to be deleted is determined based on the memory level value;

[0122] Delete the template image to be deleted.

[0123] In one embodiment, the method further includes:

[0124] If the credibility of the template image is greater than the first preset value, then the memory level of the template image is determined based on the initial memory level value of the template image and the first reinforcement parameter.

[0125] If the credibility of the template image is less than or equal to the first preset value, then the memory level of the template image is determined based on the initial memory level value, the first reinforcement parameter, the temporal value, and the second reinforcement parameter.

[0126] The preset parameters include the initial memory level value, the first reinforcement parameter, the time sequence value, and the second reinforcement parameter.

[0127] In one embodiment, the first determining module 901 is specifically used for:

[0128] Determine the application scenario of the target area, as well as the location information and field of view of each camera in the target area;

[0129] Based on the location information of each camera, its field of view, and the application scenario, the path information between the cameras is determined.

[0130] In one embodiment, the first determining module 901 is specifically used for:

[0131] The identification information of each camera is determined based on the path information;

[0132] Associate the identification information and the path information.

[0133] In one embodiment, the method further includes:

[0134] Determine the first weight value for face recognition and the second weight value for appearance detection;

[0135] The target detection algorithm is determined based on the first weight value, the second weight value, the memory sequence, the channel attention module, the spatial attention module, and the convolutional network management model.

[0136] The target user is detected according to the target detection algorithm.

[0137] Figure 10 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 10As shown, the electronic device may include: a processor 1010, a communication interface 1020, a memory 1030, and a communication bus 1040, wherein the processor 1010, the communication interface 1020, and the memory 1030 communicate with each other via the communication bus 1040. The processor 1010 can call a computer program in the memory 1030 to execute the steps of the cross-camera tracking method, such as including:

[0138] Determine the path information between each camera;

[0139] The first camera to be detected is determined based on the first direction of movement of the target user and the path information, and the image of the target user is acquired based on the first camera to be detected.

[0140] The credibility of the target user image is determined based on the first image features of the target user image and the second image features of the template image.

[0141] The target user image is stored in a memory sequence based on the credibility level.

[0142] The second camera to be detected is determined based on the memory sequence, the path information, and the second movement direction of the target user, and the target user is tracked based on the second camera to be detected.

[0143] Furthermore, the logical instructions in the aforementioned memory 1030 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0144] On the other hand, embodiments of this application also provide a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the steps of the cross-camera tracking method provided in the above embodiments, such as including:

[0145] Determine the path information between each camera;

[0146] The first camera to be detected is determined based on the first direction of movement of the target user and the path information, and the image of the target user is acquired based on the first camera to be detected.

[0147] The credibility of the target user image is determined based on the first image features of the target user image and the second image features of the template image.

[0148] The target user image is stored in a memory sequence based on the credibility level.

[0149] The second camera to be detected is determined based on the memory sequence, the path information, and the second movement direction of the target user, and the target user is tracked based on the second camera to be detected.

[0150] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0151] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A cross-camera tracking method, characterized in that, include: Determine the path information between each camera; The first camera to be detected is determined based on the target user's first direction of movement and the path information, and the target user's image is acquired based on the first camera to be detected; the first camera to be detected refers to the next camera to track the target user. The credibility of the target user image is determined based on the first image features of the target user image and the second image features of the template image. The target user image is stored in a memory sequence based on the credibility level. The second camera to be detected is determined based on the memory sequence, the path information, and the second movement direction of the target user, and the target user is tracked based on the second camera to be detected. The step of determining the credibility of the target user image based on the first image features of the target user image and the second image features of the template image includes: Determine the third image feature of the initial template image, wherein the template image includes the initial template image; Determine the maximum response value between the first image feature and the third image feature; The credibility of the target user image is determined based on the maximum response value, the second image feature, and the number of template images. The formula for calculating the credibility is: in, This represents the first image feature; This represents the third image feature; The maximum response value is defined as the peak value of the first image feature and the third image feature. This indicates the memory length, i.e., the number of template images. This represents the second image feature; The numerator represents the average value of the target response in the memory sequence; the ratio of the numerator to the denominator represents the confidence level of the target user image. When the ratio is greater than 1, the confidence level is determined by the numerator. This indicates that the target user image prediction result is reliable, and the target user image is stored in a memory sequence.

2. The cross-camera tracking method according to claim 1, characterized in that, The method further includes: If the number of template images in the memory sequence is greater than or equal to a set number, then the memory degree value of each template image is determined according to preset parameters and the credibility of each template image. The template image to be deleted is determined based on the memory level value; Delete the template image to be deleted.

3. The cross-camera tracking method according to claim 2, characterized in that, The step of determining the memorability value of each template image based on preset parameters and the credibility of each template image includes: If the credibility of the template image is greater than the first preset value, then the memory level of the template image is determined based on the initial memory level value of the template image and the first reinforcement parameter. If the credibility of the template image is less than or equal to the first preset value, then the memory level of the template image is determined based on the initial memory level value, the first reinforcement parameter, the temporal value, and the second reinforcement parameter. The preset parameters include the initial memory level value, the first reinforcement parameter, the time sequence value, and the second reinforcement parameter.

4. The cross-camera tracking method according to claim 1, characterized in that, The determination of path information between each camera includes: Determine the application scenario of the target area, as well as the location information and field of view of each camera in the target area; Based on the location information of each camera, its field of view, and the application scenario, the path information between the cameras is determined.

5. The cross-camera tracking method according to claim 1, characterized in that, After determining the path information between each camera, the process includes: The identification information of each camera is determined based on the path information; Associate the identification information and the path information.

6. The cross-camera tracking method according to claim 1, characterized in that, The method further includes: Determine the first weight value for face recognition and the second weight value for appearance detection; The target detection algorithm is determined based on the first weight value, the second weight value, the memory sequence, the channel attention module, the spatial attention module, and the convolutional network management model. The target user is detected according to the target detection algorithm.

7. A cross-mirror tracking device, characterized in that, include: The first determining module is used to determine the path information between each camera; The second determining module is used to determine the first camera to be detected based on the first movement direction of the target user and the path information, and to acquire the target user image based on the first camera to be detected; the first camera to be detected refers to the next camera to track the target user; The third determining module is used to determine the credibility of the target user image based on the first image features of the target user image and the second image features of the template image; A storage module is used to store the target user image into a memory sequence according to the credibility level; The tracking module is used to determine the second camera to be detected based on the memory sequence, the path information, and the second movement direction of the target user, and to track the target user based on the second camera to be detected. The step of determining the credibility of the target user image based on the first image features of the target user image and the second image features of the template image includes: Determine the third image feature of the initial template image, wherein the template image includes the initial template image; Determine the maximum response value between the first image feature and the third image feature; The credibility of the target user image is determined based on the maximum response value, the second image feature, and the number of template images. The formula for calculating the credibility is: in, This represents the first image feature; This represents the third image feature; The maximum response value is defined as the peak value of the first image feature and the third image feature. This indicates the memory length, i.e., the number of template images. This represents the second image feature; The numerator represents the average value of the target response in the memory sequence; the ratio of the numerator to the denominator represents the confidence level of the target user image. When the ratio is greater than 1, the confidence level is determined by the numerator. This indicates that the target user image prediction result is reliable, and the target user image is stored in a memory sequence.

8. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the cross-camera tracking method according to any one of claims 1 to 6.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the cross-camera tracking method according to any one of claims 1 to 6.

Images