A commercial vehicle right turn test evaluation method, system and device

Abstract

The application discloses a commercial vehicle right-turn test evaluation method, system and equipment, relates to the vehicle automatic driving test evaluation technical field, and the method comprises the following steps: generating a test scene based on traffic data; taking a general precision time protocol as a global clock reference, and acquiring a space-time data stream of a commercial vehicle running in the test scene; performing three-dimensional Boolean operation based on the space-time data stream to obtain a dynamic absolute blind area; performing calculation on a target in and across the dynamic absolute blind area, a dynamic trajectory conflict time, and an evaluation on a blind area space-time prediction topology deviation to obtain an identity tracking penalty value, the dynamic trajectory conflict time and the topology prediction deviation degree; generating a comprehensive space-time continuity risk index based on the dynamic trajectory conflict time, the identity tracking penalty value and the topology prediction deviation degree, and outputting an evaluation report. The method can improve the comprehensiveness of the commercial vehicle test evaluation system.

Description

Technical Field

[0001] This invention relates to the field of vehicle autonomous driving testing and evaluation technology, and in particular to a method, system and equipment for testing and evaluating right turns of commercial vehicles. Background Technology

[0002] With the widespread adoption of autonomous driving and advanced driver assistance systems, testing and evaluation technologies for intelligent connected vehicles are developing rapidly. However, for commercial vehicles such as heavy trucks, buses, and dump trucks, right turns are a high-risk area for traffic accidents. Due to their long wheelbase and tall body, commercial vehicles generate a significant inner wheel difference and physical blind spots at the front and sides when turning right, making them highly susceptible to running over pedestrians, two-wheeled vehicles, and other vulnerable road users in these areas.

[0003] The shortcomings of existing technologies are as follows: First, current tests mostly use minimum collision time or whether a collision occurs as evaluation criteria, which are typical black-box tests. This method can only evaluate whether the vehicle eventually comes to a stop, but cannot judge whether the perception coverage of the intelligent driving system is reasonable or whether target loss is frequent during a right turn. Second, there is a lack of an evaluation system deeply coupled with physical blind spots. Existing evaluation indicators such as multi-target tracking accuracy do not incorporate the dynamic inner wheel difference and visual cone occlusion geometric model unique to commercial vehicles. When a target is briefly obscured by a large vehicle body, existing methods cannot reasonably determine whether it is a reasonable physical occlusion or an algorithmic miss. Third, in real-world extreme right-turn scenarios, after vulnerable road users enter the absolute blind spot, excellent intelligent driving systems should have the logical reasoning ability for object persistence and the ability to predict temporal trajectories. However, existing technologies separate spatial occlusion from temporal tracking, lacking spatiotemporal continuity and object existence evaluation methods. Summary of the Invention

[0004] To address the problem that existing technologies are not comprehensive enough in evaluating right-turn tests of commercial vehicles, this invention provides a method, system, and device for evaluating right-turn tests of commercial vehicles. The specific technical solution is as follows: In a first aspect, the present invention provides a method for evaluating right turns of commercial vehicles, comprising the following steps: S1: Generating test scenarios based on traffic data; S2: Using a universal precision time protocol as the global clock reference, acquire the spatiotemporal data stream of the commercial vehicle running in the test scenario; S3: Perform three-dimensional Boolean operations based on the spatiotemporal data stream to obtain the dynamic absolute blind zone; S4: Calculate the cross-blind zone identity tracking penalty value, dynamic trajectory conflict time, and evaluate the spatiotemporal prediction topology deviation within the blind zone for targets that are in or cross the dynamic absolute blind zone, and obtain the identity tracking penalty value, dynamic trajectory conflict time, and topology prediction deviation. S5: Generate a comprehensive spatiotemporal continuity risk index based on the dynamic trajectory conflict time, the identity tracking penalty value, and the topology prediction deviation, and output an evaluation report.

[0005] Furthermore, the process of generating the test scenario in step S1 is as follows: A scenario feature space is constructed, traffic data is decoded into structured feature vectors, a deep generative model is used to learn the joint probability distribution of the feature vectors, and the feature weights of the preset feature vectors are adjusted to generate a test scenario of right-turn conflict for commercial vehicles.

[0006] Further, the spatiotemporal data stream in step S2 includes a motion state matrix, a target true time-series trajectory, and an operation output matrix. The motion state matrix includes the real-time vehicle speed, front wheel steering angle, yaw rate, wheelbase, and global coordinates of the commercial vehicle. The target true time-series trajectory is the absolute true position sequence of the target in the global coordinate system. The operation output matrix includes the target identification code sequence and the target predicted time-series trajectory.

[0007] Furthermore, the process of obtaining the dynamic absolute blind zone in step S3 is as follows: Based on Ackerman steering geometry, current front wheel angle and wheelbase, calculate the dynamic inner wheel difference sweep area for right turns, and combine it with the three-dimensional outer contour of the commercial vehicle to calculate the physical occlusion shadow area of ​​the vehicle body; obtain the three-dimensional effective coverage area of ​​the commercial vehicle sensor cone projected on the ground; After taking the union of the swept area and the occluded shadow area, a Boolean difference operation is performed with the three-dimensional effective coverage area to obtain the dynamic absolute blind zone.

[0008] Furthermore, the calculation of the cross-blind-spot identity tracking penalty value in step S4 is as follows: Obtain the initial identification code when the target enters the absolute blind zone, and the exit identification code when it exits the absolute blind zone; if the initial identification code and the exit identification code are inconsistent, it is determined that an identity change has occurred; calculate the identity tracking penalty value based on the identity change indication function and the recovery hysteresis time of the target reassigning the identification code after exiting the blind zone.

[0009] The calculation of the dynamic trajectory conflict time in step S4 is as follows: Based on the kinematic state matrix and the right-turn dynamic inner wheel difference sweep region, combined with the target's true time-series trajectory, the remaining time when the commercial vehicle's inner wheel difference sweep boundary and the target intersect and overlap in the global coordinate system is extrapolated and calculated. The remaining time is the dynamic trajectory conflict time.

[0010] The evaluation of the spatiotemporal prediction topology bias within the blind zone in step S4 is as follows: Based on the target ground time-series trajectory and the target predicted time-series trajectory, the spatial topological alignment distance between the two time-series trajectories is calculated using the dynamic time warping algorithm, and the corresponding distance matrix terminal value is extracted as the topological prediction deviation.

[0011] Furthermore, the comprehensive spatiotemporal continuity risk index described in step S5 is generated by weighted and normalized fusion of the reciprocal of the dynamic trajectory conflict time, the identity tracking penalty value, and the topology prediction deviation.

[0012] Secondly, this invention proposes a commercial vehicle right-turn test evaluation system, including a scenario construction module, a data acquisition module, an edge evaluation calculation module, and a report generation module. The scenario construction module is used to generate a test scenario based on traffic data. The data acquisition module is used to acquire the spatiotemporal data stream of the commercial vehicle running in the test scenario using a universal precise time protocol as the global clock reference. The edge evaluation calculation module is used to receive the spatiotemporal data stream and perform three-dimensional Boolean operations to obtain the dynamic absolute blind zone. For targets that are in or cross the dynamic absolute blind zone, the system calculates the cross-blind zone identity tracking penalty value, the dynamic trajectory conflict time, and evaluates the spatiotemporal prediction topology deviation within the blind zone, obtaining the identity tracking penalty value, the dynamic trajectory conflict time, and the topology prediction deviation degree. The report generation module is used to generate a comprehensive spatiotemporal continuity risk index based on the dynamic trajectory conflict time, the identity tracking penalty value, and the topology prediction deviation degree, and outputs an evaluation report.

[0013] Thirdly, the present invention proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the program to implement the commercial vehicle right-turn test evaluation method proposed in the present invention.

[0014] Beneficial effects: This invention proposes a method, system, and device for evaluating right-turn tests of commercial vehicles. The method acquires the spatiotemporal data stream of the commercial vehicle operating in the test scenario, performs three-dimensional Boolean operations based on the spatiotemporal data stream to obtain the dynamic absolute blind zone, and when the target true value enters the absolute blind zone, quantitatively calculates the target identity jump penalty value and the topological deviation degree of the predicted trajectory within the blind zone, and simultaneously extrapolates the dynamic trajectory conflict time based on the inner wheel difference region to generate a comprehensive evaluation result. The method can overcome the limitations of static field of view evaluation, quantitatively evaluate the object persistence and blind zone prediction capabilities in right-turn tests of commercial vehicles, and improve the comprehensiveness of the right-turn test evaluation system for commercial vehicles. Attached Figure Description

[0015] Figure 1 This is a flowchart of a commercial vehicle right-turn test evaluation method according to an embodiment of the present invention; Figure 2This is a flowchart illustrating a specific implementation of a commercial vehicle right-turn test and evaluation method according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating the dynamic absolute blind zone spatial delineation in an embodiment of the present invention; Figure 4 This is an architectural diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0017] Example 1 This embodiment proposes a test and evaluation method for right turns of commercial vehicles, and the flowchart of the method is as follows. Figure 1 As shown, it includes the following steps: S1: Generating test scenarios based on traffic data; It should be noted that the generation process of the test scenario is as follows: A scenario feature space is constructed, traffic data is decoded into structured feature vectors, a deep generative model is used to learn the joint probability distribution of the feature vectors, and the feature weights of the preset feature vectors are adjusted to generate a test scenario of right-turn conflict for commercial vehicles.

[0018] In one specific embodiment, the preset feature vector includes the traffic participant's initial position, movement speed, and intrusion angle.

[0019] S2: Using a universal precision time protocol as the global clock reference, acquire the spatiotemporal data stream of the commercial vehicle running in the test scenario; It should be noted that the spatiotemporal data stream includes a motion state matrix, a target ground truth time-series trajectory, and an operational output matrix. The motion state matrix includes the commercial vehicle's real-time speed, front wheel steering angle, yaw rate, wheelbase, and global coordinates. The target ground truth time-series trajectory is the absolute true position sequence of the target in the global coordinate system. The operational output matrix includes a target identification code sequence and a target predicted time-series trajectory. The target mainly refers to vulnerable road users (VRUs) in the commercial vehicle's driving environment, including but not limited to pedestrians, bicycles, and electric two-wheelers.

[0020] S3: Perform three-dimensional Boolean operations based on the spatiotemporal data stream to obtain the dynamic absolute blind zone; It should be noted that the process of obtaining the dynamic absolute blind zone is as follows: Based on Ackerman steering geometry, current front wheel angle and wheelbase, calculate the dynamic inner wheel difference sweep area for right turns, and combine it with the three-dimensional outer contour of the commercial vehicle to calculate the physical occlusion shadow area of ​​the vehicle body; obtain the three-dimensional effective coverage area of ​​the commercial vehicle sensor cone projected on the ground; After taking the union of the swept area and the occluded shadow area, a Boolean difference operation is performed with the three-dimensional effective coverage area to obtain the dynamic absolute blind zone.

[0021] S4: Calculate the cross-blind zone identity tracking penalty value, dynamic trajectory conflict time, and evaluate the spatiotemporal prediction topology deviation within the blind zone for targets that are in or cross the dynamic absolute blind zone, and obtain the identity tracking penalty value, dynamic trajectory conflict time, and topology prediction deviation. It should be noted that the calculation of the cross-blind-spot identity tracking penalty value is as follows: Obtain the initial identification code when the target enters the absolute blind zone, and the exit identification code when it exits the absolute blind zone; if the initial identification code and the exit identification code are inconsistent, it is determined that an identity change has occurred; calculate the identity tracking penalty value based on the identity change indication function and the recovery hysteresis time of the target reassigning the identification code after exiting the blind zone.

[0022] The calculation of the dynamic trajectory conflict time is as follows: Based on the kinematic state matrix and the right-turn dynamic inner wheel difference sweep region, combined with the target's true time-series trajectory, the remaining time when the commercial vehicle's inner wheel difference sweep boundary and the target intersect and overlap in the global coordinate system is extrapolated and calculated. The remaining time is the dynamic trajectory conflict time.

[0023] The evaluation of the spatiotemporal prediction topology bias within the blind zone includes the following: Based on the target ground time-series trajectory and the target predicted time-series trajectory, the spatial topological alignment distance between the two time-series trajectories is calculated using the dynamic time warping algorithm, and the corresponding distance matrix terminal value is extracted as the topological prediction deviation.

[0024] S5: Generate a comprehensive spatiotemporal continuity risk index based on the dynamic trajectory conflict time, the identity tracking penalty value, and the topology prediction deviation, and output an evaluation report.

[0025] It should be noted that the comprehensive spatiotemporal continuity risk index is generated by weighted and normalized fusion of the reciprocal of the dynamic trajectory conflict time, the identity tracking penalty value, and the topology prediction deviation.

[0026] This invention provides a dynamic physical blind zone definition mathematical model by constructing a model using Ackerman steering geometry and Boolean operations to dynamically calculate the absolute blind zone that cannot be effectively covered by sensors and belongs to the inner wheel difference danger zone. This overcomes the limitations of static field of view evaluation and provides a spatial geometric basis for white-box evaluation. This invention provides a quantitative evaluation algorithm for cross-blind zone tracking and prediction: it introduces a cross-blind zone identity change determination and recovery hysteresis time penalty function, and combines a dynamic time warping algorithm to calculate the trajectory prediction deviation under occlusion conditions, thereby realizing a quantitative evaluation of the object persistence and blind zone prediction capabilities of the intelligent driving system. The evaluation method of this invention has a highly generalizable environment decoupling architecture: by uniformly defining the data acquisition standards for the vehicle state, the target true value and the predicted value, the evaluation method can seamlessly adapt to and be compatible with virtual simulation software testing and real vehicle closed-field testing.

[0027] Example 2 This embodiment provides a specific implementation method for testing and evaluating right turns of commercial vehicles, as shown in the flowchart below. Figure 2 As shown, the specific content is as follows: Step S1: Test Scenario Generation and Distribution In one specific embodiment, a scene feature space is constructed, traffic data is decoded into structured feature vectors, and a deep generative model is used to learn the joint probability distribution of the feature vectors. The initial positions, speeds, and intrusion angles of traffic participants are then adjusted to generate a commercial vehicle right-turn conflict scenario, which is then distributed to the test execution platform. The traffic data comes from raw data sources including real-world driving datasets, the CIDAS accident database, intersection surveillance videos, or raw point cloud data collected by LiDAR. The deep learning model includes common Generative Adversarial Networks (GANs). The test execution platform includes real physical commercial vehicles as well as virtual simulation testing software or a hardware-in-the-loop test bench.

[0028] Step S2: High-frequency synchronous acquisition of multi-source spatiotemporal data In one specific embodiment, the control test commercial vehicle operates in the right-turn conflict scenario, using a universal precision time protocol as the global clock reference, and synchronously acquires the following three types of spatiotemporal data streams in real time: The first category is the vehicle's kinematic state matrix, which includes the commercial vehicle's real-time speed, front wheel steering angle, yaw rate, wheelbase, and global coordinates; The second category is the target true time-series trajectory, which is the absolute true position sequence of the target in the global coordinate system; The third category is the runtime output matrix, which is the real-time parsing of the target identification code sequence and the predicted time-series trajectory.

[0029] It should be noted that the target identification code sequence and the predicted time-series trajectory are outputs of the System Under Test (SUT), including the domain controller or algorithm under test, and not outputs from vehicle operation.

[0030] Step S3: Dynamic absolute blind zone spatial delineation, flowchart as follows Figure 3As shown In one specific embodiment, the spatiotemporal data stream is received and a three-dimensional Boolean operation is performed: Based on Ackerman steering geometry, current front wheel angle and wheelbase, the dynamic inner wheel difference sweep area for right turns is calculated. Combined with the three-dimensional outer contour of the commercial vehicle, the physical occlusion shadow area of ​​the vehicle body is calculated, and the three-dimensional effective coverage area of ​​the visual cones of each sensor of the commercial vehicle on the ground is obtained.

[0031] After taking the union of the swept area and the occluded shadow area, a Boolean difference operation is performed with the effective coverage area to obtain the dynamic absolute blind zone.

[0032] The target's true coordinates are compared in real time. If the target's true value falls into the dynamic absolute blind zone, step S4 is triggered.

[0033] Step S4: Evaluation of the spatiotemporal continuity and physical performance of the right turn In one specific embodiment, the target that is in or crosses the dynamic absolute blind zone is quantitatively evaluated, specifically including the following three calculation dimensions: (1) Calculate the cross-blind zone identity tracking continuity penalty value: Obtain the initial identity code assigned when the target enters the absolute blind zone, and the exit identity code assigned when it exits the absolute blind zone. If the initial identity code and the exit identity code are inconsistent, an identity jump is determined to have occurred. Calculate the identity tracking penalty value based on the identity jump indication function and the recovery hysteresis time of the reassigned identity code after the target exits the blind zone.

[0034] (2) Evaluate the spatiotemporal prediction topology bias within the blind zone: Extract the predicted time-series trajectory sequence and the absolute true time-series trajectory sequence of the target within the absolute blind zone. Calculate the spatial topology alignment distance between the two time-series trajectories using a dynamic time warping algorithm, and extract the corresponding distance matrix terminal value as the topology prediction bias.

[0035] (3) Calculate the dynamic trajectory conflict time: Based on the kinematic state matrix obtained in step S2 and the right-turn dynamic inner wheel difference sweep area calculated in step S3, combined with the target true time series trajectory, extrapolate and calculate the remaining time when the inner wheel difference sweep boundary of this vehicle and the target intersect and overlap in the global coordinate system, and define it as the dynamic trajectory conflict time (i.e., inner wheel difference dynamic TTC).

[0036] Step S5: Generation of comprehensive evaluation results In one specific embodiment, the reciprocal of the dynamic trajectory conflict time, the identity tracking penalty value, and the topology prediction deviation are weighted, normalized, and fused to generate a comprehensive spatiotemporal continuity risk index and output an evaluation report.

[0037] Example 3 This embodiment provides an electronic device. Figure 4The diagram shows the structure of the device, which may include: a processor 402, a communications interface 404, a memory 406, and a communications bus 408.

[0038] The processor 402, communication interface 404, and memory 406 communicate with each other via communication bus 408. Communication interface 404 is used to communicate with other network elements such as clients or other servers. Processor 402 executes program 410, specifically performing the relevant steps of the commercial vehicle right-turn test evaluation method described in this invention.

[0039] Specifically, program 410 may include program code, which includes computer-executable instructions.

[0040] Processor 402 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The electronic device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0041] Memory 406 is used to store program 410. Memory 406 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0042] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. 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 claims of the present invention.

Claims

1. A method for testing and evaluating right turns of commercial vehicles, characterized in that, Includes the following steps: S1: Generating test scenarios based on traffic data; S2: Using a universal precision time protocol as the global clock reference, acquire the spatiotemporal data stream of the commercial vehicle running in the test scenario; S3: Perform three-dimensional Boolean operations based on the spatiotemporal data stream to obtain the dynamic absolute blind zone; S4: Calculate the cross-blind zone identity tracking penalty value, dynamic trajectory conflict time, and evaluate the spatiotemporal prediction topology deviation within the blind zone for targets that are in or cross the dynamic absolute blind zone, and obtain the identity tracking penalty value, dynamic trajectory conflict time, and topology prediction deviation. S5: Generate a comprehensive spatiotemporal continuity risk index based on the dynamic trajectory conflict time, the identity tracking penalty value, and the topology prediction deviation, and output an evaluation report. The spatiotemporal data stream in step S2 includes a motion state matrix, a target ground truth time-series trajectory, and an operation output matrix. The motion state matrix includes the real-time vehicle speed, front wheel steering angle, yaw rate, wheelbase, and global coordinates of the commercial vehicle. The target ground truth time-series trajectory is the absolute true position sequence of the target in the global coordinate system. The operation output matrix includes the target identification code sequence and the target predicted time-series trajectory. The calculation of the cross-blind zone identity tracking penalty value in step S4 is as follows: Obtain the initial identification code when the target enters the absolute blind zone, and the exit identification code when it exits the absolute blind zone; if the initial identification code and the exit identification code are inconsistent, it is determined that an identity change has occurred; calculate the identity tracking penalty value based on the identity change indication function and the recovery hysteresis time of the target reassigning the identification code after exiting the blind zone. The calculation of the dynamic trajectory conflict time in step S4 is as follows: Based on the kinematic state matrix and the right-turn dynamic inner wheel difference sweep area, combined with the target's true time-series trajectory, the remaining time when the commercial vehicle's inner wheel difference sweep boundary and the target intersect and overlap in the global coordinate system is extrapolated and calculated. The remaining time is the dynamic trajectory conflict time. The evaluation of the spatiotemporal prediction topology bias within the blind zone in step S4 is as follows: Based on the target ground time-series trajectory and the target predicted time-series trajectory, the spatial topological alignment distance between the two time-series trajectories is calculated using the dynamic time warping algorithm, and the corresponding distance matrix terminal value is extracted as the topological prediction deviation.

2. The commercial vehicle right-turn test and evaluation method according to claim 1, characterized in that, The process of generating the test scenario in step S1 is as follows: A scenario feature space is constructed, traffic data is decoded into structured feature vectors, a deep generative model is used to learn the joint probability distribution of the feature vectors, and the feature weights of the preset feature vectors are adjusted to generate a test scenario of right-turn conflict for commercial vehicles.

3. The commercial vehicle right-turn test and evaluation method according to claim 1, characterized in that, The process of obtaining the dynamic absolute blind zone in step S3 is as follows: Based on Ackerman steering geometry, current front wheel angle and wheelbase, calculate the dynamic inner wheel difference sweep area for right turns, and combine it with the three-dimensional outer contour of the commercial vehicle to calculate the physical occlusion shadow area of ​​the vehicle body; obtain the three-dimensional effective coverage area of ​​the commercial vehicle sensor cone projected on the ground; After taking the union of the swept area and the occluded shadow area, a Boolean difference operation is performed with the three-dimensional effective coverage area to obtain the dynamic absolute blind zone.

4. The commercial vehicle right-turn test and evaluation method according to claim 1, characterized in that, The comprehensive spatiotemporal continuity risk index described in step S5 is generated by weighted and normalized fusion of the reciprocal of the dynamic trajectory conflict time, the identity tracking penalty value, and the topology prediction deviation.

5. A testing and evaluation system for right turns of commercial vehicles, characterized in that, The system includes a scenario construction module, a data acquisition module, an edge assessment calculation module, and a report generation module. The scenario construction module generates test scenarios based on traffic data. The data acquisition module acquires spatiotemporal data streams of commercial vehicles operating in the test scenarios, using a universal precision time protocol as the global clock reference. The edge assessment calculation module receives the spatiotemporal data streams and performs three-dimensional Boolean operations to obtain dynamic absolute blind zones. It then calculates cross-blind zone identity tracking penalty values, dynamic trajectory conflict times, and assesses spatiotemporal prediction topology deviations for targets within and crossing the dynamic absolute blind zones, obtaining identity tracking penalty values, dynamic trajectory conflict times, and topology prediction deviation degrees. The report generation module generates a comprehensive spatiotemporal continuity risk index based on the dynamic trajectory conflict time, the identity tracking penalty value, and the topology prediction deviation degree, and outputs an evaluation report.

6. An electronic device comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that, When the processor executes the program, it implements the method of any one of claims 1 to 4.

Images