Real-time detection and management and control system and method for three-dimensional contour of road large vehicle

Abstract

The invention provides a road large vehicle three-dimensional contour real-time detection and management and control system and method, the system comprises a three-dimensional sensing module, an image acquisition module, an edge calculation unit and a management and control platform, point clouds are obtained through a laser radar, and quaternion leveling, segmented voxel downsampling and DBSCAN clustering are sequentially executed to directly calculate the length, width, height and wheelbase of a vehicle; then, multi-frame target tracking is achieved in combination with Hungary matching-Kalman filtering, vehicle type and axle information is synchronously output from a point cloud projection gray depth map through a neural network, then structured vehicle data is generated by fusing license plate numbers according to timestamps, graded alarm is triggered by dynamically comparing a large-piece transportation record size library, and over-limit early warning with low misjudgment is achieved. All-weather stable operation is achieved, and the precision, the timeliness and the coverage range of overload control of the large vehicle are integrally and remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of vehicle traffic monitoring equipment technology, and more specifically, to a system and method for real-time detection and control of the three-dimensional contours of large highway vehicles. Background Technology

[0002] With the continuous growth in demand for special transportation on highways, the inter-provincial circulation of large structural components (such as wind turbine blades, bridge sections, and aerial work platforms) is becoming increasingly frequent. Oversized transport vehicles, due to their size and weight far exceeding that of conventional freight vehicles, pose a significantly increased safety threat to bridges, tunnels, gantries, and other facilities when traveling on main highways at high speeds. If not identified and warned in a timely manner, they can easily lead to structural collisions, traffic congestion, and even serious accidents. Therefore, there is an urgent need to establish a high-precision, wide-coverage, and end-to-end mechanism for the detection and management of oversized vehicles. Currently, vehicle detection systems on highways are mostly deployed in low-speed areas such as tollbooth entrances, exits, or service areas, typically relying on single or combined methods such as video surveillance, laser height measurement, and weighbridge systems to achieve vehicle type identification and verification.

[0003] The identification system closest to existing technology, application publication number: CN119028035A, discloses a highway toll auditing method and system based on video and lidar, which includes: analyzing target data obtained by preprocessing vehicle data collected using a pre-set data acquisition tool, obtaining license plate recognition results and vehicle type detection results for vehicles entering and exiting toll stations; and combining and analyzing the license plate recognition results and vehicle type detection results to complete the entry and exit audit of toll stations.

[0004] The system's identification is limited to vehicle type classification and license plate matching; it does not acquire the vehicle's actual three-dimensional dimensions (length, width, height) or cargo configuration information. Furthermore, it can only determine whether the vehicle type matches the license plate, but cannot determine whether the vehicle exceeds the permitted transport dimensions.

[0005] To address the aforementioned technical shortcomings, there is an urgent need to propose a method that can acquire the three-dimensional dimensions of vehicles in real time in high-speed scenarios and perform dual threshold judgments, thereby achieving accurate identification, graded alarms, and closed-loop management of oversized vehicles. Summary of the Invention

[0006] The purpose of this invention is to provide a vehicle model recognition device that enables real-time detection and accurate identification of three-dimensional contours.

[0007] Firstly, this application proposes a real-time detection and control system for the three-dimensional contours of heavy-duty highway vehicles, including:

[0008] A 3D perception module is installed on the mainline gantry of the highway to acquire 3D point clouds of vehicles passing through the gantry.

[0009] An image acquisition module, deployed in conjunction with the three-dimensional perception module, is used to acquire images of the license plate or exterior of the target vehicle.

[0010] The edge computing unit is communicatively connected to both the 3D perception module and the image acquisition module. The edge computing unit is used to: sequentially perform quaternion flattening, lane line filtering, ground point segmentation, and distance segmentation adaptive downsampling on a single frame point cloud; extract vehicle point clusters using a density clustering algorithm and directly calculate the vehicle's length, width, height, and wheelbase; overlay continuous frame point clouds and use Hungarian bipartite graph matching and Kalman filtering for multi-target tracking to update vehicle status; project the point cloud into a 2D grayscale depth map and output wheel axle information and vehicle type category through an axle type classification network and a vehicle type classification network; and fuse the vehicle's 3D dimensions, vehicle type category, and license plate information according to timestamps to generate structured vehicle data.

[0011] The control platform, which is connected to the edge computing unit, is used to: compare the structured vehicle data with the oversized transport registration database using a dual-threshold size comparison, and manually review the results for car carriers; trigger tiered warnings for vehicles determined to be oversized or suspected of being oversized, and link the law enforcement terminal and the toll collection or toll collection correction system to complete closed-loop control.

[0012] In some implementations, the three-dimensional perception module is a lidar, and the image acquisition module is a license plate recognition camera. The lidar and license plate recognition camera are installed at a height of 6m-7m with a pitch angle of 15°±2°, and a single system covers 4+1 lanes.

[0013] In some implementations, the quaternion flattening is performed by fitting the ground plane with RANSAC, calculating Euler angles, and constructing quaternions to complete tilt compensation.

[0014] In some implementations, the distance segmented adaptive downsampling uses voxel sizes of 0.3m, 0.2m, and 0.1m in the ranges of 0-30m, 30-40m, and 40-50m, respectively, and downsampling is not performed beyond 50m.

[0015] In some implementations, the density clustering algorithm uses DBSCAN, with a neighborhood radius ε = 1m and a minimum number of points minPts = 5.

[0016] In some implementations, the multi-target tracking uses a preset loss count threshold when a target mismatch occurs. Once the threshold is reached, tracking is terminated and historical trajectories are output.

[0017] In some implementations, the axle type classification network uses a lightweight convolutional neural network to output the number of axles and wheelbase information, and the vehicle type classification network uses an improved ResNet structure to output a vehicle type category.

[0018] In some implementations, when the control platform triggers an early warning, it will activate a wide-angle camera to capture the target vehicle and simultaneously upload the alarm information to the traffic police control platform and the ETC system.

[0019] In some implementations, the lidar has an IP69K protection rating and supports an operating temperature range of -40°C to 85°C.

[0020] Secondly, this application proposes a method for real-time detection and control of the three-dimensional contour of large vehicles on highway mainline gantry, including the following steps:

[0021] S1: Data acquisition and database loading: raw point cloud frames are acquired using LiDAR; corresponding license plate images are acquired using license plate recognition cameras; the oversized transport registration database and its size threshold table are loaded and cached from the management platform; the above data are uniformly marked with the same timestamp.

[0022] S2: Point cloud preprocessing: Quaternion flattening, lane line filtering, ground point segmentation, and distance segmentation adaptive voxel downsampling are sequentially performed on the original point cloud frame to obtain the preprocessed point cloud of the current frame;

[0023] S3: Single-frame target extraction: The preprocessed point cloud obtained in step S2 is processed using a density clustering algorithm to extract vehicle point clusters, and the length, width, height and wheelbase of each point cluster are calculated; the obtained vehicle dimensions and their timestamps are cached in the trajectory queue;

[0024] S4: Multi-frame target tracking. Repeat steps S1 to S3 to form continuous multi-frame vehicle point cluster data. Cache the multi-frame vehicle point cluster data and use Hungarian bipartite graph matching combined with Kalman filtering to perform multi-target tracking of vehicle targets and update vehicle trajectory and speed status.

[0025] S5: Vehicle type and axle type recognition: For the preprocessed point cloud obtained in step S2, project it into a two-dimensional grayscale depth map in the current frame; input it into the axle type classification network and the vehicle type classification network respectively to obtain the number of vehicle axles, wheelbase information and a type of traffic vehicle;

[0026] S6: Data Fusion: The information obtained in steps S3, S4, and S5, based on the timestamp association, is fused with the license plate image recognition results to generate structured vehicle data containing vehicle 3D dimensions, vehicle type, wheel axle information, and license plate number.

[0027] S7: Size Comparison and Closed-Loop Management: The structured vehicle data generated in step S6 is compared with the oversized transport registration database loaded in step S1 using a dual-threshold size comparison, and the results for car carriers are manually reviewed; when a vehicle is determined to be oversized or suspected of being oversized, a graded warning is triggered and linked with wide-angle capture, road administration enforcement and ETC toll collection systems to form a closed-loop management record and upload it to the traffic police management platform.

[0028] In some implementations, the distance segmented adaptive downsampling in step S2 uses 0.3m×0.3m×0.3m cubic voxels for the distance segment from 0m to 30m; 0.2m×0.2m×0.2m cubic voxels for the distance segment from 30m to 40m; 0.1m×0.1m×0.1m cubic voxels for the distance segment from 40m to 50m; and no downsampling is performed for the distance segment greater than 50m.

[0029] In some implementations, the density clustering in step S3 uses the DBSCAN algorithm, with a neighborhood radius ε of 1m and a minimum number of points minPts of 5.

[0030] In some implementations, the multi-target tracking in step S4 terminates after the number of consecutive target mismatches reaches a preset loss count threshold.

[0031] In some implementations, the structured vehicle data generated in step S6 is uploaded to the traffic police management platform in real time and stored in a local log file for traceability.

[0032] In some implementations, step S7 further includes reading the vehicle length, width, height, and wheelbase from the structured vehicle data generated in step S6; calling the oversized transport registration database loaded in step S1 to retrieve the registration dimensions of the corresponding license plate; calculating the percentage of any vehicle dimension exceeding the registration value relative to the registered value, Δ = |(actual size − registered size)| / registered size × 100%; and uploading the warning level, handling measures, and structured vehicle data to the traffic police management platform to form a traceable closed-loop record.

[0033] The following grading criteria will trigger the warning:

[0034] When 0 < Δ ≤ 5%, a Level 1 warning, i.e. a yellow warning, is triggered, and the warning information is uploaded and pushed to the road administration backend for investigation.

[0035] When Δ > 5%, a Level II warning, i.e. a red warning, is triggered, and wide-angle capture, road administration enforcement terminals and ETC toll collection systems are linked to carry out on-site handling; for cases identified as car carriers and where the excess dimension is the height of the car body, an additional manual review process is initiated.

[0036] The beneficial effects of this invention are as follows: This invention proposes a real-time detection and control system and method for the three-dimensional contours of oversized vehicles on highways, including a three-dimensional perception module, an image acquisition module, an edge computing unit, and a control platform. It acquires point clouds using lidar and sequentially performs quaternion flattening, segmented voxel downsampling, and DBSCAN clustering to directly calculate the vehicle's length, width, height, and wheelbase. Then, it combines Hungarian matching-Kalman filtering to achieve multi-frame target tracking. The point cloud projection grayscale depth map is synchronously output as vehicle type and axle information via a neural network. Subsequently, it fuses the license plate number with timestamps to generate structured vehicle data. By dynamically comparing with the oversized transport registration size database, it triggers graded alarms, achieving low-false-judgment over-limit warnings. It operates stably around the clock, significantly improving the accuracy, timeliness, and coverage of oversized vehicle control. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the structure of a real-time detection and control system for the three-dimensional contours of heavy vehicles on highways, as described in this application.

[0038] Figure 2 This is a flowchart illustrating a method for real-time detection and control of the three-dimensional contours of heavy-duty highway vehicles according to this application.

[0039] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0040] The following embodiments are described to aid in understanding this application. These embodiments are not, and should not be construed as, limiting the scope of protection of this application.

[0041] In the following description, those skilled in the art will recognize that throughout this discussion, components may be described as individual functional units (which may include subunits), but those skilled in the art will recognize that various components or portions thereof may be divided into individual components or may be integrated together (including integrated within a single system or component).

[0042] Furthermore, the connection between components or systems is not intended to be limited to a direct connection; on the contrary, data between these components may be modified, reformatted, or otherwise altered by intermediate components. Additionally, other or fewer connections may be used. It should also be noted that the terms "connection," "link," or "input" should be understood to include direct connections, indirect connections via one or more intermediate devices, and wireless connections. Example 1:

[0043] like Figure 1 The diagram shown is a structural schematic of a real-time three-dimensional contour detection and control system for heavy-duty highway vehicles according to this application; Figure 2 The diagram shown is a flowchart illustrating a method for real-time detection and control of the three-dimensional contours of heavy-duty vehicles for highways, as proposed in this application.

[0044] A real-time 3D contour detection and control system for heavy-duty highway vehicles includes:

[0045] A 3D perception module is installed on the mainline gantry of the highway to acquire 3D point clouds of vehicles passing through the gantry.

[0046] An image acquisition module, deployed in conjunction with the three-dimensional perception module, is used to acquire images of the license plate or exterior of the target vehicle.

[0047] The edge computing unit is communicatively connected to both the 3D perception module and the image acquisition module. The edge computing unit is used to: sequentially perform quaternion flattening, lane line filtering, ground point segmentation, and distance segmentation adaptive downsampling on a single frame point cloud; extract vehicle point clusters using a density clustering algorithm and directly calculate the vehicle's length, width, height, and wheelbase; overlay continuous frame point clouds and use Hungarian bipartite graph matching and Kalman filtering for multi-target tracking to update vehicle status; project the point cloud into a 2D grayscale depth map and output wheel axle information and vehicle type category through an axle type classification network and a vehicle type classification network; and fuse the vehicle's 3D dimensions, vehicle type category, and license plate information according to timestamps to generate structured vehicle data.

[0048] The control platform, which is connected to the edge computing unit, is used to: compare the structured vehicle data with the oversized transport registration database using a dual-threshold size comparison, and manually review the results for car carriers; trigger tiered warnings for vehicles determined to be oversized or suspected of being oversized, and link the law enforcement terminal and the toll collection or toll collection correction system to complete closed-loop control.

[0049] The three-dimensional perception module is a lidar, and the image acquisition module is a license plate recognition camera. The lidar and license plate recognition camera are installed at a height of 6m-7m with a pitch angle of 15°±2°. A single system covers 4+1 lanes.

[0050] The quaternion flattening is achieved by fitting the ground plane with RANSAC, calculating Euler angles, and constructing quaternions to complete tilt compensation.

[0051] The distance segmented adaptive downsampling uses voxel sizes of 0.3m, 0.2m, and 0.1m in the 0-30m, 30-40m, and 40-50m ranges, respectively, and does not perform downsampling beyond 50m.

[0052] The density clustering algorithm used is DBSCAN, with a neighborhood radius of ε = 1m and a minimum number of points minPts = 5.

[0053] The multi-target tracking uses a preset loss count threshold when a target mismatch occurs. Once the threshold is reached, the tracking is terminated and the historical trajectory is output.

[0054] The axle type classification network uses a lightweight convolutional neural network to output the number of axles and wheelbase information, while the vehicle type classification network uses an improved ResNet structure to output a vehicle type category.

[0055] When an alert is triggered, the control platform will activate a wide-angle camera to capture the target vehicle and simultaneously upload the alarm information to the traffic police control platform and the ETC system.

[0056] The lidar has an IP69K protection rating and supports an operating temperature range of -40℃ to 85℃.

[0057] A method for real-time detection and control of the three-dimensional contour of large vehicles on highway mainline gantry, comprising the following steps:

[0058] S1: Data acquisition and database loading: raw point cloud frames are acquired using LiDAR; corresponding license plate images are acquired using license plate recognition cameras; the oversized transport registration database and its size threshold table are loaded and cached from the management platform; the above data are uniformly marked with the same timestamp.

[0059] S2: Point cloud preprocessing: Quaternion flattening, lane line filtering, ground point segmentation, and distance segmentation adaptive voxel downsampling are sequentially performed on the original point cloud frame to obtain the preprocessed point cloud of the current frame;

[0060] S3: Single-frame target extraction: The preprocessed point cloud obtained in step S2 is processed using a density clustering algorithm to extract vehicle point clusters, and the length, width, height and wheelbase of each point cluster are calculated; the obtained vehicle dimensions and their timestamps are cached in the trajectory queue;

[0061] S4: Multi-frame target tracking. Repeat steps S1 to S3 to form continuous multi-frame vehicle point cluster data. Cache the multi-frame vehicle point cluster data and use Hungarian bipartite graph matching combined with Kalman filtering to perform multi-target tracking of vehicle targets and update vehicle trajectory and speed status.

[0062] S5: Vehicle type and axle type recognition: For the preprocessed point cloud obtained in step S2, project it into a two-dimensional grayscale depth map in the current frame; input it into the axle type classification network and the vehicle type classification network respectively to obtain the number of vehicle axles, wheelbase information and a type of traffic vehicle;

[0063] S6: Data Fusion: The information obtained in steps S3, S4, and S5, based on the timestamp association, is fused with the license plate image recognition results to generate structured vehicle data containing vehicle 3D dimensions, vehicle type, wheel axle information, and license plate number.

[0064] S7: Size Comparison and Closed-Loop Management: The structured vehicle data generated in step S6 is compared with the oversized transport registration database loaded in step S1 using a dual-threshold size comparison, and the results for car carriers are manually reviewed; when a vehicle is determined to be oversized or suspected of being oversized, a graded warning is triggered and linked with wide-angle capture, road administration enforcement and ETC toll collection systems to form a closed-loop management record and upload it to the traffic police management platform.

[0065] In step S2, distance segmentation adaptive downsampling uses 0.3m×0.3m×0.3m cubic voxels for distances from 0m to 30m; 0.2m×0.2m×0.2m cubic voxels for distances from 30m to 40m; and 0.1m×0.1m×0.1m cubic voxels for distances from 40m to 50m. No downsampling is performed for distances greater than 50m. Step S3 uses the DBSCAN algorithm for density clustering, with a neighborhood radius ε of 1m and a minimum number of points minPts of 5.

[0066] Step S4 of the multi-target tracking terminates after the number of consecutive target mismatches reaches a preset loss count threshold.

[0067] The structured vehicle data generated in step S6 is uploaded to the traffic police management platform in real time and stored in a local log file for traceability.

[0068] S7 also includes reading the vehicle length, width, height and wheelbase from the structured vehicle data generated in step S6; calling the oversized transport registration database loaded in step S1 to retrieve the registration dimensions of the corresponding license plate; calculating the percentage of any vehicle dimension exceeding the registration value Δ, Δ=|(actual size − registration size)| / registration size×100%; and uploading the warning level, handling measures and structured vehicle data to the traffic police management platform to form a traceable closed-loop record.

[0069] The following grading criteria will trigger the warning:

[0070] When 0 < Δ ≤ 5%, a Level 1 warning, i.e. a yellow warning, is triggered, and the warning information is uploaded and pushed to the road administration backend for investigation.

[0071] When Δ > 5%, a Level II warning, i.e. a red warning, is triggered, and wide-angle capture, road administration enforcement terminals and ETC toll collection systems are linked to carry out on-site handling; for cases identified as car carriers and where the excess dimension is the height of the car body, an additional manual review process is initiated.

[0072] The beneficial effects of this invention are as follows: This invention proposes a real-time detection and control system and method for the three-dimensional contours of oversized vehicles on highways, including a three-dimensional perception module, an image acquisition module, an edge computing unit, and a control platform. It acquires point clouds using lidar and sequentially performs quaternion flattening, segmented voxel downsampling, and DBSCAN clustering to directly calculate the vehicle's length, width, height, and wheelbase. Then, it combines Hungarian matching-Kalman filtering to achieve multi-frame target tracking. The point cloud projection grayscale depth map is synchronously output as vehicle type and axle information via a neural network. Subsequently, it fuses the license plate number with timestamps to generate structured vehicle data. By dynamically comparing with the oversized transport registration size database, it triggers graded alarms, achieving low-false-judgment over-limit warnings. It operates stably around the clock, significantly improving the accuracy, timeliness, and coverage of oversized vehicle control.

[0073] Although this application discloses several aspects and embodiments, other aspects and embodiments will be obvious to those skilled in the art. Various modifications and improvements can be made without departing from the concept of this application, and these all fall within the scope of protection of this application. The various aspects and embodiments disclosed in this application are for illustrative purposes only and are not intended to limit this application. The actual scope of protection of this application is determined by the claims.

Claims

1. A highway oversize vehicle three-dimensional profile real-time detection and management system, characterized in that, The application relates to a large vehicle management and control system based on edge computing. The application comprises: a three-dimensional perception module arranged on a highway main line gantry, which is used for acquiring a three-dimensional point cloud of a vehicle passing through the gantry; an image acquisition module cooperatively arranged with the three-dimensional perception module, which is used for acquiring a license plate or appearance image of a target vehicle; an edge computing unit in communication connection with the three-dimensional perception module and the image acquisition module, wherein the edge computing unit is used for: sequentially performing quaternion flattening, lane line filtering, ground point segmentation and distance segmented adaptive downsampling on a single frame point cloud; extracting a vehicle point cluster by using a density clustering algorithm, and directly calculating three-dimensional dimensions such as length, width, height and wheelbase of the vehicle; superimposing continuous frame point clouds, and performing multi-target tracking by using Hungarian bipartite graph matching and Kalman filtering to update the vehicle state; projecting the point cloud into a two-dimensional gray depth map, and outputting wheel shaft information and vehicle type category through an axle type classification network and a vehicle type classification network; fusing the vehicle three-dimensional dimensions, the vehicle type category and the license plate information according to timestamps, and generating structured vehicle data; 2. The highway oversize vehicle three-dimensional profile real-time detection and management system of claim 1, wherein: a management and control platform in communication connection with the edge computing unit, which is used for: comparing the structured vehicle data with a large vehicle transportation record database according to double-threshold dimensions, and manually checking the results of the large vehicle transportation; triggering a hierarchical early warning for a vehicle determined as over-limit or suspected over-limit, and connecting the law enforcement terminal and the toll or toll rectification system to complete closed-loop management and control.

3. The highway oversize vehicle three-dimensional profile real-time detection and management system of claim 1, wherein: The three-dimensional perception module is a laser radar, the image acquisition module is a license plate recognition camera, the installation height of the laser radar and the license plate recognition camera is 6m-7m, the pitch angle is 15 DEG + 2 DEG, and a single set of system covers 4+1 lanes.

4. The highway oversize vehicle three-dimensional profile real-time detection and management system of claim 1, wherein: The quaternion flattening is completed by fitting a ground plane through RANSAC, calculating Euler angles and constructing a quaternion to complete tilt compensation.

5. The highway heavy-duty vehicle three-dimensional profile real-time detection and management system of claim 1, wherein: The distance segmented adaptive downsampling corresponds to 0.3m, 0.2m and 0.1m voxel sizes in 0-30m, 30-40m and 40-50m intervals respectively, and no downsampling is performed when the distance is greater than 50m.

6. The method for real-time detection and management of the three-dimensional profile of an OOV according to any one of claims 1-5, characterized in that, The density clustering algorithm adopts DBSCAN, the domain radius epsilon is 1m, and the minimum point number minPts is 5. The application comprises the following steps: S1: data acquisition and database loading, collecting original point cloud frames by using a laser radar; collecting corresponding license plate images by using a license plate recognition camera; loading and buffering a large vehicle transportation record database and a size threshold table from a management and control platform; and marking the same timestamp for the above data; S2: point cloud preprocessing: sequentially performing quaternion flattening, lane line filtering, ground point segmentation and distance segmented adaptive voxel downsampling on the original point cloud frames to obtain preprocessed point clouds of a current frame; S3: single frame target extraction: extracting a vehicle point cluster by using a density clustering algorithm on the preprocessed point clouds obtained in step S2, and calculating the length, width, height and wheelbase of each point cluster; and buffering the obtained vehicle dimensions and timestamps in a track queue; S4: multi-frame target tracking, repeatedly performing steps S1 to S3 to form continuous multi-frame vehicle point cluster data, buffering the multi-frame vehicle point cluster data, and performing multi-target tracking on the vehicle target by using Hungarian bipartite graph matching combined with Kalman filtering to update the vehicle track and speed state; S5: Vehicle type and axle type recognition, for the pre-processed point cloud obtained in step S2, project it into a two-dimensional gray depth map in the current frame; input the axle type classification network and the vehicle type classification network respectively, obtain the number of vehicle axles, axle distance information and a type of traffic vehicle category; S6: Data fusion, according to the time stamp, associate the information obtained in steps S3, S4, S5, and fuse with the license plate image recognition result, generate structured vehicle data containing vehicle three-dimensional size, vehicle type category, axle information and license plate number; S7: Size comparison and closed-loop control, perform double-threshold size comparison on the structured vehicle data generated in step S6 and the large transport record database loaded in step S1, and perform manual review on the results of the car transport vehicle; when the vehicle is judged to be over-limit or suspected to be over-limit, trigger warning and link wide-angle snapshot, road management law enforcement and ETC charging system, form closed-loop control record and upload to the traffic police control platform.

7. The method of claim 6, wherein the method further comprises: The distance segmentation adaptive downsampling in step S2 uses a cubic voxel of 0.3m x 0.3m x 0.3m for the distance segment of 0m to 30m; uses a cubic voxel of 0.2m x 0.2m x 0.2m for the distance segment of 30m to 40m; uses a cubic voxel of 0.1m x 0.1m x 0.1m for the distance segment of 40m to 50m; and does not perform downsampling for the distance segment greater than 50m.

8. The method of claim 6, wherein the method further comprises: The density clustering of step S3 uses DBSCAN algorithm, the domain radius ε is 1m, and the minimum point number minPts is 5.

9. The method of claim 6, wherein the method further comprises: The multi-target tracking of step S4 terminates tracking after the number of consecutive target mismatches reaches the preset missing count threshold.

10. The method of claim 6, wherein the method further comprises: The structured vehicle data generated in step S6 is uploaded to the traffic police control platform in real time, and stored in the local log file for tracing.

Images