Cross-camera passenger trajectory tracking method, system and device based on video technology

Abstract

The invention belongs to the technical field of trajectory tracking, and particularly relates to a cross-camera passenger trajectory tracking method, system and device based on a video technology, and the method comprises the steps: enabling each camera to represent a vertex, enabling an edge to represent a connection relation between cameras, determining the direction of the edge, constructing a monitoring camera network topology structure, and carrying out the tracking of a passenger trajectory; according to the method, weighting is carried out after the vertex weight is calculated, and a camera monitoring scene is constructed into a weighted directed graph, so that the time delay of video processing can be reduced, and the processing efficiency of a monitoring system is improved; during path selection, a vertex weight, a historical path weight and a path length are subjected to weighted combination with a proper weight coefficient to select a video stream path with optimal performance and resource utilization rate as a target path, so that the probability that a passenger to be tracked is searched is increased, and the intelligent level of a monitoring system is improved. The influence of human errors and subjective factors on monitoring decisions is reduced, and the problem of resource waste in a traditional monitoring system is avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of trajectory tracking, and in particular relates to a method, system and device for tracking passenger trajectories across cameras based on video technology. Background Art

[0002] With the advancement of technology and the reduction of costs, the number of surveillance cameras has exploded. Traditional surveillance systems have problems of resource waste and low processing efficiency when processing a large number of surveillance camera videos. Most of the current video selection methods are based on static rules or manual settings. When searching for specific events or targets, they generally use traversal search to search all camera videos or topological network search and repeated search. The traversal search includes breadth-first search and depth-first search. Breadth-first search starts from a certain vertex and searches in hierarchical order; depth-first search starts from a certain starting vertex, first visits the vertex, and then recursively visits all unvisited vertices adjacent to the vertex. Tracking passenger trajectories based on the above traditional method of selecting surveillance cameras not only consumes a lot of computing resources, but also leads to low efficiency and poor accuracy in personnel tracking and searching, and cannot fully utilize system resources and cope with dynamic scene changes. Summary of the invention

[0003] In order to solve the problems raised by the background technology, the present invention provides a method, system and device for tracking passenger trajectories across cameras based on video technology.

[0004] The technical solution of the present invention is as follows:

[0005] The present invention provides a method for tracking passenger trajectories across cameras based on video technology, comprising the following steps:

[0006] S1. Build network topology:

[0007] Obtain the historical flow of people in the area to be detected, divide the area to be detected into different density areas according to the density of people flow, calculate the density weight of people flow in each density area, and take the density area with a density weight greater than the preset density weight as a node;

[0008] After obtaining the historical weight data of the cameras in the area to be detected, the cameras are taken as vertices and the connections between the cameras are edges;

[0009] According to the historical traffic and camera weight data, determine the direction of different nodes and build the network topology structure;

[0010] S2. Calculate vertex weights:

[0011] Based on the network topology, the probability of the target passenger appearing at the vertex is calculated;

[0012] The vertex weight is obtained by multiplying the probability of the target passenger appearing at the vertex, the time the target passenger passes the vertex, and the clarity value of the vertex;

[0013] S3. Construct a weighted directed graph:

[0014] According to the vertex weights, after weighting operations are performed on the vertices in the network topology structure, a weighted directed graph is obtained;

[0015] S4. Select path:

[0016] For each vertex on a node, the vertex weight, historical path weight, and path length are weighted and summed to obtain the path value of the current vertex. The path consisting of the vertices with the smallest path value at the same node is taken as the first path. The first paths of all nodes are connected to obtain the target path.

[0017] S5. Tracking the trajectory of the passengers to be tracked:

[0018] Obtain target monitoring data, and track the trajectory of the passenger to be tracked based on the target path;

[0019] If the vertex in the target path does not track the passenger to be tracked, a path consisting of vertices with the next smallest path value at the node where the vertex is located is selected for tracking.

[0020] The step S1 determines the direction of different nodes according to the historical flow of people and the historical weight data of the camera, specifically:

[0021] Calculate the traffic weight of each edge based on the historical traffic;

[0022] According to the camera's historical weight data, calculate the historical weight of each vertex;

[0023] Based on the traffic weight and the historical weight of the vertex, the direction weight of the edge is calculated and the direction of the edge under different nodes is determined.

[0024] After calculating the direction weight of the edge based on the traffic weight and the historical weight of the vertex, the direction of the edge under different nodes is determined, specifically, according to the formula:

[0025] W c =aW a +bW b , calculate the direction weight of the edge;

[0026] Where W c is the direction weight of the edge; a is the coefficient of the traffic weight; W a is the traffic weight; b is the coefficient of the vertex's historical weight; W b is the historical weight of the vertex;

[0027] According to the direction weight of the edge, the direction of the edge under different nodes is determined.

[0028] The step S2 is to calculate the probability of the target passenger appearing at the vertex based on the network topology, specifically, according to the formula:

[0029] Calculate the probability of the target passenger appearing at the vertex;

[0030] Where A is the probability of the target passenger appearing at the vertex; H is the historical number of target passengers; T is the historical number of passengers passing through the vertex; W d It is the crowd density weight of the density zone to which the vertex belongs.

[0031] In step S4, for each vertex on a node, the vertex weight, the historical path weight, and the path length are weighted and summed to obtain the path value of the current vertex, specifically, according to the formula:

[0032] p i =argmin(αW(v)+βH(v)+γd(v)), get the path value of the current vertex;

[0033] In the formula, p i represents the path value of the current vertex, 1≤i≤n; v represents the vertex on the same node; argmin() represents the minimum function; α is the coefficient of vertex weight; W(v) is the vertex weight; β is the coefficient of historical path weight; H(v) is the historical path weight; γ is the coefficient of path length; d(v) is the path length from the starting node to the current vertex.

[0034] The step S5, obtaining target monitoring data and tracking the trajectory of the passenger to be tracked based on the target path, also includes:

[0035] If the current vertex tracks the passenger to be tracked, the path tracking operation from the current vertex to the next vertex will continue. When the vertex of the last node of the target path is reached, the path execution will be terminated and the tracked path will be recorded.

[0036] The step S5, if the vertex in the target path does not track the passenger to be tracked, then a path consisting of vertices with a node having a second smallest path value at the vertex is selected for tracking, further comprising:

[0037] If the passenger to be tracked is not tracked by traversing all vertices of the current node, the trajectory of the passenger to be tracked is recorded as terminating at the vertex of the previous node.

[0038] The present invention also provides a cross-camera passenger trajectory tracking system based on video technology, comprising:

[0039] Network topology structure building module: used to obtain the historical flow of people in the area to be detected, divide the area to be detected into different density areas according to the density of people flow, calculate the density weight of people flow in each density area, and take the density area with a density weight greater than the preset density weight as a node;

[0040] After obtaining the historical weight data of the cameras in the area to be detected, the cameras are taken as vertices and the connections between the cameras are edges;

[0041] According to the historical traffic and camera weight data, determine the direction of different nodes and build the network topology structure;

[0042] Vertex weight calculation module: Based on the network topology, calculate the probability of the target passenger appearing at the vertex;

[0043] The vertex weight is obtained by multiplying the probability of the target passenger appearing at the vertex, the time the target passenger passes the vertex, and the clarity value of the vertex;

[0044] Weighted directed graph construction module: used to obtain a weighted directed graph after performing weighted operations on vertices in the network topology structure according to vertex weights;

[0045] Path selection module: for each vertex on a node, the vertex weight, historical path weight, and path length are weighted and summed to obtain the path value of the current vertex. The path consisting of the vertices with the smallest path value at the same node is taken as the first path. The first paths of all nodes are connected to obtain the target path.

[0046] Passenger trajectory tracking module: used to obtain target monitoring data and track the trajectory of the passenger to be tracked based on the target path;

[0047] If the vertex in the target path does not track the passenger to be tracked, a path consisting of vertices with the next smallest path value at the node where the vertex is located is selected for tracking.

[0048] The present invention also provides a cross-camera passenger trajectory tracking device based on video technology, comprising a processor and a memory, wherein the processor implements the cross-camera passenger trajectory tracking method based on video technology when executing a computer program stored in the memory.

[0049] Beneficial Effects

[0050] The present invention constructs a monitoring camera network topology structure by representing each camera as a vertex, edges representing the connection relationship between cameras, and determining the direction of the edges. The vertex weights are calculated and weighted, and the camera monitoring scene is constructed as a weighted directed graph. This can reduce the time delay of video processing and improve the processing efficiency of the monitoring system.

[0051] When selecting a path, the present invention defines a comprehensive evaluation function, which weights and combines the vertex weight, historical path weight, and path length with an appropriate weight coefficient to select a video stream path with the best performance and resource utilization as the target path, so as to increase the probability of searching for the passengers to be tracked, improve the intelligence level of the monitoring system, reduce manual intervention, and reduce the impact of human errors and subjective factors on monitoring decisions, thereby avoiding the problem of resource waste in traditional monitoring systems and improving resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 Schematic diagram of the process of a cross-camera passenger trajectory tracking method based on video technology in an embodiment.

[0053] Figure 2 It is a weighted directed graph schematic diagram of a cross-camera passenger trajectory tracking method based on video technology in an embodiment. DETAILED DESCRIPTION

[0054] The following examples are intended to illustrate the present invention rather than to further limit the present invention.

[0055] The present invention provides a method for tracking passenger trajectories across cameras based on video technology, such as Figure 1 As shown, the following steps are included:

[0056] S1. Build network topology:

[0057] Obtain the historical pedestrian flow of the area to be detected, divide the area to be detected into different density areas according to the pedestrian density, calculate the pedestrian density weight of each density area, and take the density area with a pedestrian density weight greater than the preset pedestrian density weight as a node.

[0058] In order to ensure the tracking effect and good adaptability to the actual tracking scenario, it is necessary to consider the flow of people information to divide the density area. The division of density areas is not static. Specifically, according to the principle of crowd density, the crowd density and congestion in the area to be detected are considered and divided in combination with historical crowd data and passenger flow information during peak hours. For example, when screening the crowd flow data during the morning peak period of 7:00-9:00, the crowd density at the entrance and ticket gate is usually high. However, if the subway station temporarily adds a fast ticket inspection channel during the morning peak period, the crowd distribution will change. The ticket gate with a high crowd density may become a medium and low crowd density area. In addition, if it is during the off-peak period, the normal crowd density may be high in the transfer area, but if there are major events, the entrance may also become a high-density area.

[0059] Then, for the subway station, the crowded entrance, ticket gate and boarding door area can be regarded as the high-density area, that is, the crowded entrance, ticket gate and boarding door area can be regarded as the nodes, and the relatively idle public area can be regarded as the low-density area.

[0060] After obtaining the historical weight data of the cameras in the area to be detected, the cameras are taken as vertices and the connections between the cameras are edges;

[0061] Based on the historical traffic and camera weight data, the directions under different nodes are determined to build the network topology.

[0062] Preferably, the directions under different nodes are determined according to the historical human flow and the historical weight data of the camera, specifically:

[0063] According to the historical flow of people, the flow weight of each edge is calculated. For example, it can be calculated based on the absolute value or relative value of the flow of people. A larger flow value can indicate the movement trend of the crowd in a certain direction.

[0064] The historical weight of each vertex is calculated based on the historical weight data of the camera. The historical weight data specifically includes the flow data of people passing through the camera (i.e., the number of people passing through the camera per unit time), the average length of time passengers stay under the camera, etc., and the historical weight can represent the importance of the node in the historical path. Different methods can be used to calculate the historical weight, for example, it can be calculated based on the frequency of the node.

[0065] Based on the traffic weight and the historical weight of the vertex, after calculating the direction weight of the edge, the direction of the edge under different nodes is determined. Specifically, according to the formula:

[0066] W c =aW a +bW b , calculate the direction weight of the edge;

[0067] Where W c is the direction weight of the edge; a is the coefficient of the traffic weight; W a is the traffic weight; b is the coefficient of the vertex's historical weight; W b is the historical weight of the vertex; coefficients a and b are used to adjust the influence of the weight of human traffic and the historical weight of the vertex in the comprehensive evaluation. According to the actual situation and needs, these coefficients can be adjusted according to importance and priority. For example, if human traffic is more important in decision-making, the value of a can be increased.

[0068] Finally, the directions of the edges under different nodes are determined according to the direction weights of the edges.

[0069] In the network topology, each camera represents a vertex, and the edge represents the connection relationship between cameras. The direction of the edge is determined, which represents the probability of a passenger transferring from one camera to another. For example, if the edge direction weight calculated when the passenger moves from camera A to camera B is greater, then the edge direction is from A to B.

[0070] S2. Calculate vertex weights:

[0071] Based on the network topology, the probability of the target passenger appearing at the vertex is calculated;

[0072] The vertex weight is obtained by multiplying the probability of the target passenger appearing at the vertex, the time the target passenger passes the vertex, and the clarity value of the vertex.

[0073] Regarding queuing theory, the present invention regards the camera monitoring scene as a queuing system, the surveillance cameras available for analysis as "customers", and the analysis capability of the surveillance camera as "service capability". Given that there are factors such as customer arrival rate, service rate capability, and queuing time in the queuing theory model, for the camera monitoring scene, it is necessary to consider factors such as the camera's processing capability, queue length, and response time, quantify the camera's performance indicators, and then calculate the vertex weight.

[0074] It can be that the vertex weight set is represented by W = {w 11 , w 12 , ...w ij}, (where w ij is the vertex v ij weight), define the vertex weight as w ij ={A ij ·T ij ·B ij}. ij For vertex v ij The probability of finding the target passenger in the surveillance video, T ij For passengers passing through vertex v ij The duration of the surveillance camera represented by B ij For vertex v ij The definition value of the surveillance video.

[0075] Preferably, based on the network topology, the probability of the target passenger appearing at the vertex is calculated, specifically, according to the formula:

[0076] Calculate the probability of the target passenger appearing at the vertex;

[0077] Where A is the probability of the target passenger appearing at the vertex; H is the historical number of target passengers; T is the historical number of passengers passing through the vertex; W d It is the crowd density weight of the density zone to which the vertex belongs.

[0078] S3. Construct a weighted directed graph:

[0079] According to the vertex weights, after weighted operations are performed on the vertices in the network topology structure, a weighted directed graph is obtained.

[0080] That is, according to the calculated vertex weights, a vertex weighting operation is performed on the network topology, and weights are assigned to vertices to reflect the processing capabilities and resource utilization of different cameras.

[0081] The present invention constructs a monitoring camera network topology structure by representing each camera as a vertex, edges representing the connection relationship between cameras, and determining the direction of the edges. The vertex weights are calculated and weighted, and the camera monitoring scene is constructed as a weighted directed graph. This can reduce the time delay of video processing and improve the processing efficiency of the monitoring system.

[0082] S4. Select path:

[0083] For each vertex on a node, the vertex weight, historical path weight, and path length are weighted and summed to obtain the path value of the current vertex. The path consisting of vertices with the smallest path value at the same node is taken as the first path, representing the optimal selected path from the previous node to the current node. The first paths of all nodes are connected to obtain the target path.

[0084] Among them, regarding the historical path weight, by obtaining the historical path data, including the camera sequence that the historical target passenger passed at different time points, the historical path weight of each camera vertex is calculated. The frequency of each camera appearing in the historical path can be counted using a histogram, and the frequency is normalized into a probability value. A higher historical path weight means that the camera is more frequently used in the historical path.

[0085] Regarding vertex weights, in order to conform to the actual tracking situation, the historical path weights and the calculated vertex weights can also be weighted averaged or proportionally adjusted to obtain the adjusted camera vertex weights. In this way, the camera vertices on the historical path will have higher weights, increasing the probability of finding the target passenger near these vertices.

[0086] Regarding the path length, for example, although a camera has a large monitoring attribute weight value, that is, a large vertex weight value, it is very far away from the starting point and the node, which may lead to inefficiency or easy loss of trajectory during the entire passenger trajectory tracking process. Therefore, it is necessary to consider the impact of path length on path selection.

[0087] Therefore, when selecting a path, the vertex weight, historical path weight, and path length are comprehensively considered for searching.

[0088] Preferably, in step S4, for each vertex on a node, the vertex weight, the historical path weight, and the path length are weighted and summed to obtain the path value of the current vertex, specifically, according to the formula:

[0089] p i =argmin(αW(v)+βH(v)+γd(v)), get the path value of the current vertex;

[0090] min(P)={(p1),(p2),...(p n )};

[0091] In the formula, p i represents the path value of the current vertex, 1≤i≤n; v represents the vertex on the same node; argmin() represents the minimum function; α is the coefficient of vertex weight; W(v) is the vertex weight; β is the coefficient of historical path weight; H(v) is the historical path weight; γ is the coefficient of path length; d(v) is the path length from the starting node to the current vertex; min(P) is the target path, which represents the optimal path to track the target passenger's movement trajectory.

[0092] When selecting a path, the present invention defines a comprehensive evaluation function, which weights and combines the vertex weights, historical path weights, and path lengths with appropriate weight coefficients to select a video stream path with the best performance and resource utilization as the target path, so as to increase the probability of searching for the passengers to be tracked, improve the intelligence level of the monitoring system, reduce manual intervention, and reduce the impact of human errors and subjective factors on monitoring decisions.

[0093] S5. Tracking the trajectory of the passengers to be tracked:

[0094] Obtain target monitoring data, and track the trajectory of the passenger to be tracked based on the target path;

[0095] According to the vertex in the target path, if the passenger to be tracked is not tracked, the path consisting of the vertices with the next smallest path value at the node where the vertex is located is selected for tracking.

[0096] Specifically, it can be:

[0097] Obtain the target monitoring data and track the movement trajectory of the passenger to be tracked based on the vertex order in the target path. This means that at the vertex of each node, it is necessary to record the timestamp of the passenger's arrival, trigger the security camera, adjust the monitoring range, etc. These operations can be defined and implemented according to specific scenarios and requirements.

[0098] According to the vertex of the target path, if the passenger to be tracked is tracked at the current vertex, it means that the action of the passenger to be tracked to select the path (ie, the first path) from the previous node to the current node is completed, and the path tracking is successful.

[0099] Based on the target path, continue to perform the path tracking operation from the current vertex to the next vertex. When the vertex of the last node is executed, it is necessary to determine whether the action of the passenger to be tracked is completed. There are two judgment conditions. One is that the passenger to be tracked reaches the end point. For example, from the time the passenger enters the station to the time he takes the subway, the subway door node is considered to be the end point (because the subway car is not within the monitoring range). The other is that the movement trajectory of the passenger to be tracked exceeds the monitoring range, that is, the passenger cannot be found within the monitoring range formed by the network topology structure, which also means that the action of the passenger to be tracked is completed.

[0100] If the action of the passenger to be tracked is completed, the path execution is terminated and the tracked path is recorded.

[0101] If the passenger to be tracked is not tracked at the current vertex, that is, if the optimal vertex (the vertex with the smallest path value) selected at the current node does not find the passenger to be tracked, other possible path selections can be considered based on historical path data and vertex weight information. The suboptimal vertex (the vertex with the second smallest path value) can be selected as the current node selection, and the path search can be continued. This can better utilize historical information and improve processing efficiency and resource utilization.

[0102] If the passenger to be tracked is not tracked by traversing all vertices of the current node, it is considered that the trajectory of the passenger to be tracked ends at the previous node.

[0103] The present invention utilizes the target path to perform accurate tracking and recording to increase the probability of the passenger to be tracked being searched, and provides real-time safety monitoring and response capabilities.

[0104] Furthermore, the present invention provides a method for tracking passenger trajectories across cameras based on video technology, which is described by taking the arrangement of surveillance cameras in a subway station and tracking passenger trajectories as an example:

[0105] (1) The cross-camera passenger trajectory tracking method based on video technology proposed in the present invention is used to construct a weighted directed graph of passenger trajectories, such as Figure 2 shown.

[0106] Among them, p1, p2, p3, and p4 are nodes, representing the entrance, security checkpoint, stairs, and subway door, respectively, and the direction is entrance-security checkpoint-stairs-subway door. ij are vertices, representing different cameras.

[0107] (2) After obtaining the target path using the cross-camera passenger trajectory tracking method based on video technology proposed in the present invention, the trajectory of the passenger to be tracked is tracked according to the target path.

[0108] For example, the target path is v 11—Security checkpoint 22 —Stairs 32 —Subway door v 41 ,Based on the order of vertices in the target path, the paths represented by the corresponding vertices (i.e., the vertices with the smallest path value) are selected in turn for execution to determine whether the passenger to be tracked is found.

[0109] If the passenger to be tracked is not found in the vertices of the target path planning, the next best vertex (the vertex with the next smallest path value) in the current node is selected as the current vertex, and the path search continues.

[0110] for example, Figure 2 Although the target path is v 11 —Security checkpoint 22 —Stairs 32 —Subway door v 41 , but in v 11 、v 22 The passengers to be tracked are found, and on the stairs v 32 If the vertex is not found, the suboptimal vertex (the vertex with the second smallest path value) in the staircase node is selected as the current vertex, that is, v is selected. 33 , so the actual path of the passenger to be tracked is v 11 —Security checkpoint 22 —Stairs 33 .

[0111] According to the target path, the tracking of the passenger trajectory to be tracked has not yet ended, and the next node subway door v is continued to be searched. 41 , since at vertex v 41 If no passenger to be tracked is found in the search results, the suboptimal vertex (the vertex with the second smallest path value) in the subway door node is selected as the current vertex, that is, vertex v 42 .

[0112] After the execution is completed, the tracking trajectory of the passenger to be tracked is the entrance v 11 —Security checkpoint 22 —Stairs 33 —Subway door v 42 .

[0113] After the target path is obtained by the cross-camera passenger trajectory tracking method based on video technology proposed in the present invention, the trajectory of the passenger to be tracked is tracked according to the target path, which can improve the intelligence level of the monitoring system, reduce manual intervention, and reduce the impact of human errors and subjective factors on monitoring decisions.

[0114] The present invention also provides a cross-camera passenger trajectory tracking system based on video technology, comprising:

[0115] Network topology structure building module: used to obtain the historical flow of people in the area to be detected, divide the area to be detected into different density areas according to the density of people flow, calculate the density weight of people flow in each density area, and take the density area with a density weight greater than the preset density weight as a node;

[0116] After obtaining the historical weight data of the cameras in the area to be detected, the cameras are taken as vertices and the connections between the cameras are edges;

[0117] According to the historical traffic and camera weight data, determine the direction of different nodes and build the network topology structure;

[0118] Vertex weight calculation module: Based on the network topology, calculate the probability of the target passenger appearing at the vertex;

[0119] The vertex weight is obtained by multiplying the probability of the target passenger appearing at the vertex, the time the target passenger passes the vertex, and the clarity value of the vertex;

[0120] Weighted directed graph construction module: used to obtain a weighted directed graph after performing weighted operations on vertices in the network topology structure according to vertex weights;

[0121] Path selection module: for each vertex on a node, the vertex weight, historical path weight, and path length are weighted and summed to obtain the path value of the current vertex. The path consisting of the vertices with the smallest path value at the same node is taken as the first path. The first paths of all nodes are connected to obtain the target path.

[0122] Passenger trajectory tracking module: used to obtain target monitoring data and track the trajectory of the passenger to be tracked based on the target path;

[0123] If the vertex in the target path does not track the passenger to be tracked, a path consisting of vertices with the next smallest path value at the node where the vertex is located is selected for tracking.

[0124] The present invention also provides a cross-camera passenger trajectory tracking device based on video technology, comprising a processor and a memory, wherein the processor implements the cross-camera passenger trajectory tracking method based on video technology when executing a computer program stored in the memory.

Claims

1. A method for tracking passenger trajectories across cameras based on video technology, characterized in that: The following steps are involved: S1. Build network topology: Obtain the historical flow of people in the area to be detected, divide the area to be detected into different density areas according to the density of people flow, calculate the density weight of people flow in each density area, and take the density area with a density weight greater than the preset density weight as a node; After obtaining the historical weight data of the cameras in the area to be detected, the cameras are taken as vertices and the connections between the cameras are edges; According to the historical traffic and camera weight data, determine the direction of different nodes and build the network topology structure; S2. Calculate vertex weights: Based on the network topology, the probability of the target passenger appearing at the vertex is calculated; The vertex weight is obtained by multiplying the probability of the target passenger appearing at the vertex, the time the target passenger passes the vertex, and the clarity value of the vertex; S3. Construct a weighted directed graph: According to the vertex weights, after weighting operations are performed on the vertices in the network topology structure, a weighted directed graph is obtained; S4. Select path: For each vertex on a node, the vertex weight, historical path weight, and path length are weighted and summed to obtain the path value of the current vertex. The path consisting of the vertices with the smallest path value at the same node is taken as the first path. The first paths of all nodes are connected to obtain the target path. S5. Tracking the trajectory of the passengers to be tracked: Obtain target monitoring data, and track the trajectory of the passenger to be tracked based on the target path; If the vertex in the target path does not track the passenger to be tracked, a path consisting of vertices with the next smallest path value at the node where the vertex is located is selected for tracking.

2. The method for tracking passenger trajectories across cameras based on video technology according to claim 1, characterized in that: The step S1 determines the direction of different nodes according to the historical flow of people and the historical weight data of the camera, specifically: Calculate the traffic weight of each edge based on the historical traffic; According to the camera's historical weight data, calculate the historical weight of each vertex; Based on the traffic weight and the historical weight of the vertex, the direction weight of the edge is calculated and the direction of the edge under different nodes is determined.

3. The method for tracking passenger trajectories across cameras based on video technology according to claim 2 is characterized in that: After calculating the direction weight of the edge based on the traffic weight and the historical weight of the vertex, the direction of the edge under different nodes is determined, specifically, according to the formula: W c =aW a +bW b , calculate the direction weight of the edge; Where W c is the direction weight of the edge; a is the coefficient of the traffic weight; W a is the traffic weight; b is the coefficient of the vertex's historical weight; W b is the historical weight of the vertex; According to the direction weight of the edge, the direction of the edge under different nodes is determined.

4. The method for tracking passenger trajectories across cameras based on video technology according to claim 1, characterized in that: The step S2 is to calculate the probability of the target passenger appearing at the vertex based on the network topology, specifically, according to the formula: Calculate the probability of the target passenger appearing at the vertex; Where A is the probability of the target passenger appearing at the vertex; H is the historical number of target passengers; T is the historical number of passengers passing through the vertex; W d It is the crowd density weight of the density zone to which the vertex belongs.

5. The method for tracking passenger trajectories across cameras based on video technology according to claim 1, characterized in that: In step S4, for each vertex on a node, the vertex weight, the historical path weight, and the path length are weighted and summed to obtain the path value of the current vertex, specifically, according to the formula: p i =argmin(αW(v)+βH(v)+γd(v)), get the path value of the current vertex; In the formula, p i represents the path value of the current vertex, 1≤i≤n; v represents the vertex on the same node; argmin() represents the minimum function; α is the coefficient of vertex weight; W(v) is the vertex weight; β is the coefficient of historical path weight; H(v) is the historical path weight; γ is the coefficient of path length; d(v) is the path length from the starting node to the current vertex.

6. The method for tracking passenger trajectories across cameras based on video technology according to claim 1, characterized in that: The step S5, obtaining target monitoring data and tracking the trajectory of the passenger to be tracked based on the target path, also includes: If the current vertex tracks the passenger to be tracked, the path tracking operation from the current vertex to the next vertex will continue. When the vertex of the last node of the target path is reached, the path execution will be terminated and the tracked path will be recorded.

7. The method for tracking passenger trajectories across cameras based on video technology according to claim 1, characterized in that: The step S5, if the vertex in the target path does not track the passenger to be tracked, then a path consisting of vertices with a node having a second smallest path value at the vertex is selected for tracking, further comprising: If the passenger to be tracked is not tracked by traversing all vertices of the current node, the trajectory of the passenger to be tracked is recorded as terminating at the vertex of the previous node.

8. A cross-camera passenger trajectory tracking system based on video technology, characterized in that: include: Network topology structure building module: used to obtain the historical flow of people in the area to be detected, divide the area to be detected into different density areas according to the density of people flow, calculate the density weight of people flow in each density area, and take the density area with a density weight greater than the preset density weight as a node; After obtaining the historical weight data of the cameras in the area to be detected, the cameras are taken as vertices and the connections between the cameras are edges; According to the historical traffic and camera weight data, determine the direction of different nodes and build the network topology structure; Vertex weight calculation module: Based on the network topology, calculate the probability of the target passenger appearing at the vertex; The vertex weight is obtained by multiplying the probability of the target passenger appearing at the vertex, the time the target passenger passes the vertex, and the clarity value of the vertex; Weighted directed graph construction module: used to obtain a weighted directed graph after performing weighted operations on vertices in the network topology structure according to vertex weights; Path selection module: for each vertex on a node, the vertex weight, historical path weight, and path length are weighted and summed to obtain the path value of the current vertex. The path consisting of the vertices with the smallest path value at the same node is taken as the first path. The first paths of all nodes are connected to obtain the target path. Passenger trajectory tracking module: used to obtain target monitoring data and track the trajectory of the passenger to be tracked based on the target path; If the vertex in the target path does not track the passenger to be tracked, a path consisting of vertices with the next smallest path value at the node where the vertex is located is selected for tracking.

9. A cross-camera passenger trajectory tracking device based on video technology, characterized in that: The method comprises a processor and a memory, wherein when the processor executes the computer program stored in the memory, the method for tracking passenger trajectories across cameras based on video technology as described in any one of claims 1 to 7 is implemented.

Images