A vehicle-road cooperative perception method and device based on vehicle-mounted point cloud

Abstract

The application discloses a kind of vehicle-road cooperative perception method and device based on vehicle-mounted point cloud, belong to the field of automatic driving, the method includes: using each vehicle in current time slot in driving environment complexity and communication link data rate of cooperation perception, construct the optimization problem of maximum total perception quality, solve it to obtain the point cloud processing task priority and optimal configuration corresponding to each vehicle;Each vehicle uses corresponding optimal configuration to sample compression and back to roadside server, so that roadside server uses point cloud processing task priority to carry out target detection and issue to each vehicle to carry out environmental perception.The optimization problem in the application considers the problem of communication resources, processing priority and the cooperative perception accuracy of multiple vehicles and roadside server, thereby solving the technical problems of poor perception quality caused by large delay, low accuracy and limited resources of existing automatic driving vehicles.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving technology, and more specifically, relates to a vehicle-road cooperative perception method and device based on vehicle point cloud. Background Technology

[0002] Autonomous driving technology heavily relies on onboard sensors to detect the driving environment and corresponding environmental perception algorithms. Point cloud data collected by LiDAR can acquire three-dimensional depth information of the surrounding environment. Simultaneously, with the help of rapidly developing deep learning models, autonomous vehicles can obtain the three-dimensional coordinate information of various targets on the road through point cloud target detection and other algorithms, thus achieving road environment perception. However, point cloud target detection algorithms require high computing power, and currently most vehicles do not possess single-vehicle point cloud target detection capabilities. Offloading point cloud analysis tasks to edge processing through vehicle-to-everything (V2X) technology has become an emerging paradigm for autonomous driving: vehicle-to-infrastructure (V2I) cooperative perception.

[0003] First, multiple vehicles coordinating perception with the same roadside server are often in different road conditions, resulting in varying degrees of urgency for environmental perception among different vehicles. Second, roadside servers have limited computing resources, making prioritization strategies crucial when multiple vehicles are accessing the system. Third, the communication connection between moving vehicles and roadside servers is unstable, affected by factors such as signal transmission power, communication distance, and link obstruction, leading to significant fluctuations in point cloud data transmission latency. All of these factors often result in unsatisfactory overall perception quality for autonomous vehicles. Summary of the Invention

[0004] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a vehicle-road cooperative perception method and device based on vehicle-mounted point clouds. Its purpose is to construct an optimization problem that maximizes the overall perception quality by considering the driving environment complexity and communication link data rate of each participating vehicle in the current time slot. Solving this problem yields the point cloud processing task priorities and optimal configurations for each vehicle. Each vehicle then uses its optimal configuration to perform sampling compression and transmits the data back to a roadside server. This server then uses the point cloud processing task priorities to perform target detection and distribute the data to each vehicle for environmental perception. This solves the technical problems of poor perception quality caused by high latency, low accuracy, and limited resources in existing autonomous vehicles.

[0005] To achieve the above objectives, according to one aspect of the present invention, a vehicle-road cooperative perception method based on vehicle-mounted point cloud is provided, applied to a roadside server, comprising:

[0006] S1: Determine the optimal configuration set for sampling and compression processing of vehicle point cloud frames.

[0007] S2: Calculate the driving environment complexity Ψ(m,t) and communication link data rate of each participating collaborative sensing vehicle in the current time slot based on the point cloud target detection results of each participating collaborative sensing vehicle in the previous time slot.

[0008] S3: Construct an optimization problem to maximize total perceived quality. And solve for the point cloud processing task priority b of the m-th vehicle participating in collaborative perception in the current time slot t. m,t and optimal configuration a m,t ; The optimal configuration a m,t Issued to vehicle m so that it can utilize the optimal configuration a. m,t The redundancy-removed point cloud frames are sampled and compressed to obtain the target point cloud data, which is then transmitted back to the roadside server.

[0009] Where Φ(m,t) represents the perceived quality. The product of the driving environment complexity Ψ(m,t); With optimal configuration a m,t The corresponding perception accuracy is positively correlated with the communication link data rate. The corresponding perception delay is negatively correlated; M is the total number of vehicles participating in collaborative perception.

[0010] S4: Receive the target point cloud data corresponding to each participating collaborative perception vehicle in the current time slot, process each target point cloud data according to the point cloud processing task priority, obtain the point cloud target detection result of the current time slot, and send it to the corresponding vehicle to enable it to perform environmental perception.

[0011] In one embodiment, S1 includes:

[0012] S11: Perform offline training on a large number of point cloud frame samples to obtain the average point cloud compression rate and the average accuracy of point cloud target detection for each optimal configuration.

[0013] S12: Based on the average point cloud compression rate and the average accuracy of point cloud target detection, the Pareto optimality principle is applied. Select multiple optimal configurations to form an optimal configuration set. Each optimal configuration includes: sampling method, sampling level, and quantization bit depth during compression;

[0014] Where 'a' represents any one of the configurations, a * For the selected optimal configuration, R(a) is the average point cloud compression rate corresponding to a. * ) is a * The corresponding average point cloud compression rate, P(a) is the average accuracy of point cloud target detection corresponding to a, P(a) * ) is a* The average accuracy of the corresponding point cloud target detection.

[0015] In one embodiment, S2 includes:

[0016] S21: Calculate the driving environment complexity Ψ(m, t; n) faced by vehicle m in the current time slot t, based on the coordinates and driving direction of vehicle n being inspected. Then, sum all Ψ(m, t; n) to obtain the driving environment complexity Ψ(m, t) corresponding to vehicle m.

[0017] S22: Obtain the data rate of the communication link for vehicle m in the current time slot t by monitoring the communication link.

[0018] In one embodiment, S21 includes:

[0019] Using the formula Ψ(m, t; n) = Ψ θ (m, t; n)Ψ d (m, t; n) Calculate the driving environment complexity Ψ(m, t; n) faced by the m-th vehicle;

[0020] Among them, Ψ θ (m, t; n) represents the angle term representing the complexity of the driving environment of the m-th vehicle facing the n-th vehicle. d0 is the upper limit threshold of the point cloud acquisition range. The distance between the m-th vehicle and the n-th vehicle in time slot t; Ψ d (m, t; n) represents the relative distance term of the driving environment complexity of the m-th vehicle facing the n-th vehicle. The angle between the m-th vehicle and the n-th vehicle in time slot t.

[0021] In one embodiment, S3 includes:

[0022] S31: The optimization objective is to maximize the total perception score of all vehicles. At the same time, the accuracy satisfies the constraint P(a) m,t If P ≥ P0, the perceived delay satisfies constraint D. m,t The optimization problem is ≤τ, where τ is the perception delay constraint and P0 is the perception accuracy constraint.

[0023] S32: Solve for the point cloud processing task priority b of the m-th vehicle participating in collaborative perception in the current time slot t. m,t and optimal configuration a m,t ;

[0024] S33: Optimal configuration a m，t Issued to vehicle m so that it can utilize the optimal configuration a. m,tThe point cloud frames after redundancy removal are sampled and compressed to obtain the target point cloud data, which is then sent back to the roadside server.

[0025] In one embodiment, in S31:

[0026] Perceived quality is defined as:

[0027] Where ω1 and ω2 are both positive weighting coefficients, ω3 is a constant, and P(a m,t ) is to configure the m-th optimal configuration a in time slot t. m,t The corresponding average accuracy of point cloud object detection, D m,t for b m,t and a m,t Corresponding to perceived delay.

[0028] In one embodiment, S32 includes: solving the optimization problem according to a greedy algorithm with local time constraints, specifically:

[0029] Set time windows W and D m,t Less than the local delay constraint D, by solving (m, a m,t b m,t =arg maxΦ(m,t), under the local delay constraint D, select the first vehicle to perform point cloud target detection and its corresponding point cloud processing configuration to maximize the local perception score;

[0030] The local delay constraint D is relaxed by pushing it backward by a time window W, and then solving for (m, a) m,t b m,t =argmaxΦ(m,t), under the local delay constraint D+W, select the second vehicle for point cloud target detection and its corresponding point cloud processing configuration, and then relax the local perception delay constraint in turn to select the task processing priority and point cloud processing configuration for subsequent vehicles; and so on, to obtain the b of all vehicles. m,t and a m,t .

[0031] According to another aspect of the present invention, a vehicle-road cooperative perception device based on vehicle-mounted point cloud is provided, applied to a roadside server, comprising:

[0032] The determination module is used to determine the optimal configuration set for sampling and compression processing of vehicle point cloud frames.

[0033] The calculation module is used to calculate the driving environment complexity Ψ(m,t) and communication link data rate of each participating collaborative perception vehicle in the current time slot based on the point cloud target detection results of each participating collaborative perception vehicle in the previous time slot.

[0034] The solver module is used to construct an optimization problem that maximizes the total perceived quality. And solve for the point cloud processing task priority b of the m-th vehicle participating in collaborative perception in the current time slot t. m,t and optimal configuration a m,t ; The optimal configuration a m,t Issued to vehicle m so that it can utilize the optimal configuration a. m，t The redundancy-removed point cloud frames are sampled and compressed to obtain the target point cloud data, which is then transmitted back to the roadside server.

[0035] Where Φ(m,t) represents the perceived quality. The product of the driving environment complexity Ψ(m,t); With optimal configuration a m,t The corresponding perception accuracy is positively correlated with the communication link data rate. The corresponding perception delay is negatively correlated; M is the total number of vehicles participating in collaborative perception.

[0036] The processing module is used to receive target point cloud data corresponding to each participating collaborative perception vehicle in the current time slot, process each target point cloud data according to the point cloud processing task priority, obtain the point cloud target detection result of the current time slot, and send it to the corresponding vehicle to enable it to perform environmental perception.

[0037] According to another aspect of the present invention, a roadside server is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described vehicle-road cooperative perception method.

[0038] According to another aspect of the present invention, a vehicle-road cooperative perception system based on vehicle point clouds is provided, characterized in that it includes:

[0039] A roadside server is used to execute the aforementioned vehicle-road cooperative perception method.

[0040] Multiple vehicles participating in collaborative perception are used to sample and compress the redundancy-free point cloud frames using the optimal configuration issued by the roadside server to obtain target point cloud data and send it back to the roadside server; they are also used to receive the point cloud target detection results of the current time slot issued by the roadside server for environmental perception.

[0041] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects:

[0042] (1) This invention provides a vehicle-road cooperative perception method based on vehicle-mounted point clouds. It utilizes the driving environment complexity and communication link data rate of each participating vehicle in the current time slot to construct an optimization problem that maximizes the overall perception quality. Solving this problem yields the point cloud processing task priority and optimal configuration for each vehicle. Each vehicle uses its corresponding optimal configuration to perform sampling compression and transmit the data back to the roadside server. The roadside server then uses the point cloud processing task priority to perform target detection and distribute the data to each vehicle for environmental perception. The optimization problem considers communication resources, processing priorities, and the collaborative perception accuracy between multiple vehicles and the roadside server, thereby solving the technical problems of poor perception quality caused by large latency, low accuracy, and limited resources in existing autonomous driving vehicles.

[0043] (2) Based on the Pareto optimality principle, this scheme selects multiple optimal configurations by using the average point cloud compression rate and the average accuracy of point cloud target detection corresponding to each optimal configuration to form an optimal configuration set. This allows the selection of the optimal compression configuration combination, thereby improving the processing accuracy of point cloud frames and ultimately improving the accuracy of environmental perception.

[0044] (3) This scheme calculates the driving environment complexity Ψ(m, t; n) of vehicle m facing the nth vehicle under inspection in the current time slot t. All Ψ(m, t; n) are summed to obtain the driving environment complexity Ψ(m, t) corresponding to vehicle m. This scheme considers the driving environment complexity Ψ(m, t) of each vehicle in a complex environment involving multiple vehicles, thereby establishing a more accurate optimization problem for the overall perception quality. This improves the accuracy of environmental perception in the entire vehicle-road cooperative perception system.

[0045] (4) This scheme uses the formula Ψ(m, t; n)=Ψ θ (m, t; n)Ψ d The (m, t; n) function calculates the driving environment complexity Ψ(m, t; n) faced by the m-th vehicle. It takes into account the environment in which the current vehicle faces multiple vehicles. Based on the relative position and relative angle of the two vehicles, it determines the driving environment complexity of each vehicle, which can reduce computational complexity, thereby reducing perception latency and improving environmental perception efficiency.

[0046] (5) The optimization objective of this scheme is to maximize the total perception score of all vehicles. At the same time, the accuracy satisfies the constraint P(a) m，t If P ≥ P0, the perceived delay satisfies constraint D. m，t The optimization problem is ≤τ, where τ is the perception delay constraint and P0 is the perception accuracy constraint. By considering both the perception delay and perception accuracy constraints, the quality of environmental perception can be improved.

[0047] (6) Perceived quality is defined as: The average accuracy P(a) of point cloud object detection was considered. m，t ) and current perceived delay D m，t This can improve the quality of environmental perception.

[0048] (7) This scheme solves the optimization problem based on a greedy algorithm with local time constraints, which can obtain the accurate point cloud processing task priority b while ensuring computational efficiency. m，t and optimal configuration a m,t Ultimately, this improves the accuracy and efficiency of environmental perception. Attached Figure Description

[0049] Figure 1 A flowchart of a vehicle-road cooperative perception method based on vehicle point cloud provided in an embodiment of the present invention.

[0050] Figure 2 A flowchart of a vehicle-road cooperative perception method based on vehicle point cloud provided in an embodiment of the present invention.

[0051] Figure 3 This is a schematic diagram illustrating the assessment of driving environment complexity provided in an embodiment of the present invention.

[0052] Figure 4 This is a schematic diagram of a vehicle-road cooperative perception system based on vehicle point cloud provided in an embodiment of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0054] like Figure 1 and Figure 2 As shown, this invention provides a vehicle-road cooperative perception method based on vehicle-mounted point clouds, applied to a roadside server, including:

[0055] S1: Determine the optimal configuration set for sampling and compression processing of vehicle point cloud frames.

[0056] S2: Calculate the driving environment complexity Ψ(m,t) and communication link data rate of each participating collaborative sensing vehicle in the current time slot based on the point cloud target detection results of each participating collaborative sensing vehicle in the previous time slot.

[0057] S3: Construct an optimization problem to maximize total perceived quality. And solve for the point cloud processing task priority b of the m-th vehicle participating in collaborative perception in the current time slot t. m,t and optimal configuration a m,t ; The optimal configuration a m,t Issued to vehicle m so that it can utilize the optimal configuration a. m,t The redundancy-removed point cloud frames are sampled and compressed to obtain the target point cloud data, which is then transmitted back to the roadside server.

[0058] Where Φ(m,t) represents the perceived quality. The product of the driving environment complexity Ψ(m,t); With optimal configuration a m,t The corresponding perception accuracy is positively correlated with the communication link data rate. The corresponding perception delay is negatively correlated; M is the total number of vehicles participating in collaborative perception.

[0059] S4: Receive the target point cloud data corresponding to each participating collaborative perception vehicle in the current time slot, process each target point cloud data according to the point cloud processing task priority, obtain the point cloud target detection result of the current time slot, and send it to the corresponding vehicle so that it can perform environmental perception.

[0060] In one embodiment, S1 includes:

[0061] S11: Perform offline training on a large number of point cloud frame samples to obtain the average point cloud compression rate and the average accuracy of point cloud target detection for each optimal configuration.

[0062] S12: Based on the average point cloud compression rate and the average accuracy of point cloud target detection, the Pareto optimality principle is applied. Select multiple optimal configurations to form an optimal configuration set. Each optimal configuration includes: sampling method, sampling level, and quantization bit depth during compression;

[0063] Where 'a' represents any one of the configurations, a * For the selected optimal configuration, R(a) is the average point cloud compression rate corresponding to a. * ) is a * The corresponding average point cloud compression rate, P(a) is the average accuracy of point cloud target detection corresponding to a, P(a) * ) is a * The average accuracy of the corresponding point cloud target detection.

[0064] Specifically, the roadside server performs offline training on a large number of point cloud frame samples to obtain the average point cloud compression rate and the average accuracy of point cloud object detection for each configuration, and selects the optimal configuration set through the Pareto optimality principle.

[0065] The point cloud samples are labeled training datasets. During offline training, a point cloud object detection model is used to detect objects in the point clouds after different configurations and analyze the accuracy. Optionally, in this embodiment, the object detection model used is PointPillars. The point cloud processing configuration includes sampling and compression of the point cloud. Optionally, in this embodiment, the sampling method, sampling level, and quantization bits during compression constitute a point cloud processing configuration. Correspondingly, in this embodiment, the sampling method includes random sampling and voxel sampling. The point cloud sampling level is the sampling ratio for random sampling and the voxel size for voxel sampling. Optionally, the point cloud compression coding framework in this embodiment is Draco.

[0066] It should be noted that the description of the point cloud processing configuration here is only an optional implementation of the present invention and should not be construed as the only limitation of this embodiment. In this embodiment, the point cloud processing configuration is composed of parameters such as uniform sampling, filtered sampling, and compression level.

[0067] In one embodiment, S2 includes:

[0068] S21: Calculate the driving environment complexity Ψ(m, t; n) faced by vehicle m in the current time slot t, based on the coordinates and driving direction of vehicle n being inspected. Then, sum all Ψ(m, t; n) to obtain the driving environment complexity Ψ(m, t) corresponding to vehicle m.

[0069] S22: Obtain the data rate of the communication link for vehicle m in the current time slot t by monitoring the communication link.

[0070] Specifically, the roadside server obtains the real-time data transmission rate by monitoring the communication link; such as Figure 3 As shown, the complexity of the vehicle's current driving environment is calculated using the vehicle's coordinates and direction of travel. Finally, the relative distances between the current vehicle and other vehicles are obtained based on the target detection results returned by the roadside server. and the difference in driving direction angle This enables environmental perception.

[0071] In one embodiment, S21 includes:

[0072] Using the formula Ψ(m, t; n) = Ψ θ (m, t; n)Ψ d (m, t; n) Calculate the driving environment complexity Ψ(m, t; n) faced by the m-th vehicle;

[0073] Among them, Ψ θ (m, t; n) represents the angle term representing the complexity of the driving environment of the m-th vehicle facing the n-th vehicle. d0 is the upper limit threshold of the point cloud acquisition range. The distance between the m-th vehicle and the n-th vehicle in time slot t; Ψ d (m, t; n) represents the relative distance term of the driving environment complexity of the m-th vehicle facing the n-th vehicle. The angle between the m-th vehicle and the n-th vehicle in time slot t.

[0074] In one embodiment, S3 includes:

[0075] S31: The optimization objective is to maximize the total perception score of all vehicles. At the same time, the accuracy satisfies the constraint P(a) m,t If P ≥ P0, the perceived delay satisfies constraint D. m,t The optimization problem is ≤τ, where τ is the perception delay constraint and P0 is the perception accuracy constraint.

[0076] S32: Solve for the point cloud processing task priority b of the m-th vehicle participating in collaborative perception in the current time slot t. m,t and optimal configuration a m,t ;

[0077] S33: Optimal configuration a m,t Issued to vehicle m so that it can utilize the optimal configuration a. m,t The point cloud frames after redundancy removal are sampled and compressed to obtain the target point cloud data, which is then sent back to the roadside server.

[0078] In one embodiment, in S31:

[0079] Perceived quality is defined as:

[0080] Where ω1 and ω2 are both positive weighting coefficients, ω3 is a constant, and P(a m,t ) is to configure the m-th optimal configuration a in time slot t. m,t The corresponding average accuracy of point cloud object detection, D m,t for b m,t and a m,t Corresponding to perceived delay.

[0081] In one embodiment, S32 includes: solving the optimization problem according to a greedy algorithm with local time constraints, specifically:

[0082] Set time windows W and D m,t Less than the local delay constraint D, by solving (m, a m,t b m,t=arg maxΦ(m,t), under the local delay constraint D, select the first vehicle to perform point cloud target detection and its corresponding point cloud processing configuration to maximize the local perception score;

[0083] The local delay constraint D is relaxed by pushing it backward by a time window W, and then solving for (m, a) m,t b m,t =argmaxΦ(m,t), under the local delay constraint D+W, select the second vehicle for point cloud target detection and its corresponding point cloud processing configuration, and then relax the local perception delay constraint in turn to select the task processing priority and point cloud processing configuration for subsequent vehicles; and so on, to obtain the b of all vehicles. m,t and a m,t .

[0084] According to another aspect of the present invention, a vehicle-road cooperative perception device based on vehicle-mounted point cloud is provided, applied to a roadside server, comprising:

[0085] The determination module is used to determine the optimal configuration set for sampling and compression processing of vehicle point cloud frames.

[0086] The calculation module is used to calculate the driving environment complexity Ψ(m,t) and communication link data rate of each participating collaborative perception vehicle in the current time slot based on the point cloud target detection results of each participating collaborative perception vehicle in the previous time slot.

[0087] The solver module is used to construct an optimization problem that maximizes the total perceived quality. And solve for the point cloud processing task priority b of the m-th vehicle participating in collaborative perception in the current time slot t. m,t and optimal configuration a m,t ; The optimal configuration a m,t Issued to vehicle m so that it can utilize the optimal configuration a. m,t The redundancy-removed point cloud frames are sampled and compressed to obtain the target point cloud data, which is then transmitted back to the roadside server.

[0088] Where Φ(m,t) represents the perceived quality. The product of the driving environment complexity Ψ(m,t); With optimal configuration a m,t The corresponding perception accuracy is positively correlated with the communication link data rate. The corresponding perception delay is negatively correlated; M is the total number of vehicles participating in collaborative perception.

[0089] The processing module is used to receive the target point cloud data corresponding to each participating collaborative perception vehicle in the current time slot, process each target point cloud data according to the point cloud processing task priority, obtain the point cloud target detection result of the current time slot, and send it to the corresponding vehicle so that it can perform environmental perception.

[0090] According to another aspect of the present invention, a roadside server is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described vehicle-road cooperative perception method.

[0091] According to another aspect of the invention, such as Figure 4 As shown, a vehicle-road cooperative perception system based on vehicle point cloud is provided, characterized by comprising:

[0092] A roadside server is used to execute the aforementioned vehicle-road cooperative perception method.

[0093] Multiple vehicles participating in collaborative perception are used to sample and compress the redundancy-free point cloud frames using the optimal configuration issued by the roadside server to obtain target point cloud data and send it back to the roadside server; they are also used to receive the point cloud target detection results of the current time slot issued by the roadside server for environmental perception.

[0094] In this embodiment, the redundancy removal method for multiple vehicles participating in collaborative perception can be as follows: The coordinates of the vehicle's 3D bounding box within the point cloud space are obtained based on the vehicle's LiDAR sensor and vehicle dimensions; redundant point clouds generated by vehicle reflection within the 3D bounding box are removed. The RANSAC algorithm is used to remove redundant ground point clouds. A ground prediction range is obtained based on the fixed height of the LiDAR sensor. Three points are randomly selected from this range each time, and a plane equation is calculated based on these three points. Any point within 0.15m of this plane equation is considered a fitting point. After 100 iterations, the plane with the most fitting points is selected as the ground plane. This ground plane and its fitting points constitute the redundant point cloud generated by ground reflection and are removed. The roadside server decompresses the received point cloud data and inputs it into the PointPillars model, outputting the detection results. Correspondingly, the roadside server packages and sends this result information to the corresponding autonomous driving vehicle.

[0095] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A vehicle-road cooperative perception method based on vehicle-mounted point cloud, applied to a roadside server, characterized in that, include: S1: Determine the optimal configuration set for sampling and compression processing of vehicle point cloud frames. ; S2: Calculate the driving environment complexity of each participating collaborative perception vehicle in the current time slot based on the point cloud target detection results of each participating collaborative perception vehicle in the previous time slot. and communication link data rate ; S3: Construct an optimization problem to maximize total perceived quality. And solve for the point cloud processing task priority of the m-th vehicle participating in collaborative perception in the current time slot t. and optimal configuration ; Optimal configuration Issued to vehicle m so that it can utilize the optimal configuration. The redundancy-removed point cloud frames are sampled and compressed to obtain the target point cloud data, which is then transmitted back to the roadside server. in, To perceive quality and the complexity of the driving environment The product; With optimal configuration The corresponding perception accuracy is positively correlated with the data rate of the communication link. The corresponding perception delay is negatively correlated; M is the total number of vehicles participating in collaborative perception. S4: Receive the target point cloud data corresponding to each participating collaborative perception vehicle in the current time slot, process each target point cloud data according to the point cloud processing task priority, obtain the point cloud target detection result of the current time slot, and send it to the corresponding vehicle so that it can perform environmental perception. S2 includes: S21: Calculating the driving environment complexity faced by the m-th vehicle using the coordinates and driving direction of the m-th vehicle facing the n-th vehicle under inspection in the current time slot t. All The driving environment complexity for the m-th vehicle is obtained by summing up the results. S22: Obtain the communication link data rate of vehicle m in the current time slot t by monitoring the communication link. ; S21 includes: using the formula Calculate the complexity of the driving environment faced by the m-th vehicle. ; Let be the angle term representing the complexity of the driving environment of vehicle m facing vehicle n. , This is the upper limit threshold for the range of point cloud data to be collected. This represents the distance between the m-th vehicle and the n-th vehicle in time slot t. Let m be the relative distance term representing the complexity of the driving environment between vehicle m and vehicle n. ; Let m be the angle between the m-th vehicle and the n-th vehicle in time slot t. S3 includes: S31: Constructing an optimization objective to maximize the total perception score of all vehicles. At the same time, the accuracy meets the constraints. Perceived delay satisfies constraints The optimization problem; To perceive the delay constraint, To constrain perception accuracy; S32: Solve for the point cloud processing task priority of the m-th vehicle participating in collaborative perception in the current time slot t. and optimal configuration S33: Optimal configuration Issued to vehicle m so that it can utilize the optimal configuration. The redundancy-removed point cloud frames are sampled and compressed to obtain the target point cloud data, which is then transmitted back to the roadside server. In S31, perceived quality is defined as: ; and All are positive weighting coefficients. It is a constant. To configure the m-th optimal configuration in time slot t The corresponding average accuracy of point cloud object detection, for , and Corresponding to perceived delay.

2. The vehicle-road cooperative perception method based on vehicle-mounted point clouds as described in claim 1, characterized in that, S1 includes: S11: Perform offline training on a large number of point cloud frame samples to obtain the average point cloud compression rate and the average accuracy of point cloud target detection for each optimal configuration. S12: Based on the average point cloud compression rate and the average accuracy of point cloud target detection, the Pareto optimality principle is applied. Select multiple optimal configurations to form an optimal configuration set. Each optimal configuration includes: sampling method, sampling level, and quantization bit depth during compression; in, For any one of these configurations, The selected optimal configuration for The corresponding average point cloud compression rate, for The corresponding average point cloud compression rate, for The corresponding average accuracy of point cloud object detection, for The average accuracy of the corresponding point cloud target detection.

3. The vehicle-road cooperative perception method based on vehicle-mounted point clouds as described in claim 1, characterized in that, S32 includes: solving the optimization problem using a greedy algorithm with local time constraints, specifically: Set time window , Less than the local delay constraint D, by solving Under the local delay constraint D, the first vehicle to perform point cloud target detection and its corresponding point cloud processing configuration are selected to maximize the local perception score. Push the local delay constraint D backward by one time window. Relaxation is performed by solving... Under the local delay constraint D+W, the second vehicle to perform point cloud target detection and its corresponding point cloud processing configuration are selected. The local perception delay constraint is then relaxed sequentially to prioritize task processing and select point cloud processing configurations for subsequent vehicles; this process is repeated until all vehicles are identified. and .

4. A vehicle-road cooperative perception device based on vehicle-mounted point clouds, applied to a roadside server, and applied to the vehicle-road cooperative perception method based on vehicle-mounted point clouds as described in any one of claims 1-3, characterized in that, include: The determination module is used to determine the optimal configuration set for sampling and compression processing of vehicle point cloud frames. ; The calculation module is used to calculate the driving environment complexity of each participating collaborative perception vehicle in the current time slot based on the point cloud target detection results of each participating collaborative perception vehicle in the previous time slot. and communication link data rate ; The solver module is used to construct an optimization problem that maximizes the total perceived quality. And solve for the point cloud processing task priority of the m-th vehicle participating in collaborative perception in the current time slot t. and optimal configuration ; Optimal configuration Issued to vehicle m so that it can utilize the optimal configuration. The redundancy-removed point cloud frames are sampled and compressed to obtain the target point cloud data, which is then transmitted back to the roadside server. in, To perceive quality and the complexity of the driving environment The product; With optimal configuration The corresponding perception accuracy is positively correlated with the data rate of the communication link. The corresponding perception delay is negatively correlated; M is the total number of vehicles participating in collaborative perception. The processing module is used to receive target point cloud data corresponding to each participating collaborative perception vehicle in the current time slot, process each target point cloud data according to the point cloud processing task priority, obtain the point cloud target detection result of the current time slot, and send it to the corresponding vehicle to enable it to perform environmental perception.

5. A roadside server, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the vehicle-road cooperative perception method based on vehicle point cloud as described in any one of claims 1-3.

6. A vehicle-road cooperative perception system based on vehicle-mounted point clouds, characterized in that, include: A roadside server for executing the vehicle-road cooperative perception method based on vehicle point cloud as described in any one of claims 1-3; Multiple vehicles participating in collaborative perception are used to sample and compress the redundancy-free point cloud frames using the optimal configuration issued by the roadside server to obtain target point cloud data and send it back to the roadside server; they are also used to receive the point cloud target detection results of the current time slot issued by the roadside server for environmental perception.

Images