Method for testing metamorphosis of automatic driving cooperative tracking system and related equipment

Abstract

Embodiments of the present application provide a kind of automatic driving coordination tracking system transmutation test method and related equipment, belong to automatic driving test technical field.The method comprises: obtaining initial data set, wherein the initial data set includes several initial scenes;From initial data set, extract the initial scene of preset scene quantity as basic scene, obtain basic scene queue;Variation operation is carried out to basic scene queue, and test scene is obtained, wherein variation operation is inserted with the target vehicle of driving behavior, and driving behavior includes following, overtaking and turning;Test scene is input into coordination tracking system for tracking, and the selection metric of each test scene is calculated according to scene selection expression;According to selection metric, scene selection is carried out to test scene, and result scene is obtained;According to result scene, update basic scene queue, and the tracking result of result scene is output as test report.The embodiments of the present application can automatically generate test scene, and effectively test automatic driving.

Description

Technical Field

[0001] This application relates to the field of autonomous driving testing technology, and in particular to a transformation testing method and related equipment for an autonomous driving cooperative tracking system. Background Technology

[0002] The field of autonomous driving has undergone long-term development, giving rise to a series of driving platforms such as Tesla's FSD, Baidu Apollo, and Waymo. The implementation and market performance of these platforms mark a new stage in the industry. As a key technology for future transportation, its core lies in the perception module responsible for understanding the surrounding environment. On the one hand, among many sensors, LiDAR has become an indispensable component due to its ability to provide accurate three-dimensional spatial information. Modern LiDAR-based perception modules mainly employ deep learning technology. These models are trained using a large amount of labeled point cloud data to perform critical tasks such as 3D object detection or object tracking. Despite continuous technological iteration, fatal accidents caused by perception module failure still occur from time to time, especially given the limitations of object detection systems in handling extreme cases, such as the collision between an Uber autonomous vehicle and a pedestrian. Therefore, conducting thorough and effective testing of such systems is of great significance.

[0003] On the other hand, current large-scale deployed systems mainly employ single-vehicle intelligent perception technology, relying solely on their own sensors for environmental perception. This model has significant limitations when facing object occlusion, sensor malfunctions, or complex traffic scenarios, easily creating perception blind spots and inducing erroneous decision-making, even leading to serious accidents. To overcome these challenges, vehicle-to-everything (V2X) technology has emerged. This technology achieves multi-source perception data sharing through real-time communication between vehicles and between vehicles and roadside infrastructure, and fuses the data in a global coordinate system, thereby constructing a more comprehensive and robust environmental model. This enables cooperative tracking systems to overcome the physical limitations of single-vehicle perception, achieving more accurate and long-range target tracking. While cooperative tracking improves performance, it also increases system complexity. Complex logic often implies more potential defects, making large-scale testing particularly crucial.

[0004] Comprehensive testing of such systems requires coverage of a large number of diverse real-world scenarios. However, collecting this data through real-vehicle road tests is both time-consuming and labor-intensive. The collected multi-view data (such as point clouds and images) requires extensive manual annotation, which is a highly challenging and time-consuming process. Summary of the Invention

[0005] The main objective of this application is to propose a transformation test method and related equipment for an autonomous driving cooperative tracking system, which aims to automatically generate test scenarios with continuous spatiotemporal consistency suitable for cooperative tracking tasks, effectively test the autonomous driving system, and thus improve its robustness.

[0006] To achieve the above objectives, one aspect of this application proposes a transformation test method for an autonomous driving cooperative tracking system, the method comprising: In some embodiments, an initial dataset is obtained, wherein the initial dataset includes several initial scenes, and each initial scene includes point cloud frames and labels of the master vehicle and the cooperating vehicles; A predetermined number of initial scenes are extracted from the initial dataset as base scenes to obtain a base scene queue; The basic scenario queue is subjected to a mutation operation to obtain a test scenario, wherein the mutation operation is to insert a target vehicle with driving behavior, including following, overtaking and turning. The test scenarios are input into the collaborative tracking system for tracking, and the selection metric for each test scenario is calculated based on the scenario selection expression. The test scenario is selected based on the selection metric to obtain the result scenario; The base scenario queue is updated based on the resulting scenario, and the tracking results of the resulting scenario are output as a test report. Then, the process returns to the step of performing a mutation operation on the base scenario queue to obtain the test scenario, until the preset maximum number of iterations is reached.

[0007] In some embodiments, performing a mutation operation on the basic scenario queue to obtain a test scenario includes the following steps: Based on the behavioral constraints of driving behavior, the coordinates of the target vehicle to be inserted are calculated to obtain the insertion position information of the target vehicle in the point cloud frame corresponding to each of the basic scenarios. The insertion position information of consecutive different point cloud frames constitutes the driving trajectory of the target vehicle. Based on the 3D vehicle model, test cases are obtained by simulating rendering at the corresponding positions in each point cloud frame according to the insertion position information. The test cases are subjected to metamorphic tests based on the preset metamorphic relationship to obtain test predictions. The test oracle detects whether the test cases violate the metamorphic relationship, and the test cases that satisfy the metamorphic relationship are identified as test scenarios.

[0008] In some embodiments, the step of calculating the coordinates of the target vehicle to be inserted based on the behavioral constraints of driving behavior to obtain the insertion position information of the target vehicle in the point cloud frame corresponding to each of the basic scenes includes the following steps: When the driving behavior is turning, the road boundary line is identified to obtain a set of road boundary points used to distinguish the driving directions of different lanes; The turning trajectory is calculated based on the turning constraints and the road boundary point set to obtain multiple turning target points. The turning constraints include turning position sub-conditions, boundary distance sub-conditions, and turning direction sub-conditions. The turning constraints are valid when all three conditions are met simultaneously. The turning position sub-condition is that the turning target point is located on the perpendicular line of the road boundary. The boundary distance sub-condition is that the distance between the turning target point and the road boundary point does not exceed the road boundary threshold. The turning direction sub-condition is that the turning target point is located on one side of the lane's driving direction. The initial position of a target vehicle to be inserted is determined from multiple turning target points according to the detection constraints. The detection constraints include a perception time sub-condition and an occlusion sub-condition. The detection constraints are valid when both the perception time sub-condition and the occlusion sub-condition are met. The perception time sub-condition is that the detection time is greater than a preset minimum perception time threshold. The occlusion sub-condition is that at least one master vehicle or cooperative vehicle detects that the occlusion rate of the target vehicle is less than a preset occlusion rate threshold. Starting from the initial position, according to the order of the turning target points on the turning trajectory, a turning target point is determined sequentially for each point cloud frame corresponding to the basic scene. The horizontal and vertical coordinates of the target vehicle to be inserted are adjusted according to the approach target constraint and the turning target point to obtain the insertion position information.

[0009] In some embodiments, the step of calculating the coordinates of the target vehicle to be inserted based on the behavioral constraints of driving behavior to obtain the insertion position information of the target vehicle in the point cloud frame corresponding to each of the basic scenes includes the following steps: When the driving behavior is following another vehicle, a vehicle is randomly selected from the main vehicle and the cooperating vehicle as the target vehicle to be inserted. A local coordinate system is established in the point cloud frame corresponding to each of the basic scenes with the target vehicle as the origin. The direction of movement of the target vehicle is the positive direction of the horizontal axis, and the left side of the direction of movement is the positive direction of the vertical axis. The initial position of the target vehicle to be inserted is determined according to the following constraints, wherein the following constraints include a following position sub-condition and a following distance sub-condition. When both the following position sub-condition and the following distance sub-condition are met, the following constraints are met. The following position sub-condition is that the horizontal axis coordinate of the target vehicle is less than the horizontal axis coordinate of the following target and the vertical axis coordinate of the target vehicle is equal to the vertical axis coordinate of the following target. The following distance sub-condition is that the absolute value of the difference between the horizontal axis coordinates of the target vehicle and the following target does not exceed the following distance threshold. Starting from the initial position, the horizontal axis coordinates of the target vehicle are adjusted according to the proximity constraint and the speed difference between the target vehicle and the following target in two adjacent point cloud frames to obtain the insertion position information.

[0010] In some embodiments, the step of calculating the coordinates of the target vehicle to be inserted based on the behavioral constraints of driving behavior to obtain the insertion position information of the target vehicle in the point cloud frame corresponding to each of the basic scenes includes the following steps: When the driving behavior is overtaking, a vehicle is randomly selected from the main vehicle and the cooperating vehicle as the target vehicle to be inserted. A local coordinate system is established in the point cloud frame corresponding to each of the basic scenes with the target vehicle as the origin. The direction of the overtaking target is the positive direction of the horizontal axis, and the left side of the direction of the forward movement is the positive direction of the vertical axis. The overtaking trajectory is calculated based on the overtaking constraints to obtain four overtaking target points, including the initial position, lane change position, acceleration position, and end position. The overtaking constraints include overtaking position sub-conditions, lane change collision avoidance sub-conditions, and speed limit sub-conditions. Starting from the initial position, according to the order of the overtaking target points, an overtaking target point is determined sequentially for each point cloud frame corresponding to the basic scene. The horizontal and vertical coordinates of the target vehicle to be inserted are adjusted according to the approach target constraint and the overtaking target point to obtain the insertion position information.

[0011] In some embodiments, performing metamorphic testing on the test cases according to a preset metamorphic relationship to obtain a test prediction includes the following steps: The true trajectory result of the target vehicle is merged with the original tracking result of the basic scene to obtain the first test result; The test results of the first test and the test tracking results of the test cases are matched based on a preset metamorphic relationship with a soft equality criterion to obtain a test prediction. The metamorphic relationship includes inserting a target vehicle without changing the original tracking result of the cooperative tracking system, and the true trajectory result is accurately tracked in the test tracking result.

[0012] In some embodiments, inputting the test scenarios into a collaborative tracking system for tracking, and calculating the selection metric for each test scenario based on a scenario selection expression, includes the following steps: The test scenario is input into the collaborative tracking system to calculate the crossover ratio (CRO) and average the mean of several tracking objects to be detected, and the average tracking CRO is obtained. The error rate is calculated based on the number of false positives and false negatives of the tracked object; The average tracking crossover ratio and the error rate are weighted and calculated to obtain the selection metric for each test scenario.

[0013] To achieve the above objectives, another aspect of this application proposes a transformation test system for an autonomous driving cooperative tracking system, the system comprising: The first module is used to obtain an initial dataset, wherein the initial dataset includes several initial scenes, and each initial scene includes point cloud frames and labels of the main vehicle and the cooperative vehicles; The second module is used to extract a preset number of initial scenes from the initial dataset as basic scenes to obtain a basic scene queue; The third module is used to perform mutation operations on the basic scenario queue to obtain test scenarios. The mutation operation is to insert target vehicles with driving behaviors, including following, overtaking, and turning. The fourth module is used to input the test scenarios into the collaborative tracking system for tracking, and to calculate the selection metric for each test scenario based on the scenario selection expression; The fifth module selects a test scenario based on the selection metric to obtain the result scenario; The sixth module updates the basic scenario queue based on the result scenario, outputs the tracking results of the result scenario as a test report, and then returns to the step of performing a mutation operation on the basic scenario queue to obtain the test scenario, until the preset maximum number of iterations is reached.

[0014] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0015] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0016] To achieve the above objectives, another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0017] The embodiments of this application include at least the following beneficial effects: This application provides a metamorphic testing method and related equipment for an autonomous driving cooperative tracking system. The method involves: acquiring an initial dataset, which includes several initial scenarios; extracting a preset number of initial scenarios from the initial dataset as base scenarios to obtain a base scenario queue; performing a mutation operation on the base scenario queue to obtain test scenarios, wherein the mutation operation involves inserting target vehicles with driving behaviors, including following, overtaking, and turning; inputting the test scenarios into the cooperative tracking system for tracking, calculating the selection metric for each test scenario based on a scenario selection expression; selecting scenarios based on the selection metric to obtain result scenarios; updating the base scenario queue based on the result scenarios, and outputting the tracking results of the result scenarios as a test report; then returning to the step of performing a mutation operation on the base scenario queue to obtain test scenarios, until a preset maximum number of iterations is reached. This application can automatically generate test scenarios suitable for cooperative tracking tasks with continuous spatiotemporal consistency, effectively testing autonomous driving systems and thus improving their robustness. Attached Figure Description

[0018] Figure 1 This is a flowchart of the transformation test method for the autonomous driving cooperative tracking system provided in the embodiments of this application; Figure 2 This is a schematic diagram of the steering driving behavior trajectory provided in the embodiments of this application; Figure 3 This is a schematic diagram of the target point in the turning behavior provided in the embodiments of this application; Figure 4 This is a schematic diagram of a vehicle turning left, provided in an embodiment of this application. Figure 5 This is a schematic diagram of a vehicle turning right, provided in an embodiment of this application. Figure 6 This is a schematic diagram of following vehicle driving behavior provided in an embodiment of this application; Figure 7 This is a schematic diagram of a vehicle following scenario generated according to an embodiment of this application; Figure 8 This is a schematic diagram of the overtaking driving behavior trajectory provided in an embodiment of this application; Figure 9 This is a schematic diagram of a vehicle overtaking scenario generated according to an embodiment of this application. Figure 10 This is a schematic diagram of the structure of the transformation test system of the autonomous driving cooperative tracking system provided in the embodiments of this application; Figure 11 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0019] 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 of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of systems and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0021] Before providing a detailed description of the embodiments of this application, the relevant technologies involved in the embodiments of this application will be described first.

[0022] In related technologies, cooperative tracking systems involve multi-agent collaboration, acquiring environmental information through multiple sources of sensors on the vehicle and roadside. Fusion technology then eliminates the limitations of single sensors through data processing, providing comprehensive and accurate environmental awareness for decision-making. Its architecture and interaction logic are more complex than single-agent systems (which rely solely on their own vehicle to complete the tracking task without sharing data with other cooperating vehicles), thus making them more prone to introducing potential defects. Cooperative tracking systems may malfunction in different scenarios, such as excessive speed, large movements, or specific vehicle trajectories, which can also degrade the system's performance. Errors in cooperative tracking systems can interfere with vehicle operation and even lead to accidents.

[0023] Existing automated testing frameworks cannot generate dynamic scenarios with continuous spatiotemporal consistency suitable for collaborative tracing tasks. Therefore, test data needs to be collected through real vehicle road tests, which is a very challenging and time-consuming process.

[0024] In view of this, this application provides a metamorphic testing method and related equipment for an autonomous driving cooperative tracking system. The method involves: acquiring an initial dataset, which includes several initial scenarios; extracting a preset number of initial scenarios from the initial dataset as base scenarios to obtain a base scenario queue; performing a mutation operation on the base scenario queue to obtain test scenarios, wherein the mutation operation involves inserting target vehicles with driving behaviors, including following, overtaking, and turning; inputting the test scenarios into the cooperative tracking system for tracking, calculating a selection metric for each test scenario based on a scenario selection expression; selecting a scenario based on the selection metric to obtain a result scenario; updating the base scenario queue based on the result scenario, and outputting the tracking results of the result scenario as a test report; then returning to the step of performing a mutation operation on the base scenario queue to obtain test scenarios, until a preset maximum number of iterations is reached. This application can automatically generate test scenarios suitable for cooperative tracking tasks with continuous spatiotemporal consistency, effectively testing autonomous driving systems and thus improving their robustness.

[0025] The transformation test method for an autonomous driving cooperative tracking system provided in this application relates to the field of autonomous driving testing technology. The transformation test method for an autonomous driving cooperative tracking system provided in this application can be applied to a terminal, a server, or software running on a terminal or server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or in-vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the transformation test method for an autonomous driving cooperative tracking system, but is not limited to the above forms.

[0026] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0027] Figure 1 This is an optional flowchart of the transformation test method for the autonomous driving cooperative tracking system provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S106.

[0028] Step S101: Obtain the initial dataset, which includes several initial scenes. Each initial scene includes point cloud frames and labels of the main vehicle and the cooperative vehicles.

[0029] Step S102: Extract a preset number of initial scenes from the initial dataset as basic scenes to obtain a basic scene queue.

[0030] Step S103: Perform a mutation operation on the basic scenario queue to obtain the test scenario. The mutation operation involves inserting a target vehicle with driving behavior, including following, overtaking, and turning.

[0031] Step S104: Input the test scenario into the collaborative tracking system for tracking, and calculate the selection metric for each test scenario based on the scenario selection expression.

[0032] Step S105: Select a test scenario based on the selected metric to obtain the result scenario.

[0033] Step S106: Update the basic scene queue according to the result scene, and output the tracking result of the result scene as a test report. Then return to the step of performing mutation operation on the basic scene queue to obtain the test scene, until the preset maximum number of iterations is reached.

[0034] In this embodiment, the initial scene in the initial dataset is obtained by organizing a dataset based on a preset autonomous driving domain dataset, which includes point cloud frames and labels of the master vehicle and each cooperating vehicle.

[0035] For example, this embodiment selects V2V4Real as the experimental dataset. As the first large-scale real-world multimodal vehicle-to-vehicle cooperative perception platform, this dataset provides effective solutions to key limitations such as occlusion sensitivity and insufficient long-distance perception, aiming to support the research and development of L5 autonomous driving technology.

[0036] This dataset integrates LiDAR point cloud frames, RGB images, labeled 3D bounding boxes, and high-precision maps covering 410 kilometers of real-world driving routes. Researchers can utilize V2V4Real to conduct research on core tasks such as cooperative 3D object detection and cooperative 3D object tracking.

[0037] Next, an equal number of initial scenes are selected from the initial dataset according to the preset number of scenes to construct a basic scene queue. The scenes in the basic scene queue represent continuous 3D point cloud frames and labels generated by the LiDAR system.

[0038] The basic scene queue is subjected to mutation operation to generate a preset number of test scenes. The mutation operation is to insert a target vehicle with driving behavior into the basic scene. One driving behavior is selected from three driving behaviors: following, overtaking, and turning. Constraints are constructed and input into the constraint solver to calculate the position of the target vehicle to be inserted in each point cloud frame of the scene. The vehicle is then rendered at the corresponding position in each point cloud frame using rendering technology to obtain the test scene.

[0039] The generated test scenarios are input into the collaborative tracking system for tracking. The selection metric for each test scenario is calculated based on the scenario selection expression. The test scenario with the higher metric value is selected as the result scenario for each round of output. The tracking result obtained from each round of output is input into the collaborative tracking system as a test report. At the same time, the basic scenario queue is cleared and the basic scenario queue is reassigned to the result scenario. Steps S103 to S106 are repeated until the preset maximum number of iterations is reached.

[0040] In some embodiments, step S103 may include, but is not limited to, steps S201 to S204.

[0041] Step S201: Calculate the coordinates of the target vehicle to be inserted based on the behavioral constraints of the driving behavior, and obtain the insertion position information of the target vehicle in the point cloud frames corresponding to each basic scene. The insertion position information of consecutive different point cloud frames constitutes the driving trajectory of the target vehicle.

[0042] Step S202: Based on the 3D vehicle model, simulate rendering is performed at the corresponding positions in each point cloud frame according to the inserted position information to obtain test cases.

[0043] Step S203: Perform metamorphic testing on the test cases according to the preset metamorphic relationship to obtain test predictions.

[0044] Step S204: Detect whether test cases violate the metamorphosis relationship according to the test oracle, and determine the test cases that satisfy the metamorphosis relationship as test scenarios.

[0045] In this embodiment, the mutation operation specifically includes three main steps: coordinate calculation, scene generation, and mutation testing.

[0046] Coordinate calculation involves determining the insertion position information of the target vehicle within all point cloud frames in the base scene queue. This insertion position information includes spatial coordinates and yaw angles. These positions constitute the driving trajectory, supporting three driving behaviors: following, overtaking, and turning. Each behavior generates multiple [coordinates] under specific behavioral constraints. x , y Coordinates. For example... Figure 2 As shown, in the turning trajectory, the vehicle will pass through in sequence. tarpos 1. tarpos 2. tarpos 3. tarpos 4. tarpos 5. This completes the turning process.

[0047] It should be noted that since inserting a point cloud requires three-dimensional coordinates, the z-coordinate of the target vehicle needs to be determined based on the coordinates of the insertion position information. x , y The z-coordinate is calculated from the road point cloud data, and then the coordinates are matched in the road point cloud. x , y The nearest point cloud region at the corresponding location is used, and the z-value is determined in conjunction with the vehicle height. Therefore, subsequent calculations only focus on the coordinates ( x , y ).

[0048] During coordinate calculation, each driving action has corresponding constraints. Using these constraints, the coordinates and angles of the target vehicle under different driving actions can be calculated for scene generation. Successive point cloud frames form the driving trajectory of the target vehicle under specific driving actions.

[0049] To ensure the realism of the generated scene, each insertion location must meet three criteria: road validity, collision avoidance, and observability. Road validity ensures that the inserted target vehicle is located on the road; collision avoidance checks for any empty overlap between the target vehicle and existing objects in the scene; and observability ensures that the target vehicle's entity can be detected by other cooperating vehicles.

[0050] Based on the calculated insertion position information, vehicles are rendered sequentially, and corresponding labels and point clouds are output.

[0051] For example, after obtaining the insertion location information, MultiTest is used for scene rendering. The specific process is as follows: A simulated LiDAR is invoked to render the target vehicle at the specified location based on the insertion location information, generating point cloud data. Subsequently, all occluded point clouds need to be removed. Finally, the framework generates complete point clouds and annotations for all collaborating vehicles (CVs) and the primary vehicle (ego), and saves the results to obtain test cases for evaluating the collaborative tracking system.

[0052] However, manually verifying the correct execution of the collaborative tracing system on each test case is both time-consuming and labor-intensive. To address this issue, this embodiment utilizes metamorphic relations (MR) to construct MRs between the base scenario and the test scenario, generating test predictions to automatically evaluate system performance. Violations of MRs typically indicate potential defects; therefore, constructing appropriate MRs is crucial for effective metamorphic testing. By detecting MR violations, testing is facilitated, thereby eliminating the need for manual annotation, discarding test cases that violate MRs, and only including test cases that satisfy MRs as test scenarios.

[0053] In some embodiments, step S201 may include, but is not limited to, steps S301 to S304.

[0054] Step S301: When the driving behavior is turning, perform boundary recognition on the road boundary lines to obtain a set of road boundary points used to distinguish the driving directions of different lanes.

[0055] Step S302: Calculate the target turning trajectory based on the turning constraints and the road boundary point set to obtain multiple turning target points.

[0056] Step S303: Based on the detection constraints, determine the initial position of the target vehicle to be inserted from multiple turning target points.

[0057] Step S304: Starting from the initial position, determine a turning target point for each basic scene's corresponding point cloud frame according to the order of the turning target points on the turning trajectory. Adjust the horizontal and vertical coordinates of the target vehicle to be inserted based on the proximity target constraint and the turning target point to obtain the insertion position information.

[0058] In this embodiment, the position of the vehicle refers to the center position in the three-dimensional dimensions of the car, such as... Figure 2 As shown, Figure 2 This shows the target vehicle (represented as) inserted in the V2X cooperative tracking scenario. car insAt the turn, the target vehicle car ins It will pass through each turning target point in sequence. tarpos i This completes the turning process.

[0059] It should be noted that a single base scene contains multiple point cloud frames, and for each frame, the target vehicle needs to be determined. car ins The points are positioned and inserted at the appropriate locations. After insertion, consecutive point cloud frames will form the driving behavior of turning.

[0060] Reference Figure 2 , Figure 2 This section presents a typical V2X cooperative tracking scenario. V2X promotes intelligent road traffic through collaboration between autonomous intelligent agents. The lead vehicle (ego) and collaborating vehicles (CVs) communicate to share road information, expanding perception capabilities and improving tracking performance. To comprehensively test the cooperative tracking system in a V2X environment and ensure that the generated scenario fully utilizes V2X capabilities, this section designs turning and detection constraints.

[0061] In the constraints applied to V2X technology, the first step is to design constraints to ensure that the characteristics of V2X can be used to generate the scene. Under the constraints of turning and detection, road information is used to calculate the turning target point in the scene.

[0062] Specifically, in the turning constraints, V2X-ATCT employs a method for generating target points. tarpos i The method, Figure 3 The process of calculating the target point is shown.

[0063] First, extract the road boundary point set {b1, b2, ..., bn} from the road boundary line to distinguish lanes in different directions. Then, connect these road boundary points sequentially to generate the target point. tarpos For example, 2: Define the road boundary line l1 using points b1 and b2 (i.e., l1 passes through b1 and b2). Then construct a road boundary perpendicular line l2 that is perpendicular to l1 and passes through b2. Finally, calculate the equations for l1 and l2.

[0064] for tarpos2. Solve the constraints based on the equation expression of l2 using the turning constraint conditions. The turning constraint conditions include the turning position sub-condition, the boundary distance sub-condition, and the turning direction sub-condition. The turning constraint conditions are valid when all three conditions are met simultaneously. The turning position sub-condition is that the turning target point is located on the perpendicular line of the road boundary. The boundary distance sub-condition is that the distance between the turning target point and the road boundary point does not exceed the road boundary threshold. The turning direction sub-condition is that the turning target point is located on one side of the lane's driving direction.

[0065] For example, the turning position sub-condition requires tarpos The coordinates of 2 ( tarpos 2. x , tarpos 2. y On the online L2, this sub-condition can be expressed as: tarpos 2. x * k 2+ b 2= tarpos 2. y (4-1) Boundary distance subcondition representation tarpos The distance between b2 and b2 is within the threshold, thus avoiding the target vehicle. car ins The sub-condition of driving beyond the road boundary can be expressed as: car ins . x / 2< Dis ( tarpos 2, b2) < N * lane . w / 2- car ins . w / 2; (4-2) in, lane . w Indicates lane width, car ins . w Indicates the width of the vehicle. N Indicates the total number of lanes. Dis (·) represents the function for calculating distance.

[0066] Assuming the target vehicle car ins Driving in the right lane, the vector cross product principle is used to ensure [the correct direction of travel]. tarpos Position 2 is located on the right side of the road. This sub-condition can be expressed as: ( tarpos 2. x -b 2. x )* b -( tarpos 2. y - b 2. y )* a >0. (4-3) set up a = b 3. x - b 2. x and b = b 3. y - b 2. y .

[0067] By solving constraints (4-1) and (4-2), two candidate coordinates are obtained, and one that violates constraint (4-3) is discarded. The candidate coordinates that satisfy constraint (4-3) are then selected. tarpos 2. Assign values ​​to ultimately determine the turning target point. tarpos 2. For subsequent turning target points, such as... tarpos 3. tarpos 4. Repeat this process until all turning target points are generated.

[0068] For turning scenarios, V2X-ATCT can theoretically use any turning target point as its initial coordinates. To ensure the effectiveness of scenario generation, the target vehicle is set through detection constraints. car ins The initial position needs to be cav i Detected, i.e., selected tarpos i When used as initial coordinates, it needs to be determined by at least one machine. cav i The detection constraints include a perception time sub-condition and an occlusion sub-condition. The detection constraints are valid when both the perception time sub-condition and the occlusion sub-condition are met. The perception time sub-condition is that the detection time is greater than the preset minimum perception time threshold. The occlusion sub-condition is that at least one master vehicle or cooperative vehicle detects that the occlusion rate of the target vehicle is less than the preset occlusion rate threshold.

[0069] For example, the detection constraints require the target vehicle car ins Need to be cav i Continuous monitoring has revealed that, among them, cav i ∈{ego,CVs}.

[0070] The perceptual time subcondition specifies cav i Keep detected car ins The shortest time, this sub-condition can be expressed as: T per > MPT ∧ Count per ≥1; (4-3) in, T per For the scene cav i Target vehicle detected car ins The actual time MPT The minimum perception time threshold, Count per To detect the target vehicle car ins of cav i quantity.

[0071] The conditions for blocking the sub-blocks are specified cav i Target vehicle detected car ins The required conditions, i.e., the target vehicle car ins The condition that the object cannot be completely obscured and must remain within a certain distance can be expressed as: Dis ( cav i , car ins ) < DRL ∧ Occ ( car ins ) < ORT (4-4) in, DRL This is the LiDAR detection range. Occ (·) is the occlusion rate calculation function. ORT It is the preset occlusion rate threshold.

[0072] Target vehicle car ins V2X-ATCT selects a specific target point as the initial position, then iterates through the remaining points until the final target is reached. This is achieved through comparison. x , y The position in two directions determines the direction of movement. Specifically, the position in the current frame... x Target vehicle in the axial direction car ins Position and turning target point tarpos i Calculate the difference between the positions and compare them, then add the corresponding value (△). x Adjustments are made to get closer to the target turning point. tarpos i , y Axial Direction Principle and x The axes are consistent, so I won't go into details here.

[0073] Once the target vehicle car ins Location and target point tarpos i The positional difference is at x , y The differences in direction are all less than the threshold. δ This indicates the target vehicle. car ins We have reached the turning point.

[0074] x-axis direction adjustment (△) x >0): (4-6) (4-7) in, express car ins In the i The position of the frame.

[0075] This embodiment supports deployment in diverse traffic environments and can generate various turning trajectories. When the target vehicle to be inserted performs a left turn, the position information of all frames in the scene is calculated based on the above constraints, resulting in the following: Figure 4 The left turn scenario shown in the image indicates that the target vehicle is circled in red. car ins According to the order of the sub-images from left to right and from top to bottom, the target vehicle is... car ins Drive straight for a certain distance before turning left.

[0076] When the target vehicle to be inserted performs a right turn, the position information of all frames in the scene is calculated based on the above constraints, resulting in the following: Figure 5 As shown in the right turn scenario, the target vehicle can be identified by following the order of the sub-images from left to right and from top to bottom. car ins Go straight for a certain distance before turning right.

[0077] In some embodiments, step S201 may include, but is not limited to, steps S401 to S403.

[0078] Step S401: When the driving behavior is following another vehicle, randomly select one vehicle from the main vehicle and the cooperating vehicle as the target vehicle to be inserted, and establish a local coordinate system in the point cloud frame corresponding to each basic scene with the target vehicle as the origin.

[0079] Step S402: Determine the initial position of the target vehicle to be inserted based on the vehicle following constraint conditions.

[0080] Step S403: Starting from the initial position, adjust the horizontal axis coordinates of the target vehicle according to the approach target constraint and the speed difference between the target vehicle and the following target in two adjacent point cloud frames to obtain the insertion position information.

[0081] In this embodiment, refer to Figure 6 Following another vehicle is a form of driving that targets the vehicle. car ins Randomly select one vehicle from the main vehicle (ego) and the partner vehicle (CV) in the basic scene as the target to follow, and maintain a reasonable distance until the scene ends.

[0082] In order to effectively evaluate V2X technology in the generated scene, the following vehicle must belong to the set of vehicles participating in cooperative perception, as shown in constraint (4-8): Car bf ∈ V (4-8) in, Car bf In order to follow the target, V This represents the set of vehicles participating in collaborative perception.

[0083] The tracked target Car bf Establish the origin with the center as the point. Figure 6 The local coordinate system shown, where the target is followed Car bf Its forward direction is the positive horizontal axis, following the target. Car bf The left side of the direction of movement is the positive direction of the vertical axis.

[0084] It should be noted that in each frame, following the target Car bf The center position is defined as the origin.

[0085] Assume the first i The target vehicle performing the following action within the frame car ins The location is marked as .

[0086] For initial coordinate generation, the following constraint condition is used to ensure that the target vehicle always follows the specified target and maintains a certain distance. The following constraint condition includes a following position sub-condition and a following distance sub-condition. The following constraint condition is valid when both the following position sub-condition and the following distance sub-condition are met. The following position sub-condition is that the horizontal axis coordinate of the target vehicle is less than the horizontal axis coordinate of the target vehicle and the vertical axis coordinate of the target vehicle is equal to the vertical axis coordinate of the target vehicle. The following distance sub-condition is that the absolute value of the difference between the horizontal axis coordinates of the target vehicle and the target vehicle does not exceed the following distance threshold.

[0087] For example, the following position subcondition ensures the target vehicle car ins The position is directly located at the target being followed Car bf Following this, the sub-condition can be expressed as: (4-9) in, Indicates the target vehicle car ins x-axis coordinates Indicates the target vehicle car ins The vertical axis coordinate.

[0088] Following distance subcondition for target vehicle car ins With the target Car bf The current following distance between vehicles is limited to a predefined maximum and minimum threshold range. This sub-condition can be expressed as: (4-10) (4-11) (4-12) in, MinDistance and MaxDistance The minimum and maximum limiting distances are calculated using equations (4-11) and (4-12), respectively. Here, the limiting distance refers to the distance between the centers of the two vehicles. car . l Indicates the length of the vehicle. MinFD Indicates the minimum following distance. MaxFD This indicates the maximum following distance.

[0089] After determining the initial position, in the subsequent process of generating the vehicle-following motion coordinates, using... and They represent the first i Frame and the i Target vehicle in frame +1 carins The coordinates of the location.

[0090] For following vehicle driving behavior, the approach target constraint conditions include a first speed relationship sub-condition, a second speed relationship sub-condition, and a third speed relationship sub-condition. The first speed relationship sub-condition is that the speed of the target vehicle is less than the speed of the following target, the second speed relationship condition is that the speed of the target vehicle is greater than the speed of the following target, and the third speed relationship condition is that the speed of the target vehicle is equal to the speed of the following target.

[0091] For example, the target vehicle car ins The relative position is dynamically adjusted according to the speed: if Car ins . v < Car bf . v ,but ,in, dx >0.

[0092] if Car ins . v > Car bf . v ,but .

[0093] if Car ins . v = Car bf . v The target vehicle in the next frame car ins The coordinates remain unchanged.

[0094] At the same time, the calculated coordinates must also meet the following distance sub-condition: (4-13) (4-14) It should be noted that velocity calculation requires converting the coordinates from the local coordinate system to the world coordinate system: ; in, Represents the coordinate values ​​in the world coordinate system. This represents the transformation matrix used to transform coordinates from the local coordinate system to the world coordinate system.

[0095] The formula for calculating speed is defined as follows: ; Among them, the i The speed of +1 frame is represented as v i+1 , △ t This indicates the time interval between consecutive frames.

[0096] Based on the above constraints, the position information of all frames in the scene can be calculated to obtain the following: Figure 7 The image shows a vehicle following scenario, in which the target vehicle... car ins Choose one vehicle in the scene and follow it, maintaining a reasonable distance, until the scene ends. The target vehicle is circled in red. car ins According to the order of the sub-images from left to right and from top to bottom, the target vehicle is... car ins Following the ego car ahead.

[0097] In some embodiments, step S201 may include, but is not limited to, steps S501 to S503.

[0098] Step S501: When the driving behavior is overtaking, a vehicle is randomly selected from the main vehicle and the cooperating vehicle as the target vehicle to be inserted, and a local coordinate system is established in the point cloud frame corresponding to each basic scene with the overtaking target as the origin.

[0099] Step S502: Calculate the overtaking trajectory based on the overtaking constraints to obtain four overtaking target points. The overtaking target points include the initial position, lane change position, acceleration position, and end position. The overtaking constraints include the overtaking position sub-condition, the lane change collision avoidance sub-condition, and the speed limit sub-condition.

[0100] Step S503: Starting from the initial position, determine an overtaking target point for each basic scene point cloud frame according to the order of overtaking target points, and adjust the horizontal and vertical coordinates of the target vehicle to be inserted according to the approach target constraint and the overtaking target point to obtain the insertion position information.

[0101] In this embodiment, in the overtaking driving scenario, one of the lead vehicle and the cooperating vehicle is first randomly selected as the overtaking target. Subsequently, four overtaking target points are calculated, such as... Figure 8 As shown, the four overtaking target points are designated P1 to P4. Target vehicle car ins It will pass through these four overtaking target points in sequence to perform overtaking maneuvers.

[0102] The overtaken target is recorded as Car boAnd establish a local coordinate system with its center as the origin, such as Figure 8 As shown, exceeding the target Car bo Its direction of travel is the positive horizontal axis, and it is moving past the target. Car bo The left side of the forward direction is the positive direction of the vertical axis.

[0103] During the overtaking process, the target vehicle car ins First, change lanes, then accelerate to overtake the target. Car bo The final target vehicle car ins Return to the original lane.

[0104] It should be noted that, in order to effectively verify V2X technology in the generated scenario, the vehicle being overtaken must belong to the vehicles participating in cooperative perception, as defined by constraint (4-15): Car bo ∈ V (4-15) In addition, to ensure the smooth execution of the overtaking maneuver and to guarantee a certain duration, this embodiment designs an additional speed limit sub-condition to restrict the speed of the overtaking operation within a specified range, as shown in constraint (4-16): car bo . v × α ≤ car ins . v ≤ car bo . v × β (4-16) This speed limit subcondition remains in effect throughout the generation of the four overtaking target points.

[0105] Specifically, the four overtaking target points are the initial position P1, the lane change position P2, the acceleration position P3, the ending position P4, and the target vehicle. car ins Pass through these points in sequence, and end the overtaking maneuver when reaching P4.

[0106] For the initial generation of coordinate points during overtaking, the overtaking position sub-condition is applied to ensure that point P1 is located at the overtaking target. Car bo The overtaking position sub-condition, which is behind and within a reasonable following range, uses the same logic and constraints (4-9) and (4-10) as the following constraint conditions in the following behavior, that is, ensuring that the target vehicle is within the reasonable following range at the initial position P1. carins The position is directly above the target. Car bo Behind, at the same time, the target vehicle car ins and exceeding the goal Car bo The current following distance between vehicles is limited to a predefined maximum and minimum threshold range. This overtaking position sub-condition can be expressed as: P1. x < car bo . x ∧P1. y = car bo . y (4-17) MinDistance ≤|P1. x - car bo . x |≤ MaxDistance (4-18) Furthermore, to prevent a collision during the subsequent movement towards P2, the overtaking position sub-condition needs to be further verified to ensure that the target vehicle is within range. car ins When the front bumper reaches P1, it is in contact with the target. Car bo The rear bumper must maintain a distance of at least one lane width; that is, this sub-condition also includes: |P1. x - car bo . x |> car bo . l / 2+ lane . w + car ins . l / 2. (4-19) For the lane change position P2 during overtaking, the overtaking position sub-condition ensures that point P2... x The position in the direction is located at P1 and beyond the target. Car bo Between, and ensure that the target vehicle is at point P2. car ins and exceeding the goal Car bo The distance along the y-axis is equal to the width of one lane; this sub-condition can be expressed as: P1. x <P2. x < carbo . x (4-20) P2. y = car bo . y + lane . w (4-21) Similarly, to prevent collisions, the overtaking position must also ensure that the target vehicle is at point P2. car ins The front bumper does not exceed the target Car bo The rear bumper, i.e., this sub-condition also includes: |P2. x - car bo . x |> car bo . l / 2+ car ins . l / 2. (4-22) For the ending position P4 in the overtaking maneuver, the overtaking position sub-condition is similar to the constraint of P1, but the overtaking position sub-condition requires P4 to be within the specified range. x Direction is above the target Car bo The overtaking position sub-condition can be expressed as follows: P4. x > car bo . x ∧P4. y = car bo . y (4-23) MinDistance ≤|P4. x - car bo . x |≤ MaxDistance (4-24) |P4. x - car bo . x |> car bo . l / 2+ lane . w + car ins . l / 2. (4-25) For the acceleration position P3 in the overtaking maneuver, its overtaking position sub-condition is similar to the constraint of P2, but the overtaking position sub-condition requires P3 to be in... x Direction is above the target Car bo The overtaking position sub-condition, which is between P4 and P5, can be expressed as: car bo . x <P3. x <P4. x (4-26) |P3. x - car bo . x |> car bo . l / 2+ car ins . l / 2; (4-27) P3. y = car bo . y + lane . w (4-28) In the subsequent coordinate generation process, P1 to P4 are sequentially assigned to the point cloud frames corresponding to each basic scene, designated as the current target points, and represented as... tarPj Target vehicle car ins It will move towards the current target point according to the target proximity constraint, which can be expressed as: x Axial direction adjustment: (4-29) (4-30) y The adjustment of the axial direction is logically consistent with the adjustment of the x-axis direction, and will not be described in detail in this embodiment.

[0107] When the target vehicle car ins Upon reaching an overtaking target point, it will switch to the next target and continue moving towards it, identifying the target vehicle. car ins The conditions for reaching the target point are the same as those for the aforementioned turning behavior, when the target vehicle... car ins The position difference between the target point and the overtaking point is x , y The differences in direction are all less than the threshold. δThis indicates the target vehicle. car ins The overtaking target point has been reached, and the overtaking action ends when the final target point (P4) is reached.

[0108] Furthermore, during the movement from P1 to P2 or from P3 to P4, to avoid the target vehicle... car ins and exceeding the goal Car bo In the event of a collision, this embodiment also imposes constraints through a lane change collision avoidance sub-condition, which requires the target vehicle... car ins exist x The time required for travel in the axial direction is less than or equal to that in y The time required for travel in the axial direction can be expressed as a time constraint: t x ≤ t y ; in, t x Indicates in x The time required to travel in the direction of travel, t y Indicates in y The time required to travel in the specified direction.

[0109] For example, taking P1 to P2 as an example, if ,but t x = n *△ t , t y The same calculation logic applies, so it will not be elaborated further.

[0110] Based on the above constraints, the position information of all frames in the scene can be calculated to obtain the following: Figure 9 The image shows a vehicle overtaking scenario, in which the target vehicle... car ins Choose a car in the scene to overtake; the target car will pass by in sequence as follows: Figure 8 The four overtaking target points are shown, with the target vehicle circled in red. car ins According to the order of the sub-images from left to right and from top to bottom, the target vehicle is... car ins Overtaking the ego car ahead.

[0111] In some embodiments, step S203 may include, but is not limited to, steps S601 to S602.

[0112] Step S601: The true trajectory result of the target vehicle is merged with the original tracking result of the basic scene to obtain the first test result.

[0113] Step S602: Match the first test result and the test tracking result of the test case with the preset metamorphic relationship with soft equality criterion to obtain the test prediction. The metamorphic relationship includes that inserting a target vehicle does not change the original tracking result of the cooperative tracking system, and the true trajectory result is accurately tracked in the test tracking result.

[0114] In this embodiment, various test cases can be generated through the aforementioned steps. However, manually verifying whether the system executes correctly on each test case is both time-consuming and labor-intensive. This embodiment utilizes metamorphic relationships to generate test predictions in order to automatically evaluate system performance.

[0115] Specifically, this embodiment defines a specific metamorphic relation MR for multi-target tracking tasks in dynamic perception, where CT represents a V2X-based cooperative tracking system designed to track large-scale real-world multimodal V2X cooperative perception datasets. Multiple targets within. Given a set of trajectories. and an object database For each trajectory Assign a unique target instance .

[0116] The significance of this transformation relationship is that for all newly added trajectories, all generated test cases will automatically complete the labeling, that is, generate the true value result of the trajectory of the inserted target vehicle. Then, the tracking system tracks the basic scenario of the target vehicle to be inserted (at this time, the target vehicle has not yet been inserted) to obtain the original tracking result. The original tracking result is merged with the trajectory true value result to obtain the first test result.

[0117] The test cases for the target vehicle are then input into the tracking system for tracking, and the test tracking results of the test cases are obtained. Theoretically, the first test result and the test tracking result should be equal. If they are not equal, it means that a problem has occurred in the tracking process.

[0118] This transformation relationship also needs to take into account the issue of other vehicles remaining unchanged, that is, after performing any driving operation, other vehicles in the tracking results should not be affected.

[0119] Therefore, the pre-established metamorphic relationship is based on two core principles: (1) Inserting an object should not change the original correct tracking result of the CT system, that is, inserting a target vehicle does not change the original tracking result of the collaborative tracking system.

[0120] (2) The inserted object should be accurately tracked, that is, the true value of the trajectory should be accurately tracked in the test tracking result.

[0121] However, due to subtle drift in the tracking results, forcing strict equality of CT output results may lead to a high false alarm rate. To address this issue, this embodiment employs a soft equality criterion, utilizing a criterion-based approach. MR 1. Construct effective test predictions to facilitate testing by detecting violations of MR, thereby eliminating the need for manual annotation.

[0122] The soft equality criterion primarily refers to the requirement that 3D bounding boxes should approximately match. In a test case, the 3D IoU threshold is the "passing grade" for judging whether the 3D bounding boxes (predicted values) output by the tracking algorithm match the ground truth 3D bounding boxes. The system considers a match successful only when the Intersection over Union (IoU) between the predicted and ground truth boxes in 3D space is greater than or equal to 0.25. The soft equality criterion, on the other hand, requires that the results before and after comparison be approximately equal. That is, for the same target, the 3D IoU between all bounding boxes in result 1 and all bounding boxes in result 2 should be greater than or equal to 0.25.

[0123] For example, this metamorphic relationship MR 1 can be formalized as: ; in, This represents an insertion operator that follows the corresponding trajectory. target instance o Put into dataset M middle, GT O Indicate target o The truth value of the object includes its bounding box, object identity, and frame number. It is a soft equivalence criterion used to evaluate the equivalence of CT results between the original dataset and the modified dataset.

[0124] In some embodiments, step S104 may include, but is not limited to, steps S701 to S703.

[0125] Step S701: Input the test scenario into the collaborative tracking system to calculate the crossover ratio and average value of several tracking objects to be detected, and obtain the average tracking crossover ratio.

[0126] Step S702: Calculate the error rate based on the number of false positives and false negatives of the tracked objects.

[0127] Step S703: The average tracking crossover ratio and error rate are weighted and calculated to obtain the selection metric for each test scenario.

[0128] In this embodiment, to improve testing efficiency, a V2X collaborative tracking guided selection algorithm is also provided. This algorithm comprehensively considers the detection error ratio and the average tracking intersection-union ratio, aiming to screen test cases with poor performance in both accuracy and precision dimensions.

[0129] Specifically, the average tracking crossover ratio is calculated using the following formula: ; in, n This represents the total number of objects to be detected in the scene. IoU i This indicates the number tracked by the collaborative tracking system. i The IoU (Intersection over Union) value between an object and the ground truth of its corresponding target.

[0130] The error rate is calculated using the following formula: ; in, It is the number of ground truths in all frames of the scene. FP (False Positive) and FN (False Negative) refer to the number of false positives (also known as false detections) and false negatives (also known as false misses).

[0131] For each test case, the selection metric is calculated using the scenario selection expression. F sm So that, based on the calculations obtained F sm The test scenarios are selected based on the use cases with higher metrics. The formula for the scenario selection expression is as follows: ; Wherein, the weights satisfy .

[0132] The following is a detailed description and explanation of the solutions in the embodiments of the present invention, using specific application examples: Current automated testing frameworks are unable to generate dynamic scenes with continuous spatiotemporal consistency suitable for cooperative tracking tasks. This application addresses the lack of effective and comprehensive testing methods for LiDAR cooperative tracking systems by selecting four cooperative tracking systems as research objects, including CoBEVT based on a single-sensor cooperative tracking architecture and DMSTrack based on a multi-sensor cooperative tracking architecture, as well as two variants of DMSTrack: DMSTrackbev and DMSTrackpos.

[0133] Two variants use either local bird's-eye view (BEV) features or positional features as input to the covariance neural network, respectively. All of these collaborative perception models are based on LiDAR technology and use multi-view point clouds as input.

[0134] Before explaining this solution in detail, a brief introduction to each system will be given first.

[0135] (1) CoBEVT is the first general-purpose multi-agent, multi-camera collaborative perception framework capable of collaboratively generating bird's-eye view semantic map predictions. To efficiently fuse camera features from multiple perspectives and agents within a unified Transformer architecture, CoBEVT proposes a Fusion Axial Attention module, which can simultaneously capture sparse local and global spatial interactions across perspectives and agents. Extensive experiments on the V2V perception benchmark dataset OPV2V demonstrate that CoBEVT achieves state-of-the-art performance on collaborative BEV semantic segmentation tasks.

[0136] Furthermore, CoBEVT also demonstrated strong generalization capabilities in other tasks: single-vehicle BEV segmentation under multi-camera input; and LiDAR-based multi-agent 3D object detection—both achieving leading results in real-time inference speed. Chiu et al. used CoBEVT as the object detector, combined with the AB3DMOT tracking algorithm based on a single-sensor Kalman filter as the object tracker, forming a strong baseline. The core of AB3DMOT is the combination of a 3D Kalman filter and the Hungarian algorithm, directly performing state estimation and trajectory management in 3D space.

[0137] (2) DMSTrack is a novel 3D multi-target cooperative tracking algorithm based on a differentiable multi-sensor Kalman filter, designed specifically for autonomous driving scenarios. Existing V2V cooperative perception methods typically input the detection results of multiple connected autonomous vehicles into a standard single-sensor Kalman filter for tracking, often ignoring the heterogeneous measurement uncertainties from sensors of different vehicles, thus limiting the realization of the theoretically optimal performance of the Kalman filter.

[0138] In contrast, DMSTrack, within a differentiable multi-sensor Kalman filter framework, explicitly learns the measurement uncertainty for each detection result, thereby achieving a more reasonable and superior fusion of multi-vehicle observation information. Experiments on the V2V4Real dataset show that DMSTrack improves tracking accuracy by 17%, while the communication overhead is only 0.037 times that of the current state-of-the-art methods (i.e., approximately 27 times lower communication costs).

[0139] The transformation test method for the autonomous driving cooperative tracking system provided in this embodiment first obtains an initial scene. The initial scene is obtained based on a pre-set dataset in the field of autonomous driving. The dataset includes point cloud frames and labels of the master vehicle and each cooperative vehicle.

[0140] The steps to generate a test scenario based on the initial scenario are as follows: Step 1: Select a preset number of scenes from the initial scene to build a basic scene queue. The scenes in the basic scene queue represent continuous 3D point cloud frames and labels generated by the LiDAR system.

[0141] Step 2: Perform mutation operations on all scenarios in the basic scenario queue to generate a preset number of test scenarios. The mutation operation involves inserting vehicles with driving behavior into the basic scenarios. The implementation module for this step specifically includes: (1) Location information calculation module: Select a driving behavior, construct constraints, input the constraints into the constraint solver to calculate the location information of the vehicle to be inserted in each point cloud frame of the scene. The location information includes spatial position and yaw angle.

[0142] (2) Scene generation module: Based on the location information, select a three-dimensional vehicle model, use simulated LiDAR rendering technology to render the vehicle at the corresponding position of each point cloud frame, and update the corresponding label.

[0143] (3) Transformation Test Module: Constructs the transformation relationship between the original scene and the test scene, and generates test predictions; Step 3: Input the test scenarios into the collaborative tracking system for tracking, calculate the scenario selection expression for each test scenario, select the test scenario as the result scenario according to the scenario selection expression, clear the basic scenario queue, and reassign the result scenario to the basic scenario queue.

[0144] Step 4: Repeat steps 1 to 3 until the preset maximum number of iterations is reached. Output the result scenario for each round and output the tracking results corresponding to each round's result scenario as a test report.

[0145] Specifically, the location information calculation module is used to calculate the coordinate positions of the inserted vehicle in all frames of the driving scene; these positions constitute the driving trajectory. Each driving action has corresponding constraints. Using these constraints, the corresponding coordinates and angles can be calculated, which will be used for scene generation. Because Microsoft's Z3 solver is fast and flexible, this embodiment uses the Z3 solver to solve the constraints.

[0146] The specific calculation process involves inputting the designed constraints into the Z3 solver for calculation, obtaining the corresponding coordinate values, and repeating this process to obtain all the coordinates of the scene. These continuous coordinates constitute the trajectory in the driving scene.

[0147] The scene generation module is mainly used for effective location calculation and LiDAR simulation and point cloud insertion.

[0148] During the effective location calculation phase, to ensure the authenticity of the generated scene, the location information must meet the requirements of road effectiveness, collision avoidance, and observability.

[0149] Road validity: First, a semantic segmentation model is applied to the point clouds of the master vehicle and each cooperating vehicle to extract the drivable road area. Then, it is determined whether each location information is in the road area, and only the location information located on the road is retained.

[0150] Collision avoidance: For each location, a 3D bounding box is generated based on the size and location information of the 3D vehicle model, and its spatial overlap with existing traffic participants in the scene is checked. A location is only accepted if the bounding box has an IoU of 0 with all traffic participants in the scene.

[0151] Observability: Requires at least one collaborating vehicle or master vehicle to observe that the occlusion rate of the inserted 3D vehicle is less than 90%.

[0152] If the location information violates the above validity, it is recalculated based on the constraints and the constraint solver, and the validity is judged. This step is repeated until all the location information meets the above validity.

[0153] During the LiDAR simulation and point cloud insertion phase, the entity is rendered using a set of high-fidelity virtual LiDARs based on the location information. These virtual sensors simulate the LiDAR configuration in a real V2X scenario, including the field of view (FOV), the number of laser beams, and the angular resolution.

[0154] For example, taking a turning scenario, constraint (4-2) ensures that the target position is maintained during each scene generation process. tarpos i The diversity, and the insertion of target vehicles in each frame. car insThe displacement introduces reasonable fluctuations, thereby ensuring the diversity of the generated test cases.

[0155] The V2X communication system establishes a unified world coordinate system for aligning perception data among multiple agents. Each vehicle acquires point clouds and perception results in its own local coordinate system, and the system uses a pose transformation matrix to perform bidirectional transformations between the local and world coordinate systems. This mechanism ensures that any newly inserted entity (such as the target vehicle) is correctly positioned. car ins It maintains geometric and spatial consistency from the perspective of all participating vehicles.

[0156] By dynamically inserting entities into the original multi-view collaborative perception scene, a new test scene with reasonable semantics is generated. The original scene contains point cloud data collected by the vehicle and several cooperating vehicles.

[0157] The insertion process includes the following two key steps: Step 1: Calculation of Effective Location. The system first applies the semantic segmentation model Salsanext to the point clouds of the ego vehicle and each CV vehicle to extract the drivable road region. Then, the candidate insertion points are projected onto the world coordinate system, and only those points that are located on the road surface in all views are retained, ensuring that the inserted vehicle is located on the road.

[0158] Subsequently, for each candidate location, the system generates a 3D bounding box based on the size and orientation of the entity to be inserted, and checks whether it spatially overlaps with existing objects in the scene. A location is only accepted if the bounding box has an IoU of 0 with all traffic participants.

[0159] Meanwhile, to ensure that the entity can be detected, V2X-ATCT evaluates the occlusion of the entity from all CV viewpoints and requires that at least one CV observes that the occlusion rate of the entity is less than 90%.

[0160] Step 2: LiDAR Simulation and Point Cloud Insertion. Once a valid location is selected, the system uses a set of high-fidelity virtual LiDAR to render the entity. The system employs a ray casting algorithm to simulate laser beam emission from the perspective of each participating vehicle, generating a point cloud that conforms to the geometric relationship of that perspective; removing background point clouds occluded by the 3D vehicle; and simultaneously removing portions of the 3D vehicle that are occluded by other objects in the scene. After the point cloud rendering operation is completed, the labels are updated frame by frame. For each frame of data, if the 3D vehicle is within the perception range of the main vehicle or a cooperating vehicle, the system synchronously adds the entity's spatial position, speed, yaw angle, and 3D bounding box status information to the vehicle's label file.

[0161] Ultimately, the framework generates complete point clouds and annotations for all collaborating vehicles (CVs) and the primary vehicle (ego), and saves the results. Through iterative processing of all data frames, this framework aims to ensure complete spatiotemporal coherence of the entire augmented scene (i.e., the test scene after inserting the target vehicle).

[0162] Through the above-described transformation test method, this embodiment can automatically generate test scenarios with continuous spatiotemporal consistency suitable for collaborative tracking tasks, effectively test the autonomous driving system, and thus improve its robustness.

[0163] Reference Figure 10 This application also provides a transformation test system for an autonomous driving cooperative tracking system, which can implement the above-mentioned method. The system includes: The first module is used to obtain the initial dataset, which includes several initial scenes. Each initial scene includes point cloud frames and labels of the main vehicle and the cooperative vehicles. The second module is used to extract a preset number of initial scenes from the initial dataset as base scenes to obtain a base scene queue. The third module is used to perform mutation operations on the basic scenario queue to obtain test scenarios. The mutation operation involves inserting target vehicles with driving behaviors, including following, overtaking, and turning. The fourth module is used to input test scenarios into the collaborative tracking system for tracking, and to calculate the selection metric for each test scenario based on the scenario selection expression. The fifth module selects test scenarios based on the selected metrics to obtain the result scenarios; The sixth module updates the basic scenario queue based on the result scenario, outputs the tracking results of the result scenario as a test report, and then returns to the step of performing mutation operations on the basic scenario queue to obtain the test scenario, until the preset maximum number of iterations is reached.

[0164] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0165] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0166] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0167] Reference Figure 11 , Figure 11 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 901 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0168] The memory 902 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 902 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and is called and executed by the processor 901 using the methods described in the embodiments of this application.

[0169] The input / output interface 903 is used to implement information input and output.

[0170] The communication interface 904 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0171] Bus 905 transmits information between various components of the device, such as processor 901, memory 902, input / output interface 903, and communication interface 904.

[0172] The processor 901, memory 902, input / output interface 903, and communication interface 904 are connected to each other within the device via bus 905.

[0173] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0174] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0175] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0176] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0177] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0178] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0179] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0180] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0181] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0182] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0183] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0184] In the embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0185] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0186] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0187] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0188] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A transformation test method for an autonomous driving cooperative tracking system, characterized in that, The method includes the following steps: Obtain an initial dataset, wherein the initial dataset includes several initial scenes, and each initial scene includes point cloud frames and labels of the main vehicle and the cooperative vehicles; A predetermined number of initial scenes are extracted from the initial dataset as base scenes to obtain a base scene queue; The basic scenario queue is subjected to a mutation operation to obtain a test scenario, wherein the mutation operation is to insert a target vehicle with driving behavior, including following, overtaking and turning. The test scenarios are input into the collaborative tracking system for tracking, and the selection metric for each test scenario is calculated based on the scenario selection expression. The test scenario is selected based on the selection metric to obtain the result scenario; The base scenario queue is updated based on the resulting scenario, and the tracking results of the resulting scenario are output as a test report. Then, the process returns to the step of performing a mutation operation on the base scenario queue to obtain the test scenario, until the preset maximum number of iterations is reached.

2. The method according to claim 1, characterized in that, The process of performing a mutation operation on the basic scenario queue to obtain the test scenario includes the following steps: Based on the behavioral constraints of driving behavior, the coordinates of the target vehicle to be inserted are calculated to obtain the insertion position information of the target vehicle in the point cloud frame corresponding to each of the basic scenarios. The insertion position information of consecutive different point cloud frames constitutes the driving trajectory of the target vehicle. Based on the 3D vehicle model, test cases are obtained by simulating rendering at the corresponding positions in each point cloud frame according to the insertion position information. The test cases are subjected to metamorphic tests based on the preset metamorphic relationship to obtain test predictions. The test oracle detects whether the test cases violate the metamorphic relationship, and the test cases that satisfy the metamorphic relationship are identified as test scenarios.

3. The method according to claim 2, characterized in that, The step of calculating the coordinates of the target vehicle to be inserted based on the behavioral constraints of driving behavior to obtain the insertion position information of the target vehicle in the point cloud frame corresponding to each of the basic scenes includes the following steps: When the driving behavior is turning, the road boundary line is identified to obtain a set of road boundary points used to distinguish the driving directions of different lanes; The turning trajectory is calculated based on the turning constraints and the road boundary point set to obtain multiple turning target points. The turning constraints include turning position sub-conditions, boundary distance sub-conditions, and turning direction sub-conditions. The turning constraints are valid when all three conditions are met simultaneously. The turning position sub-condition is that the turning target point is located on the perpendicular line of the road boundary. The boundary distance sub-condition is that the distance between the turning target point and the road boundary point does not exceed the road boundary threshold. The turning direction sub-condition is that the turning target point is located on one side of the lane's driving direction. The initial position of a target vehicle to be inserted is determined from multiple turning target points according to the detection constraints. The detection constraints include a perception time sub-condition and an occlusion sub-condition. The detection constraints are valid when both the perception time sub-condition and the occlusion sub-condition are met. The perception time sub-condition is that the detection time is greater than a preset minimum perception time threshold. The occlusion sub-condition is that at least one master vehicle or cooperative vehicle detects that the occlusion rate of the target vehicle is less than a preset occlusion rate threshold. Starting from the initial position, according to the order of the turning target points on the turning trajectory, a turning target point is determined sequentially for each point cloud frame corresponding to the basic scene. The horizontal and vertical coordinates of the target vehicle to be inserted are adjusted according to the approach target constraint and the turning target point to obtain the insertion position information.

4. The method according to claim 2, characterized in that, The step of calculating the coordinates of the target vehicle to be inserted based on the behavioral constraints of driving behavior to obtain the insertion position information of the target vehicle in the point cloud frame corresponding to each of the basic scenes includes the following steps: When the driving behavior is following another vehicle, a vehicle is randomly selected from the main vehicle and the cooperating vehicle as the target vehicle to be inserted. A local coordinate system is established in the point cloud frame corresponding to each of the basic scenes with the target vehicle as the origin. The direction of movement of the target vehicle is the positive direction of the horizontal axis, and the left side of the direction of movement is the positive direction of the vertical axis. The initial position of the target vehicle to be inserted is determined according to the following constraints, wherein the following constraints include a following position sub-condition and a following distance sub-condition. When both the following position sub-condition and the following distance sub-condition are met, the following constraints are met. The following position sub-condition is that the horizontal axis coordinate of the target vehicle is less than the horizontal axis coordinate of the following target and the vertical axis coordinate of the target vehicle is equal to the vertical axis coordinate of the following target. The following distance sub-condition is that the absolute value of the difference between the horizontal axis coordinates of the target vehicle and the following target does not exceed the following distance threshold. Starting from the initial position, the horizontal axis coordinates of the target vehicle are adjusted according to the proximity constraint and the speed difference between the target vehicle and the following target in two adjacent point cloud frames to obtain the insertion position information.

5. The method according to claim 2, characterized in that, The step of calculating the coordinates of the target vehicle to be inserted based on the behavioral constraints of driving behavior to obtain the insertion position information of the target vehicle in the point cloud frame corresponding to each of the basic scenes includes the following steps: When the driving behavior is overtaking, a vehicle is randomly selected from the main vehicle and the cooperating vehicle as the target vehicle to be inserted. A local coordinate system is established in the point cloud frame corresponding to each of the basic scenes with the target vehicle as the origin. The direction of the overtaking target is the positive direction of the horizontal axis, and the left side of the direction of the forward movement is the positive direction of the vertical axis. The overtaking trajectory is calculated based on the overtaking constraints to obtain four overtaking target points, including the initial position, lane change position, acceleration position, and end position. The overtaking constraints include overtaking position sub-conditions, lane change collision avoidance sub-conditions, and speed limit sub-conditions. Starting from the initial position, according to the order of the overtaking target points, an overtaking target point is determined sequentially for each point cloud frame corresponding to the basic scene. The horizontal and vertical coordinates of the target vehicle to be inserted are adjusted according to the approach target constraint and the overtaking target point to obtain the insertion position information.

6. The method according to claim 2, characterized in that, The process of performing metamorphic testing on the test cases according to a preset metamorphic relationship to obtain test predictions includes the following steps: The true trajectory result of the target vehicle is merged with the original tracking result of the basic scene to obtain the first test result; The test results of the first test and the test tracking results of the test cases are matched based on a preset metamorphic relationship with a soft equality criterion to obtain a test prediction. The metamorphic relationship includes inserting a target vehicle without changing the original tracking result of the cooperative tracking system, and the true trajectory result is accurately tracked in the test tracking result.

7. The method according to claim 1, characterized in that, The step of inputting the test scenarios into the collaborative tracking system for tracking, and calculating the selection metric for each test scenario based on the scenario selection expression, includes the following steps: The test scenario is input into the collaborative tracking system to calculate the crossover ratio (CRO) and average the mean of several tracking objects to be detected, and the average tracking CRO is obtained. The error rate is calculated based on the number of false positives and false negatives of the tracked object; The average tracking crossover ratio and the error rate are weighted and calculated to obtain the selection metric for each test scenario.

8. A transformation test system for an autonomous driving cooperative tracking system, characterized in that, The system includes: The first module is used to obtain an initial dataset, wherein the initial dataset includes several initial scenes, and each initial scene includes point cloud frames and labels of the main vehicle and the cooperative vehicles; The second module is used to extract a preset number of initial scenes from the initial dataset as basic scenes to obtain a basic scene queue; The third module is used to perform mutation operations on the basic scenario queue to obtain test scenarios. The mutation operation is to insert target vehicles with driving behaviors, including following, overtaking, and turning. The fourth module is used to input the test scenarios into the collaborative tracking system for tracking, and to calculate the selection metric for each test scenario based on the scenario selection expression; The fifth module selects a test scenario based on the selection metric to obtain the result scenario; The sixth module updates the basic scenario queue based on the result scenario, outputs the tracking results of the result scenario as a test report, and then returns to the step of performing a mutation operation on the basic scenario queue to obtain the test scenario, until the preset maximum number of iterations is reached.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images