A multi-target merging method based on effective cut points in a cross-camera scene

Abstract

The application discloses a multi-target merging method, medium and equipment based on an effective cut point in a cross-camera scene, and relates to the technical field of image analysis. Through hierarchical clustering and merging, particularly similar targets under different cameras are first merged into a cluster; threshold constraint is performed on the feature similarity between clusters, the clusters are merged into an undirected graph, the undirected graph is segmented through an effective cut point by using a graph analysis method, and the mismerging phenomenon caused by jump noise is reduced, so that the accuracy of multi-target merging in the cross-camera scene is improved.

Description

[0001] Cross-references to related applications

[0002] This application is a divisional application of Chinese patent application No. 2023112418830, filed on September 25, 2023, entitled "A method, medium and device for merging multiple targets in a cross-camera scene". Technical Field

[0003] This invention relates to the field of image analysis technology, and in particular to a method, medium, and device for merging multiple targets in cross-camera scenes. Background Technology

[0004] In the field of video image analysis, multi-target merging across cameras is a crucial algorithmic area. Especially in multi-target tracking within video analysis, it's necessary to categorize and merge identical targets from different cameras simultaneously to obtain different angles and poses of the same target for subsequent logical analysis and processing. Factors such as target angle, shape, pose, and illumination levels from different cameras can lead to significant differences in targets. Therefore, effectively analyzing logical relationships based on target position and features, and performing efficient target merging, is key to the algorithm's success.

[0005] Currently, when merging targets across cameras, multiple hierarchical clustering is used to constrain the feature distance and spatiotemporal constraints between clusters using different threshold conditions, thereby achieving the merging of the same targets and the differentiation of different targets. However, the threshold conditions have a "one-size-fits-all" phenomenon, and some abrupt noise images may cause many different targets to be merged into one category, resulting in frequent false merging phenomena.

[0006] Therefore, how to solve the erroneous merging phenomenon caused by jump noise is the key problem that this technical solution aims to address. Summary of the Invention

[0007] In order to solve at least one of the technical problems mentioned in the background art, the present invention aims to provide a method, medium and device for multi-target merging in cross-camera scenes, which can reduce the false merging phenomenon caused by jump noise and improve the accuracy of cluster merging.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] In a first aspect, embodiments of the present invention provide a method for merging multiple targets in a cross-camera scene, including:

[0010] S101, acquire multiple targets from images captured by multiple cameras within a preset time period;

[0011] S102, clustering is performed based on the feature similarity between the targets to obtain multiple clusters; each cluster includes a cluster center;

[0012] S103, For any two cluster centers, calculate the feature similarity to obtain a similarity matrix; each row / column of the similarity matrix corresponds to a cluster center, and the value of the element in the i-th row and j-th column of the similarity matrix represents the feature similarity between the i-th cluster center and the j-th cluster center;

[0013] S104. Based on the values ​​of the elements in the similarity matrix, construct one or more undirected graphs. The undirected graphs include several nodes and edges. The nodes of the undirected graphs correspond one-to-one with the cluster centers of the clusters. When the value of an element is greater than a first preset threshold, there is an edge between the two nodes corresponding to the element. The sum of the number of nodes in all undirected graphs is equal to the number of clusters.

[0014] S105, For each undirected graph, traverse it separately, calculate the degree of all cut vertices in the undirected graph, and filter out cut vertices with a degree greater than n, where n is greater than or equal to 2;

[0015] S106, based on the feature similarity of the cut edges corresponding to the selected cut points, select cut points whose feature similarity to the cut edges is less than the second preset threshold as valid cut points.

[0016] S107, Based on the effective cut points, the undirected graph is divided into multiple subgraphs, and the nodes corresponding to each subgraph are used as the merging result.

[0017] Furthermore, following S104, the following is also included:

[0018] For each undirected graph, determine whether the number of nodes in the undirected graph is greater than a third preset threshold. If it is greater than the third preset threshold, execute steps S105 to S107; if it is not greater than the third preset threshold, use all nodes corresponding to the undirected graph as the merging result.

[0019] Furthermore, S107 includes:

[0020] Traverse the undirected graph without traversing the effective cut vertices until every node in the undirected graph has been traversed, resulting in multiple subgraphs.

[0021] Based on the degree of the effective cut point in each subgraph, the first association degree between the effective cut point and each subgraph is obtained;

[0022] When there is only one maximum value for the first correlation degree, delete the effective cut point in the subgraph where the first correlation degree is not the maximum value;

[0023] When there are p maximum values ​​for the first correlation degree, delete the effective cut vertices in any p-1 subgraphs with the largest first correlation degree, and delete the effective cut vertices in subgraphs where the first correlation degree is not the maximum value, wherein the... ;

[0024] The nodes corresponding to each subgraph are used as the merged result.

[0025] Furthermore, when there are p maximum values ​​for the first correlation degree, the sum of the edge values ​​of the effective cut point in the corresponding p subgraphs is calculated to obtain the second correlation degree between the effective cut point and the p subgraphs, wherein the edge value is the feature similarity corresponding to the edge;

[0026] If there is only one maximum value for the second correlation degree, delete the valid cut points in the subgraph where the second correlation degree is not the maximum value;

[0027] If there are q maximum values ​​for the second correlation degree, delete the effective cut vertices in any q-1 subgraphs with the largest second correlation degree, and delete the effective cut vertices in subgraphs where the second correlation degree is not the maximum value, wherein... .

[0028] Furthermore, the formula for calculating the feature similarity is as follows:

[0029] ;

[0030] in, This represents the feature similarity between target X and target Y. The features of target X and Y are both K-dimensional vectors, denoted as: , .

[0031] Furthermore, the feature of the cluster center is the average of the features of all targets in the cluster to which the cluster center belongs, or the feature of the target closest to the cluster center.

[0032] Furthermore, in S102, the clustering algorithm adopts a hierarchical clustering algorithm.

[0033] Furthermore, in S105, the undirected graph is traversed using a depth-first traversal.

[0034] Secondly, embodiments of the present invention also provide a computer storage medium storing a computer program that, when executed by a processor, implements the multi-target merging method in any cross-camera scene as described above.

[0035] Thirdly, embodiments of the present invention also provide a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the multi-target merging method in any cross-camera scene as described above.

[0036] Compared with the prior art, the beneficial effects of the present invention are: by hierarchical clustering and merging, targets that are particularly similar under different cameras are first merged into one cluster; by using the feature similarity between clusters for threshold constraints, they are merged into an undirected graph; by using graph analysis methods, the undirected graph is segmented by effective cut points, reducing the erroneous merging phenomenon caused by jump noise, thereby improving the accuracy of multi-target merging in cross-camera scenes. Attached Figure Description

[0037] Figure 1 A flowchart illustrating a multi-target merging method in a cross-camera scene provided by an embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram of the merging result without removing transition noise, provided by an embodiment of the present invention.

[0039] Figure 3 A schematic diagram of a target image corresponding to a first sub-image provided in an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram of a target image corresponding to a second sub-image provided in an embodiment of the present invention. Detailed Implementation

[0041] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] To address the aforementioned problems in the prior art, this invention discloses a method, medium, and device for multi-target merging in cross-camera scenarios. It should be noted that the executing entity in this invention can be a computer or any electronic device with data processing capabilities. For ease of description, electronic devices will be used as the executing entity in the following detailed descriptions.

[0043] Example 1:

[0044] like Figure 1 As shown, Figure 1 This is a flowchart of a multi-target merging method in a cross-camera scene provided by an embodiment of the present invention.

[0045] S101, acquire multiple targets from images captured by multiple cameras within a preset time period;

[0046] Get the set of targets appearing under multiple cameras within a preset time period: detections ,..., },in, This indicates information such as the target's position (x, y) and its width and height dimensions (width, height) in the video image. , indicating that there are a total of m targets. The preset duration is set according to the actual situation, such as ten minutes, thirty minutes, etc., and this embodiment of the invention does not limit this.

[0047] The target can be any object to be observed, such as pedestrians or animals. In this embodiment of the invention, the target refers to the pedestrian captured by the camera. The targets in the target set may belong to the same person or different people.

[0048] S102, clustering is performed based on the feature similarity between targets to obtain multiple clusters; each cluster includes a cluster center;

[0049] Each target image has corresponding features, which can be deep learning-based person re-identification embedding features, or features such as color histograms. The embedding features describing the target images are denoted as Feas: { , ,..., },in, It is a 1*K dimensional vector.

[0050] Clustering is performed based on the feature similarity between targets, ensuring that targets with similar features are grouped into one cluster; different targets are classified into different clusters, resulting in the clustering results: clusterers: { , ,..., There are n clusters in total (n <= m).

[0051] Record the characteristics of the target content , They are respectively: , Corresponding , The formula for calculating the image content similarity between two images is:

[0052] ;

[0053] Preferably, the above clustering algorithm adopts a hierarchical clustering algorithm.

[0054] The features corresponding to the cluster centers are denoted as:

[0055] { , ,..., },in, It is a 1*K dimensional vector representing the cluster centers. Its characteristics.

[0056] Cluster center characteristics The method for obtaining the value can be either the selected target feature closest to the cluster center, or all target features belonging to the same cluster. , ,..., The average of the same dimensions of} is calculated using the following formula:

[0057] ;

[0058] Where C refers to the number of targets in the same cluster.

[0059] For all cluster centers in the clusters, calculate the pairwise feature distance to obtain an n*n similarity matrix M, where, ,express , Feature similarity. Simultaneously, for n clusters of data, the merging flag... Set to true. .

[0060] S104. Based on the values ​​of the elements in the similarity matrix, construct one or more undirected graphs. The undirected graphs include several nodes and edges. The nodes of the undirected graphs correspond one-to-one with the cluster centers of the clusters. When the value of an element is greater than a first preset threshold, there is an edge between the two nodes corresponding to the element. The sum of the number of nodes in all undirected graphs is equal to the number of clusters.

[0061] By each Analyze the values ​​and construct one or more undirected graphs G(Vertexes,E) consisting of all cluster centers belonging to the same objective. Where Vertexes: {Vertex...} m Vertex n Vertex k} represents nodes belonging to the same target, and {m,n,..k} is the cluster index set belonging to that target. E:{e m,n ,e k,n ,...} represents the set of valid edge connections where these nodes exist, e m,n Represents the node {Vertex m Vertex nThe value of an edge for a pair of similar targets with the same objective, and the value e of the corresponding edge. n,m =e m,n The process of forming an undirected graph G by finding all clusters belonging to the same goal is as follows:

[0062] from Starting at (i=0), if the current merge flag... If it is false, it means that the cluster i If the cluster has already been traversed and analyzed, proceed to the next cluster traversal; if not found... When a cluster is true, it means that all clusters in the current clusters have been processed.

[0063] like A value of true indicates a cluster. i No relationship analysis with other clusters was performed, and the cluster was considered... i As a node Vertex i Adding nodes to the vertex set; traversing and clustering i All adjacent clusters that meet the conditions are appended to the Vertexes set as new nodes, as follows:

[0064] If mergedFlag j If true, search for cluster. i and Feature similarity ,like If the value is greater than the first preset threshold thresh, it means , The corresponding nodes , They are a set of similar pairs, therefore Vertex j Add to the node set Vertexes; and in the node , Connecting edges between them The value assigned is ;at the same time, Setting it to false is preferable, with a thresh value of 0.45; it should be noted that... , The values ​​are equal, therefore It was also assigned the value No need for repeated calculations.

[0065] Traverse the cluster as described above iMethods for adjacent nodes, recursive traversal, and clustering j Other nodes with a similarity distance greater than thresh (s≠i), if the constraint conditions are met, cluster s It is added to Vertexes as a new node.

[0066] If the feature similarity value between all nodes in Vertexes and the remaining nodes in clusters is not greater than thresh, it indicates that the current cluster is in a state of equilibrium. i The graph G(Vertexes,E) starting from the given point is now constructed.

[0067] If there are still unprocessed clusters, arbitrarily select one of the remaining clusters as the starting point to begin traversal, and continue constructing the next undirected graph in the same manner as above, until all clusters are processed. i , All of them are incorporated into the undirected graph; thus, one or more undirected graphs are obtained.

[0068] S105. For each undirected graph, traverse it separately, calculate the degree of all cut vertices in the undirected graph, and filter out cut vertices with a degree greater than n, where n is greater than or equal to 2.

[0069] Preferably, the traversal algorithm adopts the depth-first search (DFS) algorithm to find the cut vertices and degrees of nodes in G. In this embodiment, n is set to 2. A node degree less than 2 indicates that the relationship between the two nodes is relatively simple, which cannot effectively divide the cluster and is likely to cause more clusters to split. Therefore, in order to effectively improve the segmentation effect of the algorithm, cut vertices with a degree greater than 2 are selected.

[0070] S106, based on the feature similarity of the cut edges corresponding to the selected cut points, select cut points whose feature similarity to the cut edges is less than the second preset threshold as valid cut points.

[0071] For each found cut vertex, iterate through the cut edges surrounding it. If a cut edge exists, analyze the feature similarity between the cut edge and the cut vertex. If any cut edge is less than the second preset threshold EdgeThresh, it is considered a valid cut vertex and added to the cut vertex set cuts:{cut0,...,cut r}, r>0.

[0072] S107, Based on the effective cut points, the undirected graph is divided into multiple subgraphs, and the nodes corresponding to each subgraph are used as the merging result;

[0073] G is divided into multiple subgraphs based on effective cut vertices. , ,...}, all nodes belonging to the same subgraph belong to the same goal, and the set of nodes contained in each subgraph is the new merging result.

[0074] The above S107 can be further subdivided into the following steps:

[0075] Traverse the undirected graph without visiting valid cut vertices during the traversal until every node in the undirected graph has been visited, resulting in multiple subgraphs.

[0076] Based on the degree of the effective cut vertex in each subgraph, the first association degree between the effective cut vertex and each subgraph is obtained;

[0077] When there is only one maximum value for the first degree of association, delete the effective cut point in the subgraph where the first degree of association is not the maximum value.

[0078] When there are p maximum values ​​for the first degree of association, delete the valid cut vertices in any p-1 subgraphs with the largest first degree of association, and delete the valid cut vertices in the subgraphs where the first degree of association is not the maximum value. ;

[0079] The nodes corresponding to each subgraph are used as the merged result.

[0080] On an undirected graph G, perform a depth-first search (DFS) traversal to find the valid cut vertices in the cut vertex set cuts. i As boundary points of the subgraph, do not cut at valid cut points. i The adjacent nodes are recursively traversed until all nodes in G have been traversed, resulting in multiple subgraphs.

[0081] Based on the first degree of association between the effective cut vertex and multiple subgraphs, the effective cut vertex is retained in the subgraph with the highest first degree of association with the effective cut vertex. At the same time, the nodes corresponding to the effective cut vertex are deleted in other subgraphs. Here, the first degree of association is the degree of the effective cut vertex in the subgraph.

[0082] When performing a Depth-First Search (DFS) algorithm on a graph, a problem arises where a valid cut vertex is repeatedly traversed as a boundary node in multiple subgraphs, resulting in its being assigned to multiple subgraphs. This invention addresses this by calculating and analyzing the degree of the subgraph containing the valid cut vertex during the DFS traversal, using this degree as the first association degree. Ultimately, it determines which subgraph the valid cut vertex belongs to and removes it from other subgraphs.

[0083] For example, graph G(V,E) has a valid cut vertex A. When performing a depth-first search (DFS) traversal on graph G, encountering the valid cut vertex A indicates reaching the subgraph boundary node. Therefore, the recursive traversal of A's adjacent nodes is no longer performed, resulting in a subgraph containing A. While traversing, count the subgraph. Find the degree of node A. Then continue the above traversal process for the remaining nodes of graph G. When encountering A, stop recursively traversing the adjacent nodes of A until all the remaining nodes of G have been traversed, thus obtaining the subgraph. During the traversal, the subgraphs are also statistically analyzed. The degree of A in the middle; ultimately, two subgraphs are obtained. , A, as an effective cut point, will appear in both subgraphs. Record them separately. The degree of node A in the middle is ;like Determine whether A matches the diagram. If the first correlation is high, then it is necessary to start from... Delete A; if A and Figure The first correlation is high, so it needs to be considered from... Delete A; if If we retain a valid cut vertex A in any one of the subgraphs, we delete A from the other subgraphs; thus ensuring that the two subgraphs... It is a complete partition. In scenarios with multiple subgraphs, the same cut vertex degree voting method is used to determine which subgraph the cut vertex should be optimally assigned to.

[0084] Suppose we get 3 subgraphs, which means that the undirected graphs originally classified as the same target have been mistakenly merged. In fact, there are 3 different targets in the undirected graph. By segmenting through effective cut points, the mistaken merging caused by the jump noise image is reduced.

[0085] like Figure 2 As shown, Figure 2 This is a schematic diagram of the merging result without removing transition noise, provided by an embodiment of the present invention. Figure 2 There are 14 target images in the image, all classified as the same target. However, there are two different pedestrians in the image. Due to the presence of a noisy image, clusters that do not belong to the same target are merged together, which is a case of mismerging.

[0086] Utilizing effective cut points Figure 2 After partitioning the corresponding undirected graph, two subgraphs are obtained. Figure 3 A schematic diagram of a target image corresponding to a first sub-image provided in an embodiment of the present invention; Figure 3 There are a total of 6 pedestrian images, all corresponding to the same pedestrian; Figure 4 This is a schematic diagram of a target image corresponding to a second sub-image provided in an embodiment of the present invention. Figure 4 There are a total of 8 pedestrian images, all corresponding to the same pedestrian; obviously, this technical solution can be used to... Figure 2 Different pedestrians were effectively separated.

[0087] By hierarchical clustering and merging, highly similar targets from different cameras are first merged into one cluster. The feature similarity between clusters is used to apply a threshold constraint, and the targets are merged into an undirected graph. Graph analysis is then used to segment the undirected graph through effective cut vertices, reducing false merging caused by jump noise and thus improving the accuracy of multi-target merging in cross-camera scenarios.

[0088] Example 2:

[0089] The difference from Embodiment 1 is that, in step S107, when there are p maximum values ​​for the first correlation degree,

[0090] Calculate the sum of edge values ​​of the effective cut points in the corresponding p subgraphs to obtain the second correlation degree between the effective cut points and the p subgraphs, where the edge value is the feature similarity of the edge.

[0091] If there is only one maximum value for the second degree of association, delete the valid cut points in the subgraph where the second degree of association is not the maximum value.

[0092] If there are q maximum values ​​for the second degree of association, delete the valid cut vertices in any q-1 subgraphs with the maximum second degree of association, and delete the valid cut vertices in the subgraphs where the second degree of association is not the maximum value. .

[0093] Using the example from Embodiment 1 above as an illustration, when At that time, calculate the sum of the boundary values ​​of the effective cut point A in G1 and G2 respectively, and record them as follows: The sum of the edge values ​​of node A is like Determine whether A matches the diagram. If the second correlation is high, then it is necessary to start from... Delete A; if A and Figure The second correlation is high, requiring... Delete A; if If we retain a valid cut vertex A in any one of the subgraphs, we delete A from the other subgraphs; thus ensuring that the two subgraphs... It is completely divided.

[0094] Although the effective cut vertices have equal degrees in the subgraphs, the feature similarity between the effective cut vertices and the connected nodes in the subgraphs may differ, meaning the sum of edge values ​​may vary. Higher feature similarity indicates that the cluster represented by the effective cut vertex is closer to the subgraph. Therefore, the effective cut vertex is assigned to the subgraph with the second highest correlation. This further improves the accuracy of cluster merging.

[0095] Example 3:

[0096] The difference from Embodiment 1 is that after constructing the undirected graph in step S104, it is determined whether the number of nodes in the undirected graph is greater than the third preset threshold ThreshCount. In this embodiment, ThreshCount is set to 5. If the number of nodes is greater than 5, then S105~S107 are executed; if it is not greater, then all nodes in the undirected graph are directly used as the merging result.

[0097] When the number of nodes is small, the probability of erroneous merging is very low, so there is no need to execute S105~S107, which can reduce the consumption of network resources and improve the efficiency of image analysis.

[0098] Example 4:

[0099] Corresponding to the embodiments of the multi-target merging method in cross-camera scenarios, the embodiments of the present invention also provide a computer storage medium storing a computer program that, when executed by a processor, implements any of the above-described multi-target merging methods in cross-camera scenarios.

[0100] Example 5:

[0101] Corresponding to the embodiment of the multi-target merging method in cross-camera scenarios, this embodiment of the invention also provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-described multi-target merging methods in cross-camera scenarios.

[0102] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of equivalents of the claims be included within the present invention.

Claims

1. A multi-target merging method based on effective cut points in cross-camera scenes, characterized in that, Includes the following steps: S101, acquire multiple targets from images captured by multiple cameras within a preset time period; S102, clustering is performed based on the feature similarity between the targets to obtain multiple clusters; each cluster includes a cluster center; S103, For any two cluster centers, calculate the feature similarity to obtain the similarity matrix; Each row / column of the similarity matrix corresponds to a cluster center, and the value of the element in the i-th row and j-th column of the similarity matrix represents the feature similarity between the i-th cluster center and the j-th cluster center; for n clusters, the merge flag is set to true; S104. Based on the values ​​of the elements in the similarity matrix, construct one or more undirected graphs. The undirected graphs include several nodes and edges. The nodes of the undirected graphs correspond one-to-one with the cluster centers of the clusters. When the value of an element is greater than a first preset threshold, there is an edge between the two nodes corresponding to the element. The sum of the number of nodes in all undirected graphs is equal to the number of clusters. The process of constructing an undirected graph is as follows: From the cluster center Starting with i=0, if the current merge flag is false, it indicates that the cluster center is... If the cluster centers have already been analyzed, proceed to the next cluster center traversal; if not found... When a cluster center is true, it means that all clusters in the current clustering result have been processed; like A value of true indicates that the cluster center is... No relationship analysis with other cluster centers was performed, and this cluster center was not included in the analysis. As a node Vertex i Add nodes to the vertexes set; traverse the nodes with respect to the cluster centers. All adjacent clustering result members, and the cluster centers that meet the conditions are added as new nodes to the node set Vertexes; S105, For each undirected graph, traverse it separately, calculate the degree of all cut vertices in the undirected graph, and filter out cut vertices with a degree greater than n, where n is greater than or equal to 2; S106, based on the feature similarity of the cut edges corresponding to the selected cut points, select cut points whose feature similarity to the cut edges is less than the second preset threshold as valid cut points. S107, Based on the effective cut points, the undirected graph is divided into multiple subgraphs, and the nodes corresponding to each subgraph are used as the merging result; S107 includes: Traverse the undirected graph without traversing the effective cut vertices until every node in the undirected graph has been traversed, resulting in multiple subgraphs. Based on the degree of the effective cut point in each subgraph, the first association degree between the effective cut point and each subgraph is obtained; When there is only one maximum value for the first correlation degree, delete the effective cut point in the subgraph where the first correlation degree is not the maximum value; When there are p maximum values ​​for the first correlation degree, delete the effective cut vertices in any p-1 subgraphs with the largest first correlation degree, and delete the effective cut vertices in subgraphs where the first correlation degree is not the maximum value, wherein the... ; The nodes corresponding to each subgraph are used as the merged result.

2. The multi-target merging method in cross-camera scenarios based on effective cut points according to claim 1, characterized in that, Following S104, the following is also included: For each undirected graph, determine whether the number of nodes in the undirected graph is greater than a third preset threshold. If it is greater than the third preset threshold, execute steps S105 to S107; if it is not greater than the third preset threshold, use all nodes corresponding to the undirected graph as the merging result.

3. The multi-target merging method in cross-camera scenes based on effective cut points according to claim 1, characterized in that, When there are p maximum values ​​of the first correlation degree, the sum of the edge values ​​of the effective cut point in the corresponding p subgraphs is calculated to obtain the second correlation degree between the effective cut point and the p subgraphs, wherein the edge value is the feature similarity corresponding to the edge; If there is only one maximum value for the second correlation degree, delete the valid cut points in the subgraph where the second correlation degree is not the maximum value; If there are q maximum values ​​for the second correlation degree, delete the effective cut vertices in any q-1 subgraphs with the largest second correlation degree, and delete the effective cut vertices in subgraphs where the second correlation degree is not the maximum value, wherein... .

4. The multi-target merging method in cross-camera scenes based on effective cut points according to claim 1, characterized in that, The formula for calculating the feature similarity is: ； in, This represents the feature similarity between target X and target Y. The features of target X and Y are both K-dimensional vectors, denoted as: , .

5. The multi-target merging method in cross-camera scenes based on effective cut points according to claim 1, characterized in that, The feature of the cluster center is the average of the features of all targets in the cluster to which the cluster center belongs, or the feature of the target that is closest to the cluster center.

6. The multi-target merging method in cross-camera scenes based on effective cut points according to claim 1, characterized in that, In step S102, the clustering algorithm used is a hierarchical clustering algorithm.

7. The multi-target merging method in cross-camera scenes based on effective cut points according to claim 1, characterized in that, In S105, the undirected graph is traversed using a depth-first traversal.

8. The multi-target merging method in cross-camera scenes based on effective cut points according to claim 1, characterized in that, In step S106, for the found cut point, the cut edges around the cut point are traversed. If the cut edge exists, the feature similarity of the cut edge corresponding to the cut point is analyzed. If one of the cut edges is less than the second preset threshold, it is determined to be a valid cut point and added to the cut point set.

Images