Expressway camera-crossing vehicle trajectory reconstruction method

Abstract

The invention discloses an expressway camera-crossing vehicle track reconstruction method, and relates to the technical field of intelligent traffic monitoring, and the method comprises the steps: automatically constructing a network topological relation according to the pose information of a monitoring camera; an entrance recognizes license plates, creates an initial track and adds the track to a track set; performing multi-dimensional similarity fusion on the main road based on vehicle color, type and appearance re-identification features, screening candidate trajectories in a topological constraint and time window, and determining a target trajectory to continue the path; the exit recognizes the license plate to realize track closing. And processing the identification error through a predefined color and vehicle type similarity matrix, and dynamically adjusting the time window based on the real-time vehicle speed. The method can achieve whole-course track reconstruction, improves the vehicle distinguishing capability, is suitable for a large-scale camera network, and reduces the dependence on license plate recognition.

Description

Technical Field

[0001] This invention relates to the field of intelligent traffic monitoring technology, and in particular to a method for reconstructing vehicle trajectories across cameras on highways. Background Technology

[0002] Currently, the core problem facing highway vehicle trajectory reconstruction technology across cameras lies in the fact that the main highway surveillance cameras, due to their installation location being far from the road and their wide field of view, cannot capture images with sufficient effective pixels for license plate recognition algorithms, thus failing to obtain stable license plate recognition results. Vehicles on the main highway are in an anonymous state, lacking identity information; only the highway entrance and exit cameras can provide reliable vehicle identity anchors. Existing technologies can only correlate license plate information from entrance and exit records, failing to reconstruct the continuous driving trajectory of a vehicle across multiple surveillance cameras on the main highway, and further unable to support advanced application requirements such as vehicle behavior analysis, road operation monitoring, anomaly event tracing, and full route reconstruction.

[0003] To address the aforementioned technical challenges, existing highway monitoring systems typically consist of multiple types of equipment, including entrance cameras, main road cameras, and exit cameras, and employ the following technical solutions for data processing. At highway entrances and exits, checkpoint cameras with license plate recognition capabilities are deployed to collect vehicle identification data, creating entry and exit records as the primary source of identification. On the main road, cameras for traffic monitoring are deployed. Due to their long installation distances and wide coverage areas, the image resolution of these cameras often fails to meet the pixel requirements for license plate recognition. Therefore, main road cameras typically only collect non-identification information such as vehicle appearance, location, and time. The backend primarily matches entrance and exit records with license plate numbers and approximate travel time windows to achieve logical closure of the travel path. The travel records generated by main road cameras are generally displayed as supplementary information and do not participate in the continuous trajectory construction process. The entire technical framework revolves around cross-camera associations at entrances and exits, forming a discrete data processing mode primarily focused on travel record management.

[0004] However, there are several objective limitations in constructing vehicle trajectories across cameras. First, the main highway section lacks identification information, making it impossible to form a continuous trajectory chain. Main highway cameras cannot provide license plate information; only entrance and exit cameras have identification anchor points. Existing processes can only form a two-dimensional logical closure from entrance to exit, failing to cover the continuous trajectory of vehicles between monitoring points on the main highway. Second, highway main highway records lack reliable association conditions and cannot be used for cross-camera matching. The time, location, and appearance features collected by main highway cameras are weak feature information, lacking unique vehicle identification information, insufficient to support stable correspondences between cameras. Therefore, it is difficult to associate multiple main highway records into a continuous trajectory. Third, the lack of multi-point trajectory data on the main highway makes it difficult to support higher-level analysis needs. Due to the inability to obtain continuous path information for vehicles on the main highway section, capabilities in behavior analysis, event backtracking, and anomaly detection are significantly limited. These limitations mean that existing technologies can only perform simple matching between entrances and exits with complete identification information, and cannot achieve continuous trajectory reconstruction across cameras on highway main highway sections where identification information is missing.

[0005] In summary, existing highway cross-camera vehicle trajectory reconstruction technologies are primarily focused on traffic record management and lack the ability to describe the continuous travel paths of vehicles across various monitoring points on the main highway. A key deficiency lies in the inability of main-road cameras to provide stable license plate recognition results, resulting in vehicles remaining anonymous on the main highway. This fundamental problem prevents existing technologies from achieving continuous trajectory association between multiple monitoring cameras on the main highway. Therefore, there is an urgent need for a method that can comprehensively utilize road topology information and vehicle appearance attribute information to reconstruct vehicle trajectories across cameras, even under conditions where license plate information is unstable or completely missing. Summary of the Invention

[0006] The technical problem to be solved by this invention is to address the shortcomings of existing technologies, specifically the problems of vehicle anonymization due to the inability to recognize license plates in highway main road monitoring systems, and the inability of existing technologies to reconstruct the continuous driving trajectory of vehicles between main road monitoring cameras. Specifically, this invention provides a method for reconstructing vehicle trajectories across cameras on highways, as detailed below: 1) In a first aspect, the present invention provides a method for reconstructing vehicle trajectories across cameras on highways, the specific technical solution of which is as follows: S1. Based on the pose information of the surveillance cameras, obtain the network topology of the surveillance cameras. The network topology of the surveillance cameras includes the upstream and downstream connection relationships between the surveillance cameras and the time window parameters. S2, Based on the first vehicle identity information collected by the highway entrance monitoring camera, an initial vehicle driving trajectory is obtained, and the initial vehicle driving trajectory is added to the vehicle driving trajectory set; the first vehicle identity information includes the license plate number, the initial vehicle driving trajectory includes the monitoring point sequence and vehicle attribute information, and the vehicle driving trajectory set includes the driving trajectories of all vehicles whose trajectories have been reconstructed. S3, Receive vehicle visual features collected by highway main road monitoring cameras, wherein the vehicle visual features do not include license plate numbers; Determine the candidate upstream monitoring set corresponding to the current main road monitoring camera based on the upstream and downstream connection relationships in the monitoring camera network topology; Filter the vehicle trajectory set from the vehicle trajectory set based on time window parameters and the candidate upstream monitoring set to obtain a candidate trajectory set; Perform multi-dimensional similarity fusion based on the vehicle visual features and the vehicle attribute information corresponding to each candidate trajectory in the candidate trajectory set to obtain a comprehensive matching result; Determine the first target trajectory from the candidate trajectory set based on the comprehensive matching result; Add the current main road monitoring point to the monitoring point sequence of the first target trajectory to obtain a trajectory continuation result; Replace the corresponding original vehicle trajectory record in the vehicle trajectory set based on the trajectory continuation result, and update the vehicle trajectory set; S4. Based on the second vehicle identity information collected by the highway exit monitoring camera, determine the second target trajectory from the vehicle trajectory set based on the license plate number in the second vehicle identity information, add the exit monitoring point to the monitoring point sequence of the second target trajectory, obtain the trajectory closure result, replace the corresponding original vehicle trajectory record in the vehicle trajectory set based on the trajectory closure result, and update the vehicle trajectory set.

[0007] The beneficial effects of the cross-camera vehicle trajectory reconstruction method for highways provided by this invention are as follows: The system automatically constructs the network topology of surveillance cameras based on their pose information, eliminating the need for manual configuration of camera relationships, reducing configuration workload and errors, and making it suitable for large-scale high-speed camera systems. In situations where license plates cannot be recognized by highway mainline surveillance, leading to vehicle anonymization, it utilizes multi-dimensional similarity fusion based on vehicle color, type, and appearance re-identification features to achieve continuous trajectory continuity between mainline surveillance cameras, solving the problem of existing technologies being unable to reconstruct continuous driving trajectories on the mainline. Based on the upstream and downstream connections in the surveillance camera network topology, it filters candidate upstream surveillance sets, narrowing the candidate trajectory search range, avoiding candidate space explosion, improving computational efficiency, and ensuring that the trajectory reconstruction process remains controllable under large-scale camera networks. It uses accurate license plate information at highway entrances to create a trajectory start point and accurate license plate information at exits to create a closed trajectory end point, forming a complete identity loop, ensuring consistency between the reconstructed trajectory and the entrance / exit identity information.

[0008] Based on the above solution, the present invention can be further improved as follows.

[0009] Furthermore, the pose information includes road marker data and driving direction information; The step of obtaining the network topology of the surveillance cameras based on their pose information includes: The surveillance cameras are sorted based on road marker data. The surveillance cameras with smaller road marker values ​​are identified as upstream surveillance cameras, and the surveillance cameras with larger road marker values ​​are identified as downstream surveillance cameras. Based on the driving direction information, the monitoring cameras are divided into groups for the uphill direction and groups for the downhill direction. The geographical distance between adjacent surveillance cameras is calculated based on the road station data of adjacent surveillance cameras; time window parameters are calculated based on the geographical distance between surveillance cameras and the standard driving speed on highways.

[0010] Furthermore, the vehicle visual features include vehicle color, vehicle type, and vehicle appearance re-identification features; The multi-dimensional similarity fusion includes: Calculate the appearance similarity between the vehicle appearance re-identification features in the current vehicle visual features and the vehicle appearance re-identification features in the vehicle attribute information corresponding to the candidate trajectory; Query the preset color similarity matrix to obtain color similarity, and query the preset vehicle type similarity matrix to obtain type similarity; Based on the highway scene conditions, the appearance similarity weight, color similarity weight, and type similarity weight are determined. The appearance similarity, color similarity, and type similarity are then weighted and fused to obtain a comprehensive matching result.

[0011] Furthermore, it also includes: Receive vehicle speed data reported by the monitoring video analysis equipment, statistically analyze the vehicle speed data within a preset time period, and obtain the average vehicle speed. The time window parameters are adjusted based on the comparison between the average vehicle speed and the preset congestion threshold, and the adjusted time window parameters are applied to the screening process of obtaining a candidate trajectory set from the set of vehicle driving trajectories.

[0012] 2) In a second aspect, the present invention also provides a highway cross-camera vehicle trajectory reconstruction system, the specific technical solution of which is as follows: a topology construction module, a trajectory creation module, a trajectory continuation module, and a trajectory closure module; The topology construction module is used to obtain the network topology relationship of the surveillance cameras based on the pose information of the surveillance cameras. The network topology relationship of the surveillance cameras includes the upstream and downstream connection relationship between the surveillance cameras and the time window parameter. The trajectory creation module is used to obtain an initial vehicle driving trajectory based on the first vehicle identity information collected by the highway entrance monitoring camera, and add the initial vehicle driving trajectory to the vehicle driving trajectory set; the first vehicle identity information includes the license plate number, the initial vehicle driving trajectory includes the monitoring point sequence and vehicle attribute information, and the vehicle driving trajectory set includes the driving trajectories of all vehicles whose trajectories have been reconstructed. The trajectory continuation module is used to receive vehicle visual features collected by highway main road monitoring cameras, wherein the vehicle visual features do not include license plate numbers; determine the candidate upstream monitoring set corresponding to the current main road monitoring camera based on the upstream and downstream connection relationship in the monitoring camera network topology; filter the candidate trajectory set from the vehicle driving trajectory set based on time window parameters and the candidate upstream monitoring set; perform multi-dimensional similarity fusion based on the vehicle visual features and the vehicle attribute information corresponding to each candidate trajectory in the candidate trajectory set to obtain a comprehensive matching result; determine a first target trajectory from the candidate trajectory set based on the comprehensive matching result; append the current main road monitoring point to the monitoring point sequence of the first target trajectory to obtain a trajectory continuation result; replace the corresponding original vehicle driving trajectory record in the vehicle driving trajectory set based on the trajectory continuation result, and update the vehicle driving trajectory set. The trajectory closure module is used to determine a second target trajectory from the vehicle trajectory set based on the license plate number in the second vehicle identity information collected by the highway exit monitoring camera, add the exit monitoring point to the monitoring point sequence of the second target trajectory, obtain the trajectory closure result, replace the corresponding original vehicle trajectory record in the vehicle trajectory set based on the trajectory closure result, and update the vehicle trajectory set.

[0013] Based on the above solution, the present invention can be further improved as follows.

[0014] Furthermore, the pose information includes road marker data and driving direction information; The step of obtaining the network topology of the surveillance cameras based on their pose information includes: The surveillance cameras are sorted based on road marker data. The surveillance cameras with smaller road marker values ​​are identified as upstream surveillance cameras, and the surveillance cameras with larger road marker values ​​are identified as downstream surveillance cameras. Based on the driving direction information, the monitoring cameras are divided into groups for the uphill direction and groups for the downhill direction. The geographical distance between adjacent surveillance cameras is calculated based on the road station data of adjacent surveillance cameras; time window parameters are calculated based on the geographical distance between surveillance cameras and the standard driving speed on highways.

[0015] Furthermore, the vehicle visual features include vehicle color, vehicle type, and vehicle appearance re-identification features; The multi-dimensional similarity fusion includes: Calculate the appearance similarity between the vehicle appearance re-identification features in the current vehicle visual features and the vehicle appearance re-identification features in the vehicle attribute information corresponding to the candidate trajectory; Query the preset color similarity matrix to obtain color similarity, and query the preset vehicle type similarity matrix to obtain type similarity; Based on the highway scene conditions, the appearance similarity weight, color similarity weight, and type similarity weight are determined. The appearance similarity, color similarity, and type similarity are then weighted and fused to obtain a comprehensive matching result.

[0016] Furthermore, it also includes: Receive vehicle speed data reported by the monitoring video analysis equipment, statistically analyze the vehicle speed data within a preset time period, and obtain the average vehicle speed. The time window parameters are adjusted based on the comparison between the average vehicle speed and the preset congestion threshold, and the adjusted time window parameters are applied to the screening process of obtaining a candidate trajectory set from the set of vehicle driving trajectories.

[0017] 3) In a third aspect, the present invention also provides a computer device, the computer device including a processor coupled to a memory, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to enable the computer device to implement any of the above methods.

[0018] 4) In a fourth aspect, the present invention also provides a computer-readable storage medium storing at least one computer program, which is loaded and executed by a processor to enable a computer to perform any of the above methods.

[0019] It should be noted that the beneficial effects of the technical solutions of the second to fourth aspects of the present invention and their corresponding possible implementations can be found in the above description of the technical effects of the first aspect and its corresponding possible implementations, and will not be repeated here. Attached Figure Description

[0020] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating the steps of a method for reconstructing vehicle trajectories across cameras on a highway, according to an embodiment of the present invention. Figure 2This is a schematic diagram illustrating the process of constructing the network topology of a highway cross-camera vehicle trajectory reconstruction method according to an embodiment of the present invention. Figure 3 This is a schematic diagram of a method for reconstructing vehicle trajectories across cameras on highways based on dynamically adjusting time window parameters according to vehicle speed, as described in an embodiment of the present invention. Figure 4 This is a structural block diagram of a computer device according to an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0022] like Figure 1 As shown in the figure, a method for reconstructing vehicle trajectories across cameras on highways according to an embodiment of the present invention includes the following steps: S1. Based on the pose information of the surveillance cameras, obtain the network topology of the surveillance cameras. The network topology of the surveillance cameras includes the upstream and downstream connection relationships between the surveillance cameras and the time window parameters. S2. Based on the first vehicle identity information collected by the highway entrance monitoring camera, the initial vehicle driving trajectory is obtained and added to the vehicle driving trajectory set. The first vehicle identity information includes the license plate number, the initial vehicle driving trajectory includes the monitoring point sequence and vehicle attribute information, and the vehicle driving trajectory set includes the driving trajectories of all vehicles whose trajectories have been reconstructed. S3 receives vehicle visual features collected by highway mainline monitoring cameras. These visual features do not include license plate numbers. Based on the upstream and downstream connections in the monitoring camera network topology, a candidate upstream monitoring set corresponding to the current mainline monitoring camera is determined. A candidate trajectory set is obtained by filtering from the vehicle trajectory set based on time window parameters and the candidate upstream monitoring set. Multi-dimensional similarity fusion is performed between the vehicle visual features and the vehicle attribute information corresponding to each candidate trajectory in the candidate trajectory set to obtain a comprehensive matching result. Based on the comprehensive matching result, a first target trajectory is determined from the candidate trajectory set. The current mainline monitoring point is added to the monitoring point sequence of the first target trajectory to obtain a trajectory continuation result. Based on the trajectory continuation result, the original vehicle trajectory record corresponding to the current mainline monitoring point is replaced in the vehicle trajectory set, thus updating the vehicle trajectory set. S4. Based on the second vehicle identity information collected by the highway exit monitoring camera, determine the second target trajectory from the vehicle trajectory set based on the license plate number in the second vehicle identity information, add the exit monitoring point to the monitoring point sequence of the second target trajectory, obtain the trajectory closure result, replace the corresponding original vehicle trajectory record in the vehicle trajectory set based on the trajectory closure result, and update the vehicle trajectory set.

[0023] The beneficial effects of the cross-camera vehicle trajectory reconstruction method for highways provided by this invention are as follows: The system automatically constructs the network topology of surveillance cameras based on their pose information, eliminating the need for manual configuration of camera relationships, reducing configuration workload and errors, and making it suitable for large-scale high-speed camera systems. In situations where license plates cannot be recognized by highway mainline surveillance, leading to vehicle anonymization, it utilizes multi-dimensional similarity fusion based on vehicle color, type, and appearance re-identification features to achieve continuous trajectory continuity between mainline surveillance cameras, solving the problem of existing technologies being unable to reconstruct continuous driving trajectories on the mainline. Based on the upstream and downstream connections in the surveillance camera network topology, it filters candidate upstream surveillance sets, narrowing the candidate trajectory search range, avoiding candidate space explosion, improving computational efficiency, and ensuring that the trajectory reconstruction process remains controllable under large-scale camera networks. It uses accurate license plate information at highway entrances to create a trajectory start point and accurate license plate information at exits to create a closed trajectory end point, forming a complete identity loop, ensuring consistency between the reconstructed trajectory and the entrance / exit identity information.

[0024] It should be noted that, for ease of understanding, the technical terms used in this solution will be explained one by one, and will not be repeated hereafter: The pose information of the surveillance cameras is a dataset containing road marker data and travel direction information. The road marker data uses K-code format to represent the highway mileage, and the travel direction information indicates the travel direction of the lane covered by the surveillance camera. In this solution, the pose information is used to automatically infer the spatial relationship between the surveillance cameras.

[0025] Surveillance camera network topology: A data structure built based on the pose information of surveillance cameras, used to describe the spatial arrangement and constraints between surveillance cameras. The surveillance camera network topology includes upstream and downstream connections between surveillance camera nodes and time window parameters. Upstream and downstream connections refer to the physical sequence determined by the road marker data, while time window parameters restrict the time range for vehicles to cross surveillance camera lines.

[0026] Upstream and downstream connectivity: The physical spatial order is determined based on the road marker data of the surveillance cameras. Surveillance cameras with smaller road marker values ​​are identified as upstream cameras, and those with larger road marker values ​​are identified as downstream cameras, forming a unidirectional logical chain. In this scheme, the upstream and downstream connectivity is used to narrow down the search range of candidate trajectories.

[0027] Time window parameter: A time constraint value calculated based on the geographical distance between adjacent surveillance cameras and the standard driving speed on highways. It is used to filter the candidate trajectory set from the vehicle trajectory set to ensure that only vehicle trajectory records that arrive at the current surveillance camera within a reasonable travel time range are selected.

[0028] Highway entrance surveillance cameras: checkpoint surveillance cameras deployed at the entrance of highway toll stations, equipped with license plate number recognition capabilities.

[0029] First, vehicle identification information: a multi-dimensional collection of vehicle data collected by highway entrance monitoring cameras, including license plate number, vehicle color, vehicle type, vehicle appearance re-identification features, and collection time. The license plate number serves as a strong identifier used to create the initial vehicle trajectory.

[0030] License plate number: A character sequence extracted by a license plate recognition algorithm from highway entrance or exit surveillance cameras. The license plate number serves as a unique identification anchor, providing reliable vehicle identification during both the trajectory creation and trajectory closure phases.

[0031] Initial vehicle trajectory: This is a data record created after the highway entrance monitoring camera identifies the license plate number. It includes a unique trajectory identifier, license plate number, trajectory start time, trajectory end time, monitoring point sequence, and vehicle attribute information. The initial vehicle trajectory is added to the vehicle trajectory set as the starting point for subsequent continuation of the main road trajectory.

[0032] Monitoring point sequence: An ordered list of monitoring camera identifiers and acquisition times associated with the vehicle's travel trajectory. New monitoring point information is added after each acquisition by a main road monitoring camera or exit monitoring camera, forming a complete spatiotemporal path description of the vehicle.

[0033] Vehicle attribute information: A set of appearance feature data extracted from (first / second) vehicle identity information or vehicle visual features, including vehicle color, vehicle type, and vehicle appearance re-identification features. Vehicle attribute information is used for multi-dimensional similarity fusion calculation in the main road phase.

[0034] Vehicle trajectory set: A dynamically maintained data structure that stores the trajectory records of all vehicles undergoing trajectory reconstruction. The vehicle trajectory set initially consists of the initial vehicle trajectories created in the entry stage. During the main road stage, the corresponding records are updated based on the trajectory continuation results, and during the exit stage, the corresponding records are updated based on the trajectory closure results, enabling parallel trajectory reconstruction for multiple vehicles.

[0035] The driving trajectory of all vehicles reconstructed: refers to the sum of vehicle trajectory records in various states such as entrance created, main road continuing, and exit closed during the vehicle trajectory reconstruction process across the highway camera, which constitutes the complete content of the vehicle driving trajectory set.

[0036] Highway main road surveillance cameras: Surveillance cameras deployed along the main road of highways for road condition monitoring but unable to identify license plate numbers.

[0037] Vehicle visual features: A set of non-identified vehicle information collected by highway mainline surveillance cameras, excluding license plate numbers, but including vehicle color, vehicle type, and vehicle appearance re-identification features. Vehicle visual features are used to achieve trajectory continuation across surveillance cameras when license plate information is missing.

[0038] Candidate upstream monitoring set: Based on the upstream and downstream connections in the network topology of the surveillance cameras, this is a set of surveillance cameras located upstream of the current main road surveillance camera in terms of physical location. The candidate upstream monitoring set is used to filter vehicle trajectory records that may reach the current main road surveillance camera from the vehicle trajectory set.

[0039] Candidate trajectory set: A subset of vehicle trajectory records selected from the vehicle trajectory set based on time window parameters and the candidate upstream monitoring set. Each candidate trajectory in the candidate trajectory set must satisfy the following conditions: the last associated monitoring point is located within the candidate upstream monitoring set, and the difference between the trajectory termination time and the current acquisition time is within the time window parameter range.

[0040] Multi-dimensional similarity fusion: This is a computational process that performs multi-dimensional matching and weighted fusion of appearance re-identification features, color features, and type features in vehicle visual features with vehicle attribute information corresponding to candidate trajectories. Multi-dimensional similarity fusion includes calculating appearance similarity, querying a preset color similarity matrix to obtain color similarity, querying a preset vehicle type similarity matrix to obtain type similarity, and then performing a weighted sum based on weight coefficients to obtain a comprehensive matching result.

[0041] Comprehensive matching result: This value, calculated through multi-dimensional similarity fusion, quantifies the degree of matching between vehicle visual features and vehicle attribute information of candidate trajectories. A higher comprehensive matching result value indicates a greater likelihood that the vehicle visual features and candidate trajectory belong to the same vehicle.

[0042] First target trajectory: The most matching vehicle trajectory record determined from the candidate trajectory set based on the comprehensive matching results. The current main road monitoring point is appended to the monitoring point sequence of the first target trajectory to obtain the trajectory continuation result.

[0043] Trajectory continuation result: The current main road monitoring point information is appended to the monitoring point sequence of the first target trajectory, and the vehicle driving trajectory record obtained after the trajectory termination time is updated. The trajectory continuation result is used to replace the corresponding original vehicle driving trajectory record in the vehicle driving trajectory set, realizing the dynamic updating of the vehicle driving trajectory set.

[0044] Original vehicle trajectory records: Vehicle trajectory records stored in the vehicle trajectory set before being updated by the current main road monitoring camera. After the original vehicle trajectory records are replaced by the trajectory continuation results, the vehicle trajectory set completes one update operation.

[0045] Highway exit surveillance cameras: These are checkpoint surveillance cameras deployed at highway toll station exits, equipped with license plate number recognition capabilities. The secondary vehicle identification information collected by the exit surveillance cameras includes the license plate number.

[0046] Second vehicle identification information: A multi-dimensional set of vehicle data collected by highway exit surveillance cameras, including license plate number, vehicle color, vehicle type, vehicle appearance re-identification features, and collection time. The license plate number in the second vehicle identification information is used to determine the trajectory of the second target from the set of vehicle travel trajectories.

[0047] Second target trajectory: Based on the license plate number in the second vehicle identity information, the vehicle's trajectory record is matched and determined from the set of vehicle trajectories. An exit monitoring point is added to the monitoring point sequence of the second target trajectory to obtain the trajectory closure result.

[0048] Trajectory Closure Result: This is the vehicle trajectory record obtained by appending the exit monitoring point information to the monitoring point sequence of the second target trajectory and marking the trajectory status as closed. The trajectory closure result is used to replace the corresponding original vehicle trajectory record in the vehicle trajectory set, completing the reconstruction of the entire vehicle trajectory.

[0049] Road marker data: This data represents highway mileage locations using K-code format, for example, K120+500 represents 120 kilometers and 500 meters. Road marker data is used to calculate geographical distances between surveillance cameras and to determine upstream and downstream connections between them.

[0050] Driving direction information: This data indicates the driving direction of the lanes covered by the surveillance cameras, divided into uphill and downhill directions. Driving direction information is used to separate uphill and downhill camera groups.

[0051] Upward-direction camera group: This is the collection of all cameras whose travel direction information points in the upward direction. The upward-direction camera group is independent of the downward-direction camera group, each forming its own independent camera network topology.

[0052] Geographic distance between surveillance cameras: The physical distance value obtained by converting the road station data difference between adjacent surveillance cameras. The geographic distance between surveillance cameras is used to calculate time window parameters.

[0053] Standard highway speed: The preset design speed or legal speed limit for highways. The standard highway speed is used to calculate time window parameters in conjunction with the geographical distance between surveillance cameras.

[0054] Vehicle color: The main color attribute of the vehicle extracted from the vehicle image, including categories such as white, silver, black, gray, blue, green, brown, yellow, and purple. Vehicle color is an important component of the visual characteristics of a vehicle.

[0055] Vehicle type: Vehicle type attributes extracted from vehicle images, including categories such as sedans, SUVs, multi-purpose passenger vehicles, vans, trucks, and heavy trucks. Vehicle type is an important component of vehicle visual characteristics.

[0056] Vehicle appearance re-identification features: These are high-dimensional feature vectors extracted from vehicle images using a neural network model, used to identify the same vehicle across different surveillance cameras. Vehicle appearance re-identification features are the core features for achieving vehicle identity association even when license plate numbers are missing.

[0057] Appearance similarity: This is a numerical value calculated using the vector distance method to determine the similarity between the vehicle appearance re-identification feature vector in the current vehicle's visual features and the vehicle appearance re-identification feature vector in the candidate trajectory's corresponding vehicle attribute information. The appearance similarity value typically ranges from -1.0 to 1.0 or from 0 to 1.0.

[0058] A preset color similarity matrix is ​​a manually calibrated matrix data structure used to query the similarity score between any two vehicle colors. Easily confused color pairs are assigned medium scores, obviously inconsistent color pairs are assigned negative scores, and identical colors are assigned the highest score. In this solution, the preset color similarity matrix effectively addresses color recognition errors.

[0059] Color similarity: The score obtained by querying the preset color similarity matrix is ​​used to quantify the degree of matching between the vehicle color in the current vehicle's visual features and the vehicle color in the candidate trajectory's corresponding vehicle attribute information. Color similarity, appearance similarity, and type similarity participate in weighted fusion.

[0060] A pre-defined vehicle type similarity matrix is ​​used to query the similarity score between any two vehicle types. Easily confused vehicle type pairs are assigned medium scores, obviously impossible combinations are assigned negative scores, and identical vehicle types are assigned the highest score. In this solution, the pre-defined vehicle type similarity matrix effectively addresses the vehicle type recognition error problem.

[0061] Type similarity: The score obtained by querying the preset vehicle type similarity matrix is ​​used to quantify the degree of matching between the vehicle type in the current vehicle's visual features and the vehicle type in the corresponding vehicle attribute information of the candidate trajectory. Type similarity, appearance similarity, and color similarity are all used in a weighted fusion process.

[0062] Appearance similarity weight: A coefficient used to control the proportion of appearance similarity in the overall matching result. The appearance similarity weight is dynamically adjusted according to the highway scene conditions. For example, in a nighttime scene, the color similarity weight is reduced, and the appearance similarity weight is increased accordingly.

[0063] Color similarity weight: A coefficient used to control the proportion of color similarity contributed to the overall matching result. The color similarity weight is dynamically adjusted according to the conditions of the highway scene; for example, the weight is reduced in nighttime scenes due to the decreased reliability of color recognition.

[0064] Type similarity weight: A coefficient used to control the proportion of type similarity in the overall matching result. The type similarity weight is set according to the needs of the scenario and remains relatively stable.

[0065] Surveillance video analysis equipment: Video analysis equipment deployed along highways is used to detect vehicles from surveillance videos and extract vehicle speed data.

[0066] Vehicle speed data: A set of instantaneous speed values ​​of individual vehicles extracted from surveillance videos using video analysis algorithms, used to calculate average vehicle speed.

[0067] Average vehicle speed: The arithmetic mean of all vehicle speed data collected within a preset time period. Average vehicle speed is used to determine whether the traffic condition is congested or smooth.

[0068] Preset congestion threshold: A pre-defined lower limit for average vehicle speed; a speed below this threshold is considered a traffic congestion state. The preset congestion threshold is determined based on historical highway traffic data or management experience.

[0069] Figure 2 A flowchart illustrating the process of building a network topology for surveillance cameras, as shown below. Figure 2 As shown, in another embodiment of this solution, S1 is specifically implemented as follows: The position information of the surveillance camera includes road marker data and travel direction information. The road marker data is represented in K-code format, for example, K120+500 represents the position at 120 km 500 m. The travel direction information indicates the direction of travel in the lane covered by the surveillance camera.

[0070] The road marker data from all surveillance cameras is analyzed and converted into numerical form. The conversion rule is as follows: multiply the kilometer value in the K-code by 1000 and add the meter value to obtain the marker value in meters.

[0071] All surveillance cameras are sorted based on their station numbers, arranged in ascending order. For the sorted camera sequence, cameras with smaller station numbers are designated as upstream cameras, and those with larger station numbers are designated as downstream cameras. For any two cameras i and j, if the station number of camera i is less than that of camera j, then camera i is located upstream of camera j, and camera j is located downstream of camera i.

[0072] Based on the travel direction information, all surveillance cameras are divided into uplink camera groups and downlink camera groups. The uplink camera group contains all cameras whose travel direction is up, and the downlink camera group contains all cameras whose travel direction is down. The uplink and downlink camera groups each form an independent surveillance camera network topology.

[0073] For each group of surveillance cameras in the uplink and downlink directions, the road station difference between adjacent surveillance cameras is calculated sequentially according to the sorted chainage values. The road station difference is the geographical distance between adjacent surveillance cameras.

[0074] The time window parameter is obtained by dividing the geographical distance between surveillance cameras by the standard speed of the highway. The standard speed of the highway is set according to the highway design standards or legal speed limits, and the unit is meters per second. The time window parameter represents the time reference value required for a vehicle to travel from the upstream surveillance camera to the downstream surveillance camera at the standard speed.

[0075] The upstream and downstream connections between surveillance cameras and the time window parameters are stored as the surveillance camera network topology. The surveillance camera network topology is represented by a graph data structure, where nodes represent surveillance cameras, directed edges represent upstream and downstream connections, and the attributes of the edges include time window parameters and the station numbers of the upstream and downstream surveillance cameras.

[0076] In another embodiment of this solution, S2 is specifically implemented as follows: The first vehicle identification information collected by the highway entrance monitoring camera includes the license plate number, vehicle color, vehicle type, vehicle appearance re-identification features, and the time of acquisition. The license plate number in the first vehicle identification information is identified and output by the entrance monitoring camera using a license plate recognition algorithm, and the format is a character sequence conforming to license plate management regulations. The vehicle color in the first vehicle identification information is extracted from the vehicle image using a color classification algorithm; color categories can include white, silver, black, gray, blue, green, brown, yellow, purple, and unknown. The vehicle type in the first vehicle identification information is extracted from the vehicle image using a vehicle type classification algorithm; vehicle type categories can include sedan, SUV, multi-purpose passenger vehicle, van, truck, heavy truck, and unknown. The vehicle appearance re-identification features in the first vehicle identification information are extracted from the vehicle image using a convolutional neural network model, with a feature dimension of 512-dimensional floating-point vectors. The acquisition time in the first vehicle identification information is the timestamp of the entrance monitoring camera capturing the vehicle image, accurate to milliseconds.

[0077] Initialize the initial vehicle trajectory, including: generating a unique identifier for the trajectory in the initial vehicle trajectory using UUID format; directly assigning the license plate number from the first vehicle identity information to the license plate number field in the initial vehicle trajectory; assigning the collection time from the first vehicle identity information to the trajectory start time field in the initial vehicle trajectory; assigning the collection time from the first vehicle identity information to the trajectory end time field in the initial vehicle trajectory; initializing the monitoring point sequence field to an empty list; and the vehicle attribute information field containing three subfields: color attribute, type attribute, and appearance re-identification feature attribute. The color attribute stores the vehicle color from the first vehicle identity information, the type attribute stores the vehicle type from the first vehicle identity information, and the appearance re-identification feature attribute stores the vehicle appearance re-identification feature vector from the first vehicle identity information.

[0078] Append the entrance surveillance camera identifier to the monitoring point sequence field. The entrance surveillance camera identifier is a unique number assigned during camera registration, in string format. Append the acquisition time from the first vehicle identification information to the monitoring point sequence field, forming a paired record with the entrance surveillance camera identifier. Each element in the monitoring point sequence field contains two sub-elements: the surveillance camera identifier and the acquisition time.

[0079] The vehicle trajectory set is implemented using a hash table data structure. The key of the hash table is a unique identifier for the trajectory, and the value is the initial vehicle trajectory object. The initial vehicle trajectory's unique identifier is used as the key, and the initial vehicle trajectory object is used as the value, to insert the vehicle trajectory set. After the insertion operation, the vehicle trajectory set contains the trajectories of all vehicles whose trajectories have been reconstructed, including the initial vehicle trajectory created in this operation. The vehicle trajectory set supports subsequent query, filtering, and update operations at the main road and exit stages.

[0080] Each trajectory record in the vehicle trajectory set represents the complete spatiotemporal path of a vehicle being tracked. The number of trajectory records in the vehicle trajectory set increases as a vehicle enters a highway entrance and decreases as a vehicle exits a highway exit. The size of the set dynamically reflects the total number of vehicles whose trajectories are being reconstructed on the highway.

[0081] In another embodiment of this solution, S3 is specifically implemented as follows: Highway mainline surveillance cameras collect vehicle visual features, including vehicle color, vehicle type, vehicle appearance re-identification features, and the acquisition time, but excluding license plate numbers. Because the mainline surveillance cameras are installed far from the road and have a wide field of view, the image resolution is insufficient to meet the effective pixel requirements of the license plate recognition algorithm, resulting in unstable license plate recognition results. The acquisition methods for vehicle color, vehicle type, and vehicle appearance re-identification features are the same as the corresponding parts in S2, and will not be repeated here. The acquisition time is the timestamp of the mainline surveillance camera capturing the vehicle image, accurate to milliseconds.

[0082] The node information of the current main road surveillance camera is read from the network topology of the surveillance camera. The node information includes the surveillance camera identifier and a list of upstream and downstream connections. The list of upstream and downstream connections is traversed, and connections with the connection direction upstream are filtered. The upstream surveillance camera identifiers corresponding to all upstream connections are extracted to form a candidate upstream surveillance set. The candidate upstream surveillance set includes at least one upstream surveillance camera identifier.

[0083] The process iterates through each vehicle trajectory record in the vehicle trajectory set, reading the trajectory termination time and the corresponding monitoring point sequence. The last element of the monitoring point sequence contains the identifier of the last associated monitoring camera; this identifier is extracted. The process checks if the identifier of the last associated monitoring camera belongs to the candidate upstream monitoring set. If it does, the subsequent filtering continues; otherwise, the current vehicle trajectory record is skipped. The time difference between the current main road monitoring camera's acquisition time and the trajectory termination time in the vehicle trajectory record is calculated. The time window parameter between the last associated monitoring camera and the current main road monitoring camera is queried from the monitoring camera network topology. The time difference is compared with the time window parameter. If the time difference is less than or equal to the time window parameter, the current vehicle trajectory record is added to the candidate trajectory set; if the time difference is greater than the time window parameter, the current vehicle trajectory record is skipped. The last associated monitoring camera is determined by its identifier. After the iteration is complete, a candidate trajectory set is obtained, containing at least one vehicle trajectory record that meets the filtering criteria (i.e., a candidate record).

[0084] For each candidate trajectory in the candidate trajectory set, read the vehicle attribute information of the candidate trajectory. The vehicle attribute information includes color attribute, type attribute, and appearance re-identification feature attribute.

[0085] Calculate appearance similarity S appearance Appearance similarity is calculated using a vector distance metric to determine the similarity between the vehicle appearance re-identification feature vector in the current vehicle's visual features and the vehicle appearance re-identification feature vector in the vehicle attribute information corresponding to the candidate trajectory. Specifically, cosine similarity can be used to calculate the similarity between the vehicle appearance re-identification feature vector in the current vehicle's visual features and the vehicle appearance re-identification feature vector in the vehicle attribute information corresponding to the candidate trajectory. The appearance similarity value ranges from -1.0 to 1.0.

[0086] Query the preset color similarity matrix to obtain the color similarity S colorThe preset color similarity matrix is ​​a manually calibrated two-dimensional matrix. Rows and columns represent color categories, and element values ​​indicate the similarity score between two colors. Examples of element values ​​from the preset color similarity matrix are given below: white and silver have a similarity score of 0.80, gray and black have a similarity score of 0.70, blue and green have a similarity score of 0.70, black and brown have a similarity score of 0.60, white and black have a similarity score of -1.0, black and yellow have a similarity score of -0.30, white and purple have a similarity score of -0.40, the same color has a similarity score of 1.0, and an unknown color has a similarity score of 0.20 with other colors. Assuming the vehicle color in the current vehicle's visual features is black, and the color attribute in the candidate trajectory's vehicle attribute information is brown, querying the color similarity matrix yields the color similarity score S. color It is 0.6.

[0087] The type similarity S is obtained by querying the preset vehicle type similarity matrix. type The preset vehicle type similarity matrix is ​​a manually calibrated two-dimensional matrix. Rows and columns represent vehicle types, and element values ​​indicate the similarity score between two vehicle types. Examples of element values ​​from the preset vehicle type similarity matrix are given below: Sedan and SUV pairing similarity is 0.85; SUV and van pairing similarity is 0.80; truck and heavy truck pairing similarity is 0.85; SUV and SUV pairing similarity is 0.70; sedan and truck pairing similarity is -0.80; sedan and heavy truck pairing similarity is -1.0; van and heavy truck pairing similarity is -1.0; same vehicle type pairing similarity is 1.0; unknown vehicle type pairing similarity is 0.20. Assuming the current vehicle's visual features indicate a sedan type, and the candidate trajectory's corresponding vehicle attribute information indicates an SUV type, querying the vehicle type similarity matrix yields the type similarity S. type It is 0.85.

[0088] The appearance similarity weight w is determined based on the highway scene conditions. appearance Color similarity weight w color and type similarity weight w type In this embodiment, taking lighting conditions as an example, the lighting conditions are divided into three states: daytime, nighttime, and tunnel. The corresponding weights are shown below: In the daytime state, w appearance = 0.5, w color =0.3, w type =0.2. At night, the reliability of color recognition decreases, w appearance =0.7, w color =0.1, w type =0.2. In tunnel mode, w appearance=0.6, w color =0.2, w type =0.2. The weighting coefficients satisfy w appearance +w color +w type =1.0.

[0089] The similarity scores of appearance, color, and type are weighted and fused to obtain the comprehensive matching result S. final The weighted fusion formula is S. final =w appearance ×S appearance +w color ×S color +w type ×S type The comprehensive matching result ranges from -1.0 to 1.0. The higher the value, the greater the probability that the current vehicle's visual features and the candidate trajectory belong to the same vehicle.

[0090] The candidate trajectory set is traversed, and a comprehensive matching result is calculated for each candidate trajectory. The comprehensive matching results of all candidate trajectories are compared, and the candidate trajectory with the largest comprehensive matching result value is selected as the first target trajectory. If the maximum value of the comprehensive matching result is less than the preset matching threshold of 0.5, it is considered that there is no matching target, the current vehicle's visual features correspond to a newly entered vehicle on the main road, and no further addition operations are performed.

[0091] Read the monitoring point sequence of the first target trajectory, which is a list structure. Create a new monitoring point element, which includes the identifier of the current main road monitoring camera and its corresponding acquisition time. Append the new monitoring point element to the end of the monitoring point sequence list. Update the trajectory termination time field of the first target trajectory, setting it to the acquisition time corresponding to the current main road monitoring camera. Update the vehicle attribute information field of the first target trajectory, updating the color attribute to the vehicle color in the current vehicle visual features, the type attribute to the vehicle type in the current vehicle visual features, and the appearance re-identification feature attribute to the vehicle appearance re-identification feature vector in the current vehicle visual features, thus obtaining the trajectory continuation result, which is a continuation of the current vehicle's driving trajectory.

[0092] Extract the unique trajectory identifier from the trajectory continuation result. Use the unique trajectory identifier as the key to locate the corresponding original vehicle trajectory record from the vehicle trajectory set. Delete the original record in the vehicle trajectory set whose key is the unique trajectory identifier. Use the trajectory continuation result as the new value, with the unique trajectory identifier as the key, and re-insert it into the vehicle trajectory set. After the replacement operation is complete, the trajectory records stored in the vehicle trajectory set now contain the main road monitoring point information.

[0093] In another embodiment of this solution, S4 is specifically implemented as follows: The highway exit surveillance camera collects second vehicle identification information, which includes license plate number, vehicle color, vehicle type, vehicle appearance re-identification features, and the time of acquisition. The vehicle color, vehicle type, and vehicle appearance re-identification features are obtained in the same way as the corresponding parts in S2, and will not be repeated here. The acquisition time is the timestamp of the exit surveillance camera capturing the vehicle image, accurate to milliseconds.

[0094] Extract the license plate number from the second vehicle identification information (hereinafter referred to as the exit license plate number for easy distinction). Traverse each vehicle trajectory record in the vehicle trajectory set, read the license plate number field from the record (hereinafter referred to as the record license plate number for easy distinction), and compare the string content of the exit license plate number with that of the record license plate number. If the strings are identical, the vehicle trajectory record is selected as a candidate closed trajectory. If multiple vehicle trajectory records with the same license plate number exist, further compare the time difference between the trajectory termination time and the collection time in the second vehicle identification information. Select the vehicle trajectory record with the smallest time difference that is within the time window parameter range as the second target trajectory. If no matching vehicle trajectory record is found, it is assumed that the current exit vehicle has not created an entry trajectory in the vehicle trajectory set, and no subsequent closure operation is performed.

[0095] Read the monitoring point sequence field of the second target trajectory. The monitoring point sequence is a list structure. Create a new exit monitoring point element. The new monitoring point element contains the exit monitoring camera identifier and the corresponding acquisition time. The exit monitoring camera identifier uses a unique number assigned when the exit monitoring camera is registered. The format is a string. Append the new monitoring point element to the end of the monitoring point sequence list. Update the trajectory termination time field of the second target trajectory, setting the trajectory termination time to the acquisition time corresponding to the exit monitoring camera identifier. Update the vehicle attribute information field of the second target trajectory. Update the color attribute to the vehicle color in the second vehicle identity information, update the type attribute to the vehicle type in the second vehicle identity information, and update the appearance re-identification feature attribute to the vehicle appearance re-identification feature vector in the second vehicle identity information.

[0096] Create a trajectory status field, with values ​​ranging from "continuous" to "closed". Set the trajectory status field of the second target trajectory to "closed". Create a trajectory integrity identifier field, a boolean type, initially set to False. Set the trajectory integrity identifier field to True, indicating that the second target trajectory includes both entry and exit monitoring points, forming a complete path description. The trajectory closure result includes the updated trajectory unique identifier, license plate number, trajectory start time, trajectory end time, monitoring point sequence, vehicle attribute information, trajectory status, and trajectory integrity identifier.

[0097] Extract the unique trajectory identifier from the trajectory closure result. Use the unique trajectory identifier as the key to locate the corresponding original vehicle trajectory record from the vehicle trajectory set. Delete the original record in the vehicle trajectory set whose key is the unique trajectory identifier. Use the trajectory closure result as the new value, with the unique trajectory identifier as the key, and re-insert it into the vehicle trajectory set. After the replacement operation is complete, the trajectory records stored in the vehicle trajectory set now contain exit monitoring point information, and the trajectory status is closed. The closed trajectory records can be used for subsequent statistical analysis, path reconstruction, and behavior analysis applications.

[0098] After the vehicle trajectory set is updated, records with a closed trajectory status and a trajectory termination time that differs from the current system time by a preset cleanup threshold can be periodically cleaned up. The preset cleanup threshold is set to 3600 seconds to control the storage size of the vehicle trajectory set.

[0099] Furthermore, the position and pose information includes road marker data and driving direction information; Based on the pose information of the surveillance cameras, the network topology of the surveillance cameras is obtained, including: The surveillance cameras are sorted based on road marker data. The surveillance cameras with smaller road marker values ​​are identified as upstream surveillance cameras, and the surveillance cameras with larger road marker values ​​are identified as downstream surveillance cameras. Based on the driving direction information, the monitoring cameras are divided into groups for the uphill direction and groups for the downhill direction. The geographical distance between adjacent surveillance cameras is calculated based on the road station data of adjacent surveillance cameras; time window parameters are calculated based on the geographical distance between surveillance cameras and the standard driving speed on highways.

[0100] Furthermore, vehicle visual features include vehicle color, vehicle type, and vehicle appearance re-identification features. Multi-dimensional similarity fusion includes: Calculate the appearance similarity between the vehicle appearance re-identification features in the current vehicle visual features and the vehicle appearance re-identification features in the vehicle attribute information corresponding to the candidate trajectory; Query the preset color similarity matrix to obtain color similarity, and query the preset vehicle type similarity matrix to obtain type similarity; Based on the highway scene conditions, the appearance similarity weight, color similarity weight, and type similarity weight are determined. The appearance similarity, color similarity, and type similarity are then weighted and fused to obtain a comprehensive matching result.

[0101] Furthermore, it also includes: Receive vehicle speed data reported by the monitoring video analysis equipment, statistically analyze the vehicle speed data within a preset time period, and obtain the average vehicle speed. The time window parameters are adjusted based on the comparison between the average vehicle speed and the preset congestion threshold. These adjusted parameters are then applied to the selection process of the candidate trajectory set obtained from the vehicle trajectory set. Specifically: Figure 3 This is a schematic diagram illustrating the dynamic adjustment of time window parameters based on vehicle speed, as shown below. Figure 3 As shown, the system receives vehicle speed data reported by the monitoring video analysis device. The vehicle speed data includes instantaneous speed values ​​and the collection time. The instantaneous speed values ​​are calculated by the monitoring video analysis device from the video stream after detecting the vehicle, and the unit is kilometers per hour. The collection time is the timestamp when the vehicle is detected.

[0102] The average vehicle speed is obtained by statistically analyzing vehicle speed data within a preset time period. The preset time period can be set to 300 seconds. A sliding time window is maintained, and all vehicle speed data whose data reporting time falls within the sliding time window are filtered. The instantaneous speed values ​​are summed and divided by the number of instantaneous speed values ​​to obtain the average vehicle speed.

[0103] The time window parameters are adjusted based on a comparison between the average vehicle speed and a preset congestion threshold. The preset congestion threshold can be set to 30 km / h, and the preset free-flow threshold can be set to 80 km / h. If the average vehicle speed is less than the preset congestion threshold, the traffic condition is determined to be congested, and the time window parameters are multiplied by a congestion amplification factor of 1.5 to obtain the adjusted time window parameters. If the average vehicle speed is greater than the preset free-flow threshold, the traffic condition is determined to be free-flowing, and the time window parameters are multiplied by a free-flow reduction factor of 0.8 to obtain the adjusted time window parameters. If the average vehicle speed is between the preset congestion threshold and the preset free-flow threshold, the adjusted time window parameters are equal to the original time window parameters.

[0104] The adjusted time window parameter is applied to the selection process of obtaining the candidate trajectory set from the vehicle trajectory set. During the selection process, the adjusted time window parameter is read as a time constraint. The time difference between the trajectory termination time of the vehicle trajectory record and the current acquisition time of the main road monitoring camera is compared. If the time difference is less than or equal to the adjusted time window parameter, the vehicle trajectory record is added to the candidate trajectory set.

[0105] The beneficial effects are as follows: By integrating road topology relationships, candidate selection mechanisms, and attribute matching strategies, stable continuity of vehicle trajectories across cameras can be achieved even when identity anchors are lacking on main road sections. In situations with dense traffic and numerous consecutive appearances of similar vehicles, multi-attribute optimization, topological constraints, and time window pruning effectively reduce mismatches and maintain trajectory consistency. Limiting the candidate range through topological structure and temporal constraints avoids candidate space explosion, improves computational efficiency, and is suitable for large-scale high-speed camera systems. It does not rely on license plate recognition capabilities on main road sections, maintaining stability of cross-camera associations even when license plates are missing or recognition is unstable.

[0106] In the above embodiments, although the steps are numbered S1, S2, etc., they are only specific embodiments given by the present invention. Those skilled in the art can adjust the execution order of S1, S2, etc. according to the actual situation, and these situations are also within the protection scope of the present invention. It can be understood that in some embodiments, some or all of the above embodiments may be included.

[0107] The present invention also provides a highway cross-camera vehicle trajectory reconstruction system, the specific technical solution of which is as follows: a topology construction module, a trajectory creation module, a trajectory continuation module, and a trajectory closure module; The topology building module is used to obtain the network topology of the surveillance cameras based on their pose information. The network topology of the surveillance cameras includes the upstream and downstream connections between the surveillance cameras and time window parameters. The trajectory creation module is used to obtain the initial vehicle driving trajectory based on the first vehicle identity information collected by the highway entrance monitoring camera, and add the initial vehicle driving trajectory to the vehicle driving trajectory set; the first vehicle identity information includes the license plate number, the initial vehicle driving trajectory includes the monitoring point sequence and vehicle attribute information, and the vehicle driving trajectory set includes the driving trajectories of all vehicles whose trajectories have been reconstructed. The trajectory continuation module receives vehicle visual features collected by highway mainline monitoring cameras (excluding license plate numbers). Based on the upstream and downstream connections in the camera network topology, it determines the candidate upstream monitoring set corresponding to the current mainline monitoring camera. Based on time window parameters and the candidate upstream monitoring set, it filters from the vehicle trajectory set to obtain a candidate trajectory set. It performs multi-dimensional similarity fusion between the vehicle visual features and the vehicle attribute information corresponding to each candidate trajectory in the candidate trajectory set to obtain a comprehensive matching result. Based on the comprehensive matching result, it determines the first target trajectory from the candidate trajectory set, appends the current mainline monitoring point to the monitoring point sequence of the first target trajectory, and obtains the trajectory continuation result. Based on the trajectory continuation result, it replaces the corresponding original vehicle trajectory record in the vehicle trajectory set and updates the vehicle trajectory set. The trajectory closure module is used to determine the second target trajectory from the vehicle trajectory set based on the license plate number in the second vehicle identity information collected by the highway exit monitoring camera, add the exit monitoring point to the monitoring point sequence of the second target trajectory, obtain the trajectory closure result, replace the corresponding original vehicle trajectory record in the vehicle trajectory set based on the trajectory closure result, and update the vehicle trajectory set.

[0108] It should be noted that the beneficial effects of the highway cross-camera vehicle trajectory reconstruction system provided in the above embodiments are the same as those of the highway cross-camera vehicle trajectory reconstruction method described above, and will not be repeated here. Furthermore, the system provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the system can be divided into different functional modules according to the actual situation to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, and will not be repeated here.

[0109] like Figure 4 As shown, an embodiment of the present invention provides a computer device 300, which includes a processor 320 coupled to a memory 310. The memory 310 stores at least one computer program 330, which is loaded and executed by the processor 320 to enable the computer device 300 to implement any of the above-described methods. Specifically: The computer device 300 can vary considerably due to differences in configuration or performance. It may include one or more processors 320 (Central Processing Units, CPUs) and one or more memories 310. The one or more memories 310 store at least one computer program 330, which is loaded and executed by the one or more processors 320 to enable the computer device 300 to implement the highway cross-camera vehicle trajectory reconstruction method provided in the above embodiment. Of course, the computer device 300 may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. The computer device 300 may also include other components for implementing device functions, which will not be elaborated here.

[0110] An embodiment of the present invention provides a computer-readable storage medium storing at least one computer program, which is loaded and executed by a processor to enable a computer to implement any of the above-described methods.

[0111] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0112] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform any of the above-described methods for reconstructing vehicle trajectories across highway cameras.

[0113] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0114] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this disclosure can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, the invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0115] Any combination of one or more computer-readable media may be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.

[0116] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for reconstructing vehicle trajectories across cameras on highways, characterized in that, include: S1. Based on the pose information of the surveillance cameras, obtain the network topology of the surveillance cameras. The network topology of the surveillance cameras includes the upstream and downstream connection relationships between the surveillance cameras and the time window parameters. S2, Based on the first vehicle identity information collected by the highway entrance monitoring camera, an initial vehicle driving trajectory is obtained, and the initial vehicle driving trajectory is added to the vehicle driving trajectory set; the first vehicle identity information includes the license plate number, the initial vehicle driving trajectory includes the monitoring point sequence and vehicle attribute information, and the vehicle driving trajectory set includes the driving trajectories of all vehicles whose trajectories have been reconstructed. S3, receive vehicle visual features collected by the highway main road monitoring camera, the vehicle visual features do not include license plate numbers; determine the candidate upstream monitoring set corresponding to the current main road monitoring camera based on the upstream and downstream connection relationship in the monitoring camera network topology, and filter the candidate trajectory set from the vehicle driving trajectory set based on the time window parameter and the candidate upstream monitoring set. Based on the vehicle's visual features and the vehicle attribute information corresponding to each candidate trajectory in the candidate trajectory set, a multi-dimensional similarity fusion is performed to obtain a comprehensive matching result; Based on the comprehensive matching results, a first target trajectory is determined from the candidate trajectory set, and the current main road monitoring point is added to the monitoring point sequence of the first target trajectory to obtain the trajectory continuation result; Replace the corresponding original vehicle driving trajectory record in the vehicle driving trajectory set based on the trajectory continuation result, and update the vehicle driving trajectory set; S4. Based on the second vehicle identity information collected by the highway exit monitoring camera, determine the second target trajectory from the vehicle trajectory set based on the license plate number in the second vehicle identity information, add the exit monitoring point to the monitoring point sequence of the second target trajectory, obtain the trajectory closure result, replace the corresponding original vehicle trajectory record in the vehicle trajectory set based on the trajectory closure result, and update the vehicle trajectory set.

2. The method for reconstructing vehicle trajectories across cameras on highways according to claim 1, characterized in that, The pose information includes road marker data and driving direction information; The step of obtaining the network topology of the surveillance cameras based on their pose information includes: The surveillance cameras are sorted based on road marker data. The surveillance cameras with smaller road marker values ​​are identified as upstream surveillance cameras, and the surveillance cameras with larger road marker values ​​are identified as downstream surveillance cameras. Based on the driving direction information, the monitoring cameras are divided into groups for the uphill direction and groups for the downhill direction. The geographical distance between adjacent surveillance cameras is calculated based on the road station data of adjacent surveillance cameras; time window parameters are calculated based on the geographical distance between surveillance cameras and the standard driving speed on highways.

3. The method for reconstructing vehicle trajectories across cameras on highways according to claim 1, characterized in that, The vehicle visual features include vehicle color, vehicle type, and vehicle appearance re-identification features. The multi-dimensional similarity fusion includes: Calculate the appearance similarity between the vehicle appearance re-identification features in the current vehicle visual features and the vehicle appearance re-identification features in the vehicle attribute information corresponding to the candidate trajectory; Query the preset color similarity matrix to obtain color similarity, and query the preset vehicle type similarity matrix to obtain type similarity; Based on the highway scene conditions, the appearance similarity weight, color similarity weight, and type similarity weight are determined. The appearance similarity, color similarity, and type similarity are then weighted and fused to obtain a comprehensive matching result.

4. The method for reconstructing vehicle trajectories across cameras on highways according to claim 1, characterized in that, Also includes: Receive vehicle speed data reported by the monitoring video analysis equipment, statistically analyze the vehicle speed data within a preset time period, and obtain the average vehicle speed. The time window parameters are adjusted based on the comparison between the average vehicle speed and the preset congestion threshold, and the adjusted time window parameters are applied to the screening process of obtaining a candidate trajectory set from the set of vehicle driving trajectories.

5. A highway cross-camera vehicle trajectory reconstruction system, characterized in that, include: Topology building module, trajectory creation module, trajectory continuation module, and trajectory closure module; The topology construction module is used to obtain the network topology relationship of the surveillance cameras based on the pose information of the surveillance cameras. The network topology relationship of the surveillance cameras includes the upstream and downstream connection relationship between the surveillance cameras and the time window parameter. The trajectory creation module is used to obtain an initial vehicle driving trajectory based on the first vehicle identity information collected by the highway entrance monitoring camera, and add the initial vehicle driving trajectory to the vehicle driving trajectory set; the first vehicle identity information includes the license plate number, the initial vehicle driving trajectory includes the monitoring point sequence and vehicle attribute information, and the vehicle driving trajectory set includes the driving trajectories of all vehicles whose trajectories have been reconstructed. The trajectory continuation module is used to receive vehicle visual features collected by the highway main road monitoring camera, and the vehicle visual features do not include license plate numbers. Based on the upstream and downstream connection relationships in the network topology of the surveillance cameras, the candidate upstream monitoring set corresponding to the current main road surveillance camera is determined, and a candidate trajectory set is obtained by filtering from the vehicle trajectory set based on the time window parameter and the candidate upstream monitoring set. Based on the vehicle's visual features and the vehicle attribute information corresponding to each candidate trajectory in the candidate trajectory set, a multi-dimensional similarity fusion is performed to obtain a comprehensive matching result; Based on the comprehensive matching results, a first target trajectory is determined from the candidate trajectory set, and the current main road monitoring point is added to the monitoring point sequence of the first target trajectory to obtain the trajectory continuation result; Replace the corresponding original vehicle driving trajectory record in the vehicle driving trajectory set based on the trajectory continuation result, and update the vehicle driving trajectory set; The trajectory closure module is used to determine a second target trajectory from the vehicle trajectory set based on the license plate number in the second vehicle identity information collected by the highway exit monitoring camera, add the exit monitoring point to the monitoring point sequence of the second target trajectory, obtain the trajectory closure result, replace the corresponding original vehicle trajectory record in the vehicle trajectory set based on the trajectory closure result, and update the vehicle trajectory set.

6. A highway cross-camera vehicle trajectory reconstruction system according to claim 5, characterized in that, The pose information includes road marker data and driving direction information; The step of obtaining the network topology of the surveillance cameras based on their pose information includes: The surveillance cameras are sorted based on road marker data. The surveillance cameras with smaller road marker values ​​are identified as upstream surveillance cameras, and the surveillance cameras with larger road marker values ​​are identified as downstream surveillance cameras. Based on the driving direction information, the monitoring cameras are divided into groups for the uphill direction and groups for the downhill direction. The geographical distance between adjacent surveillance cameras is calculated based on the road station data of adjacent surveillance cameras; time window parameters are calculated based on the geographical distance between surveillance cameras and the standard driving speed on highways.

7. A highway cross-camera vehicle trajectory reconstruction system according to claim 5, characterized in that, The vehicle visual features include vehicle color, vehicle type, and vehicle appearance re-identification features. The multi-dimensional similarity fusion includes: Calculate the appearance similarity between the vehicle appearance re-identification features in the current vehicle visual features and the vehicle appearance re-identification features in the vehicle attribute information corresponding to the candidate trajectory; Query the preset color similarity matrix to obtain color similarity, and query the preset vehicle type similarity matrix to obtain type similarity; Based on the highway scene conditions, the appearance similarity weight, color similarity weight, and type similarity weight are determined. The appearance similarity, color similarity, and type similarity are then weighted and fused to obtain a comprehensive matching result.

8. A highway cross-camera vehicle trajectory reconstruction system according to claim 5, characterized in that, Also includes: Receive vehicle speed data reported by the monitoring video analysis equipment, statistically analyze the vehicle speed data within a preset time period, and obtain the average vehicle speed. The time window parameters are adjusted based on the comparison between the average vehicle speed and the preset congestion threshold, and the adjusted time window parameters are applied to the screening process of obtaining a candidate trajectory set from the set of vehicle driving trajectories.

9. A computer device, characterized in that, The computer device includes a processor coupled to a memory storing at least one computer program, which is loaded and executed by the processor to enable the computer device to implement a method for reconstructing vehicle trajectories across cameras on a highway as described in any one of claims 1 to 4.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to enable the computer to implement a method for reconstructing vehicle trajectories across cameras on a highway as described in any one of claims 1 to 4.

Images