Target identification method and device, and vehicle-road-cloud integrated test system

Abstract

The application relates to a target identification method and device and a car-road-cloud integrated test system, which comprises the following steps: acquiring test scene configuration information uploaded by a test scene simulation module, and issuing the test scene configuration information to a roadside sensing system; acquiring roadside sensing data and a preliminary target identification result uploaded by the roadside sensing system, and acquiring vehicle-end sensing data uploaded by a vehicle-end control system; the preliminary target identification result is determined by the roadside sensing system according to the roadside sensing data and the test scene configuration information, and the preliminary target identification result comprises candidate controlled targets and candidate uncontrolled targets; the preliminary target identification result is subjected to secondary verification according to the roadside sensing data, the vehicle-end sensing data and current running state data of the test scene; and the final target identification result is determined according to the secondary verification result. Therefore, the controlled targets and the uncontrolled targets in the test scene can be accurately distinguished, and the safety of unmanned test is significantly improved.

Description

Technical Field

[0001] This application relates to the field of vehicle testing, and in particular to a target recognition method, device, and vehicle-road-cloud integrated testing system. Background Technology

[0002] With the rapid development of intelligent connected vehicle technology, unmanned testing has become a crucial step in verifying the performance of vehicle driver assistance systems. Vehicle-road-cloud integrated technology, through the collaboration of "smart vehicles, intelligent roads, and powerful cloud," provides a new path for automated testing in closed test tracks. The roadside perception system, as the core perception terminal, is responsible for identifying, locating, and tracking targets within the test track. In closed test tracks, there are two types of core targets: controlled targets (such as the vehicle under test, simulated pedestrians, etc.), whose movement is controlled by test commands; and uncontrolled targets (such as stray personnel, unrelated vehicles, etc.), whose movement is uncontrollable. Failure to accurately distinguish between these two types of targets can easily lead to test interruptions or safety accidents.

[0003] Existing roadside perception technologies primarily focus on improving target detection accuracy, only able to identify target types (such as vehicles and pedestrians), but unable to determine whether a target is a controlled object within the testing scenario. They also fail to leverage the advantages of vehicle-road-cloud collaboration to integrate multi-source perception data with prior scenario knowledge. Therefore, a method capable of accurately distinguishing between controlled and uncontrolled targets is urgently needed to ensure the safety of unmanned testing. Summary of the Invention

[0004] To address or at least partially address the aforementioned technical problems, this application provides a target recognition method, apparatus, and vehicle-road-cloud integrated testing system.

[0005] In a first aspect, this application provides a target recognition method, including: Obtain the test scenario configuration information uploaded by the test scenario simulation module, and send the test scenario configuration information to the roadside perception system; The system acquires roadside perception data and preliminary target recognition results uploaded by the roadside perception system, as well as vehicle-side perception data uploaded by the vehicle-side control system under test. The preliminary target recognition results are determined by the roadside perception system based on the roadside perception data and the test scenario configuration information. The preliminary target recognition results include candidate controlled targets and candidate uncontrolled targets. The preliminary target recognition result is then verified a second time based on the roadside perception data, the vehicle-side perception data, and the current operating status data of the test scenario. Based on the results of the secondary verification, the final target recognition result is determined.

[0006] In one possible implementation, the roadside perception system determines the preliminary target recognition result based on the roadside perception data and the test scenario configuration information, including: The roadside sensing system extracts the features of the target from the roadside sensing data; The features are matched with the controlled target features contained in the test scenario configuration information; Based on the matching results, the target is determined to be either a candidate controlled target or a candidate uncontrolled target.

[0007] In one possible implementation, the secondary verification of the preliminary target recognition result based on the roadside perception data, the vehicle-side perception data, and the current operating status data of the test scenario includes at least one of the following: Based on the roadside perception data and the vehicle-side perception data, verify whether the candidate controlled target exists in the list of controlled targets included in the test scenario configuration information; Based on the roadside perception data, the vehicle-side perception data, and the current operating status data of the test scenario, verify whether the motion state of the candidate controlled target is consistent with the control commands contained in the current operating status data of the test scenario. Based on the roadside perception data and the vehicle-side perception data, verify whether the candidate uncontrolled target is located outside the planned range of the test scenario. Based on the roadside perception data and the vehicle-side perception data, verify whether the motion trajectory of the candidate uncontrolled target is abnormal; For fuzzy targets, multi-source cross-validation is performed based on the roadside perception data from at least one roadside node and the vehicle-side perception data to determine whether the fuzzy target is a controlled target or an uncontrolled target; wherein, the fuzzy target is a candidate controlled target or a candidate uncontrolled target whose absolute value of the difference between the corresponding feature matching degree and a preset threshold is less than a preset deviation value.

[0008] In one possible implementation, verifying whether the candidate controlled target exists in the controlled target list included in the test scenario configuration information based on the roadside perception data and the vehicle-side perception data includes: The roadside perception data and the vehicle-side perception data are fused to obtain the fusion features of the candidate controlled target; The fused features are matched with the controlled target features contained in the test scenario configuration information; If a controlled target feature that matches the fusion feature exists, the candidate controlled target is determined as a controlled target.

[0009] In one possible implementation, verifying whether the trajectory of the candidate uncontrolled target is abnormal based on the roadside perception data and the vehicle-mounted perception data includes: Based on the roadside perception data and the vehicle-side perception data, obtain the position information of the candidate uncontrolled target in multiple consecutive frames, and generate the motion trajectory of the candidate uncontrolled target; Determine whether the motion trajectory exhibits at least one abnormal characteristic among sudden intrusion, irregular reversal, sudden speed change, or sudden direction change; If so, it is determined that the movement trajectory is abnormal.

[0010] In one possible implementation, the multi-source cross-validation includes: The first feature of the fuzzy target is extracted from the roadside perception data of the at least one roadside node, and the second feature of the fuzzy target is extracted from the vehicle-side perception data; Determine a cross-source feature consistency measure between the first feature and the second feature; Based on the cross-source feature consistency metric, the fuzzy target is determined to be either a controlled target or an uncontrolled target.

[0011] In one possible implementation, the at least one roadside node includes multiple roadside nodes; the multi-source cross-validation includes: Multiple first features of the fuzzy target are extracted from the roadside perception data of multiple roadside nodes, and second features of the fuzzy target are extracted from the vehicle-side perception data. Determine a cross-perspective consistency measure among the plurality of first features, and determine a cross-source consistency measure between the plurality of first features and the second feature; Based on the cross-perspective consistency metric and the cross-source consistency metric, the fuzzy target is determined to be either a controlled target or an uncontrolled target.

[0012] In one possible implementation, it also includes: Perform test management operations corresponding to the final target identification result.

[0013] Secondly, this application provides a target recognition device, comprising: The scenario configuration information synchronization module is used to obtain the test scenario configuration information uploaded by the test scenario simulation module and send the test scenario configuration information to the roadside perception system. The roadside data acquisition module is used to acquire roadside sensing data and preliminary target recognition results uploaded by the roadside sensing system; wherein, the preliminary target recognition results are determined by the roadside sensing system based on the roadside sensing data and the test scenario configuration information, and the preliminary target recognition results include: candidate controlled targets and candidate uncontrolled targets; The vehicle-side data acquisition module is used to acquire vehicle-side perception data uploaded by the vehicle's control system under test. The secondary verification module is used to perform secondary verification on the preliminary target recognition result based on the roadside perception data, the vehicle-side perception data, and the current operating status data of the test scenario. The result determination module is used to determine the final target recognition result based on the secondary verification result.

[0014] In one possible implementation, the roadside perception system determines the preliminary target recognition result based on the roadside perception data and the test scenario configuration information, including: The roadside sensing system extracts the features of the target from the roadside sensing data; The features are matched with the controlled target features contained in the test scenario configuration information; Based on the matching results, the target is determined to be either a candidate controlled target or a candidate uncontrolled target.

[0015] In one possible implementation, the secondary verification module performs a secondary verification on the preliminary target recognition result based on the roadside perception data, the vehicle-side perception data, and the current operating status data of the test scenario, including at least one of the following: Based on the roadside perception data and the vehicle-side perception data, verify whether the candidate controlled target exists in the list of controlled targets included in the test scenario configuration information; Based on the roadside perception data, the vehicle-side perception data, and the current operating status data of the test scenario, verify whether the motion state of the candidate controlled target is consistent with the control commands contained in the current operating status data of the test scenario. Based on the roadside perception data and the vehicle-side perception data, verify whether the candidate uncontrolled target is located outside the planned range of the test scenario. Based on the roadside perception data and the vehicle-side perception data, verify whether the motion trajectory of the candidate uncontrolled target is abnormal; For fuzzy targets, multi-source cross-validation is performed based on the roadside perception data from at least one roadside node and the vehicle-side perception data to determine whether the fuzzy target is a controlled target or an uncontrolled target; wherein, the fuzzy target is a candidate controlled target or a candidate uncontrolled target whose absolute value of the difference between the corresponding feature matching degree and a preset threshold is less than a preset deviation value.

[0016] In one possible implementation, the secondary verification module verifies whether the candidate controlled target exists in the controlled target list included in the test scenario configuration information based on the roadside perception data and the vehicle-side perception data, including: The roadside perception data and the vehicle-side perception data are fused to obtain the fusion features of the candidate controlled target; The fused features are matched with the controlled target features contained in the test scenario configuration information; If a controlled target feature that matches the fusion feature exists, the candidate controlled target is determined as a controlled target.

[0017] In one possible implementation, the secondary verification module verifies whether the motion trajectory of the candidate uncontrolled target is abnormal based on the roadside perception data and the vehicle-side perception data, including: Based on the roadside perception data and the vehicle-side perception data, obtain the position information of the candidate uncontrolled target in multiple consecutive frames, and generate the motion trajectory of the candidate uncontrolled target; Determine whether the motion trajectory exhibits at least one abnormal characteristic among sudden intrusion, irregular reversal, sudden speed change, or sudden direction change; If so, it is determined that the movement trajectory is abnormal.

[0018] In one possible implementation, the multi-source cross-validation includes: The first feature of the fuzzy target is extracted from the roadside perception data of the at least one roadside node, and the second feature of the fuzzy target is extracted from the vehicle-side perception data; Determine a cross-source feature consistency measure between the first feature and the second feature; Based on the cross-source feature consistency metric, the fuzzy target is determined to be either a controlled target or an uncontrolled target.

[0019] In one possible implementation, the at least one roadside node includes multiple roadside nodes; the multi-source cross-validation includes: Multiple first features of the fuzzy target are extracted from the roadside perception data of multiple roadside nodes, and second features of the fuzzy target are extracted from the vehicle-side perception data. Determine a cross-perspective consistency measure among the plurality of first features, and determine a cross-source consistency measure between the plurality of first features and the second feature; Based on the cross-perspective consistency metric and the cross-source consistency metric, the fuzzy target is determined to be either a controlled target or an uncontrolled target.

[0020] In one possible implementation, the device further includes: The test management module is used to perform test management operations corresponding to the final target identification result.

[0021] Thirdly, this application provides a vehicle-road-cloud integrated testing system, including: The test scenario simulation module is used to generate test scenario configuration information; A roadside perception system is used to collect roadside perception data and determine preliminary target recognition results based on the roadside perception data and the test scenario configuration information; the preliminary target recognition results include candidate controlled targets and candidate uncontrolled targets. The vehicle-side control system under test is used to collect vehicle-side sensing data; The cloud platform is communicatively connected to the test scenario simulation module, the roadside perception system, and the vehicle-to-the-test control system, respectively. The cloud platform is configured as follows: Obtain the test scenario configuration information uploaded by the test scenario simulation module, and send the test scenario configuration information to the roadside perception system; Acquire the roadside perception data and the preliminary target recognition result uploaded by the roadside perception system, and acquire the vehicle-side perception data uploaded by the vehicle-side control system under test; The preliminary target recognition result is then verified a second time based on the roadside perception data, the vehicle-side perception data, and the current operating status data of the test scenario. Based on the results of the secondary verification, the final target recognition result is determined.

[0022] Fourthly, this application provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the target recognition method described in any of the first aspects.

[0023] Fifthly, this application provides a computer-readable storage medium storing a program for a target recognition method, wherein when the program for the target recognition method is executed by a processor, it implements the steps of the target recognition method described in any of the first aspects.

[0024] The technical solution provided in this application pre-acquires test scenario configuration information containing controlled target information and sends it to the roadside perception system. This enables the roadside perception system to perform preliminary identification of targets within the test site based on prior knowledge, generating preliminary results of candidate controlled and candidate uncontrolled targets. Furthermore, it integrates roadside perception data, vehicle-side perception data, and the current operating status data of the test scenario to perform a secondary verification of the preliminary identification results. Because the secondary verification incorporates vehicle-side perception data (providing a close-up view) and current operating status data (providing real-time control commands and scene progress), it effectively cross-verifies the accuracy of the preliminary roadside identification, correcting identification deviations caused by single-view occlusion, lighting changes, or insufficient static prior information. Therefore, this solution can accurately distinguish between controlled and uncontrolled targets in the test scenario, avoiding test interruptions or safety accidents caused by misclassifying uncontrolled targets as controlled targets, and also avoiding incomplete test scenarios caused by misclassifying controlled targets as uncontrolled targets, thereby significantly improving the safety of unmanned testing. Attached Figure Description

[0025] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0026] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 A schematic diagram of a vehicle-road-cloud integrated testing system provided in an embodiment of this application; Figure 2 A flowchart illustrating a target recognition method provided in an embodiment of this application; Figure 3 A flowchart illustrating another target recognition method provided in this application embodiment; Figure 4 A flowchart illustrating yet another target recognition method provided in this application embodiment; Figure 5 A flowchart illustrating another target recognition method provided in this application embodiment; Figure 6 A block diagram illustrating an embodiment of a target recognition device provided in this application; Figure 7 This is a structural diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0029] See Figure 1 This is a schematic diagram of a vehicle-road-cloud integrated testing system provided in an embodiment of this application. Figure 1 As shown, the vehicle-road-cloud integrated testing system includes: a test scenario simulation module 11, a roadside perception system 12, a vehicle-to-the-world control system 13, and a cloud platform 14. Additionally, it may include an interactive terminal, such as a tablet, smartphone, or dedicated handheld terminal. This interactive terminal can be used to display vehicle status, perception data, and warning information issued by the cloud platform in real time, and supports manual intervention by test personnel in emergency situations (such as sending deceleration or stopping commands). Simultaneously, the interactive terminal can upload its own status (such as location and battery level) and the test personnel's operation records to the cloud platform, forming a closed-loop test management system with human-machine collaboration.

[0030] The cloud platform 14 includes a cloud control infrastructure platform and a cloud control application platform. For example, the cloud control infrastructure platform provides underlying computing, storage, and network communication infrastructure services, supporting the real-time processing and distribution of massive amounts of data. The cloud control application platform, built on top of the cloud control infrastructure platform, carries specific testing business logic, such as test scenario management, data fusion and verification, target identification and decision-making, and test report generation. Through the collaboration of the cloud control infrastructure platform and the cloud control application platform, the cloud platform 14 can achieve real-time aggregation, fusion processing, instruction issuance, and test process management of test data, providing core computing and decision-making capabilities for integrated vehicle-road-cloud unmanned testing.

[0031] Test scenario simulation module 11 is used to generate test scenario configuration information; specifically: The test scenario simulation module 11 refers to software tools used to generate or edit unmanned test scenarios, such as simulation platforms like SUMO and CARLA. This module can pre-set various parameters of the test scenario according to test requirements, i.e., the test scenario configuration information.

[0032] Test scenario configuration information is a priori data set, which includes at least a list of controlled targets (including the type, initial position, motion trajectory parameters, and control command protocol of the controlled targets), and may also include scenario description information such as road topology, traffic rules, and environmental conditions. It should be noted that test scenario configuration information is usually static, that is, once the test scenario is determined, this information remains unchanged throughout the entire test process, which is different from the "current running status data of the test scenario" that will be updated in real time.

[0033] For example, testers can use the test scenario simulation module 11 to generate unmanned test scenarios (such as meeting oncoming traffic at intersections, pedestrians crossing the road, etc.) that involve multi-target interactions. They can then determine the list of controlled targets within the test scenario, which includes 3 vehicles with driver assistance capabilities, 2 simulated vehicles, and 3 simulated pedestrians. The testers can also define the type, initial position, trajectory parameters (such as the simulated pedestrian crossing the road at a speed of 1.2 m / s and a straight line) and V2X (Vehicle-to-Everything) control command protocol for each type of controlled target.

[0034] The roadside perception system 12 is used to collect roadside perception data and determine preliminary target recognition results based on the roadside perception data and the aforementioned test scenario configuration information; the preliminary target recognition results include candidate controlled targets and candidate uncontrolled targets; specifically: The roadside perception system 12 is a perception terminal deployed at key nodes (such as intersections and curves) of a closed test site. It typically consists of high-definition cameras and mobile edge computing (MEC) units, used to collect target image data within the test site and perform preliminary processing. In practical applications, the roadside perception system 12 includes multiple roadside nodes, each containing at least one high-definition camera and optional other sensors (such as millimeter-wave radar, lidar, etc.). These roadside nodes are distributed at key locations such as intersections, curves, and straight sections of the test site to achieve comprehensive coverage. Different roadside nodes have different perception ranges, and the perception ranges of adjacent nodes can overlap, thereby eliminating blind spots and ensuring continuous tracking and omnidirectional perception of targets within the test site.

[0035] Roadside perception data refers to the information of targets within the test site collected in real time by the roadside perception system 12 through sensors such as high-definition cameras. After preprocessing by the edge computing unit (such as Gaussian filtering for noise reduction, perspective transformation for distortion correction, image enhancement, and spatiotemporal synchronization), the target's position coordinates, speed, direction of movement, shape features, contour information, and attitude information can be extracted. In addition, roadside perception data also includes structured data (such as target lists, tracking trajectories, and feature vectors) after preprocessing by the edge computing unit. After time synchronization and spatial alignment, this data can be used for subsequent target recognition, feature extraction, and secondary verification.

[0036] The vehicle-side control system 13 under test is used to collect vehicle-side sensing data; specifically: The vehicle-to-the-world (V2D) control system 13 is an onboard control system installed on the vehicle under test, which possesses driver assistance capabilities. It typically consists of onboard sensors (such as cameras, millimeter-wave radar, and lidar), positioning equipment (such as GPS / inertial navigation systems), a V2X communication unit, and a decision control unit. It is used to collect the vehicle's own status data and surrounding environmental information, and to execute test commands issued by the cloud platform. In practical applications, the V2D control system 13 may include one or more onboard nodes (such as forward-facing cameras, surround-view cameras, and millimeter-wave radar). These sensors are distributed at the front, rear, left, and right of the vehicle to achieve omnidirectional perception of the vehicle's surrounding environment. Different onboard sensors have different detection ranges, and the detection ranges of adjacent sensors can overlap, thereby reducing blind spots and ensuring that the vehicle can obtain continuous and reliable environmental information during testing.

[0037] Vehicle-side perception data refers to the vehicle's own state data and environmental perception data collected in real time by the vehicle-side control system 13 through onboard sensors and positioning equipment. This includes, but is not limited to: the vehicle's position coordinates, driving speed, heading angle, steering angle, acceleration, brake / throttle opening, control command execution status (i.e., the degree of conformity between the vehicle's actual execution and the commands issued by the cloud platform), and image data collected by the vehicle-side cameras (such as visual information of the road ahead and surrounding objects). After time synchronization and spatial coordinate system transformation, this data can be uploaded to the cloud platform for cross-validation with roadside perception data, motion state consistency verification, and secondary fusion recognition.

[0038] The cloud platform 14 is communicatively connected to the test scenario simulation module 11, the roadside perception system 12, and the vehicle-to-the-test control system 13. In the technical solution provided in this application embodiment, the cloud platform 14 is configured as follows: Obtain the test scenario configuration information uploaded by the test scenario simulation module 11, and send the test scenario configuration information to the roadside perception system 12; Acquire roadside perception data and preliminary target recognition results uploaded by the roadside perception system 12, and acquire vehicle-side perception data uploaded by the vehicle-side control system 13 under test; Based on roadside perception data, vehicle-side perception data, and the current operating status data of the test scenario, the preliminary target recognition results are verified a second time. Based on the results of the secondary verification, the final target recognition result is determined.

[0039] The overall architecture of the vehicle-road-cloud integrated testing system provided in the embodiments of this application has been explained above. The following, in conjunction with specific embodiments, provides a detailed explanation of each step in the target recognition method provided in the embodiments of this application.

[0040] Figure 2 This is a flowchart illustrating an embodiment of a target recognition method provided in this application. In one embodiment, the method is executed by a cloud platform, for example... Figure 1 The cloud platform 14 is shown in the example. (For example...) Figure 2 As shown, the method includes the following steps: Step 201: Obtain the test scenario configuration information uploaded by the test scenario simulation module, and send the test scenario configuration information to the roadside perception system.

[0041] In one embodiment, the test scenario simulation module (e.g.) Figure 1 The test scenario simulation module 11 shown in the example sends data to the cloud platform (e.g., Figure 1 In the example shown, cloud platform 14) uploads test scenario configuration information, so that in step 201, cloud platform can obtain test scenario configuration information uploaded by test scenario simulation module.

[0042] For example, the test scenario simulation module proactively uploads test scenario configuration information to the cloud platform. For instance, during test scenario initialization, the test scenario simulation module automatically packages the test scenario configuration information and sends it to the cloud platform.

[0043] For example, the cloud platform proactively sends a configuration retrieval request to the test scenario simulation module. For instance, when starting the test process, the cloud platform sends a request to the test scenario simulation module, which responds by uploading the configuration information of the current scenario to the cloud platform.

[0044] In one embodiment, after obtaining complete test scenario configuration information from the test scenario simulation module, the cloud platform first verifies the test scenario configuration information. For example, it performs completeness and validity checks. Completeness check involves verifying whether the configuration information contains all necessary fields (such as the list of controlled targets, initial position, trajectory parameters, control command protocol, etc.) and whether there are any missing data or empty fields. Validity check involves verifying whether the data format conforms to preset specifications (such as whether the coordinates are within the site range, whether the speed exceeds the limit, whether the command protocol is the standard V2X format, etc.) and whether the parameters are within reasonable thresholds. Through these verifications, the accuracy and reliability of the configuration information can be ensured, avoiding test anomalies or system failures due to data errors.

[0045] Furthermore, after successful verification, the cloud platform will distribute the test scenario configuration information to the roadside perception system (e.g., Figure 1 The roadside perception system 12 is shown in the example. The purpose of this action is to allow the roadside perception system to know in advance which targets in the test scene are controlled, their appearance characteristics (such as size, outline, color, and texture), and their motion patterns, thereby providing a feature comparison benchmark for subsequent preliminary target recognition. This step ensures that the roadside perception system can pre-build a controlled target image feature comparison library.

[0046] The controlled target image feature comparison library is a database stored in the edge computing unit of the roadside perception system. It contains preset image features of each controlled target extracted from the test scenario configuration information (such as static features like size, outline, color, and texture, as well as dynamic features like motion trajectory and speed change patterns). In actual testing, after the roadside perception system acquires target images, it extracts the real-time image features of the targets and matches them with the features in this comparison library to preliminarily determine whether the target belongs to the controlled target category.

[0047] Furthermore, in one embodiment, the cloud platform can also distribute test scenario configuration information to the control system of the vehicle under test (e.g., Figure 1 The vehicle-to-vehicle control system 13 shown in the example enables the vehicle to understand basic information about the test scenario, such as which targets are controlled and the motion rules to be followed, which helps the vehicle to maintain coordination with the test scenario when executing driving commands.

[0048] Step 202: Obtain the roadside perception data and preliminary target recognition results uploaded by the roadside perception system, and obtain the vehicle-side perception data uploaded by the vehicle-side control system under test; wherein, the preliminary target recognition results are determined by the roadside perception system based on the roadside perception data and test scenario configuration information, and the preliminary target recognition results include: candidate controlled targets and candidate uncontrolled targets.

[0049] The preliminary target identification result is an intermediate result obtained by the roadside perception system after comparing roadside perception data and test scenario configuration information. Specifically, it includes two categories of labels: "candidate controlled targets" and "candidate uncontrolled targets." Candidate controlled targets refer to targets that the roadside perception system initially determines to be controlled objects within the test scenario plan; candidate uncontrolled targets refer to irrelevant targets that the roadside perception system initially determines to be outside the test scenario plan and possibly unintentionally intruded.

[0050] In one embodiment, the roadside perception system uploads roadside perception data and preliminary target recognition results to the cloud platform every set time interval (e.g., 100ms). Here, there is a correlation between the roadside perception data and the preliminary target recognition results; specifically, the preliminary target recognition results are generated based on the roadside perception data. Therefore, in step 202, the cloud platform can obtain the roadside perception data uploaded by the roadside perception system and its corresponding preliminary target recognition results.

[0051] At the same time, the cloud platform also receives vehicle-side perception data (including vehicle motion status and vehicle-side images) uploaded by the vehicle's control system under test, thus enabling the cloud platform to aggregate multi-source data from the roadside and the vehicle, providing sufficient raw information for subsequent secondary verification.

[0052] Therefore, the key to step 202 is that the cloud platform does not directly rely on perception data from a single source, but rather aggregates data from the roadside and the vehicle to provide a data foundation for subsequent fusion verification.

[0053] Step 203: Based on the roadside perception data, vehicle-side perception data, and the current operating status data of the test scenario, perform a second verification on the preliminary target recognition results.

[0054] The current operational status data of the test scenario is dynamic information recorded in real time by the cloud platform during test execution, distinct from static test scenario configuration information. This data includes at least: the execution progress of the test scenario (current stage), the scheduling status of the controlled targets (which instructions have been issued and which have not yet been executed), and the theoretical control instruction values ​​issued by the cloud platform to each controlled target at the current moment (e.g., the speed a simulated vehicle should maintain and the trajectory it should follow). This data reflects the real-time evolution of the test scenario and serves as a crucial benchmark for determining whether the target's movement aligns with the test plan.

[0055] Secondary verification is a further verification and correction based on the initial roadside identification results, using multi-source data (roadside perception data, vehicle-side perception data, and current scene operation status data). It aims to eliminate misjudgments caused by a single perspective or a single sensor and improve the reliability of identification.

[0056] As for how the preliminary target recognition results are secondarily verified based on roadside perception data, vehicle-side perception data, and the current operational status data of the test scenario, this will be explained in the following text. Figure 4 The embodiments shown are provided as examples and will not be elaborated upon here.

[0057] Step 204: Determine the final target recognition result based on the secondary verification results.

[0058] The final target identification result is the cloud platform's final judgment on each candidate target based on the secondary verification results, confirming whether the target is a "controlled target" or an "uncontrolled target". The final target identification result can serve as the basis for subsequent process management of the entire testing system.

[0059] In one embodiment, the cloud platform summarizes the secondary verification results in step 203 and makes a final controlled / uncontrolled determination for each target. For example, if the secondary verification confirms that a candidate controlled target meets all the controlled target verification conditions, it is output as a "controlled target"; if the secondary verification confirms that a candidate uncontrolled target is outside the planning range or has an abnormal trajectory, it is output as an "uncontrolled target".

[0060] The technical solution provided in this application pre-acquires test scenario configuration information containing controlled target information and sends it to the roadside perception system. This enables the roadside perception system to perform preliminary identification of targets within the test site based on prior knowledge, generating preliminary results of candidate controlled and candidate uncontrolled targets. Furthermore, it integrates roadside perception data, vehicle-side perception data, and the current operating status data of the test scenario to perform a secondary verification of the preliminary identification results. Because the secondary verification incorporates vehicle-side perception data (providing a close-up view) and current operating status data (providing real-time control commands and scene progress), it effectively verifies the accuracy of the preliminary roadside identification and corrects identification deviations caused by single-view occlusion, lighting changes, or insufficient static prior information. Therefore, this solution can accurately distinguish between controlled and uncontrolled targets in the test scenario, avoiding test interruptions or safety accidents caused by misclassifying uncontrolled targets as controlled targets, and also avoiding incomplete test scenarios caused by misclassifying controlled targets as uncontrolled targets, thereby significantly improving the safety of unmanned testing.

[0061] Figure 3 A flowchart illustrating an embodiment of another target recognition method provided in this application. Figure 3 The process shown is in Figure 2 Based on the illustrated process, an exemplary implementation of the roadside perception system determining preliminary target recognition results based on roadside perception data and test scenario configuration information is described. For example... Figure 3 As shown, it includes the following steps: Step 301: The roadside perception system extracts the features of the target from the roadside perception data.

[0062] In this context, target characteristics refer to quantitative information that describes the target's appearance and motion attributes. Appearance attributes include static features, such as the target's dimensions, outline shape, color distribution, and surface texture; motion attributes include dynamic features, such as the target's trajectory, velocity variation patterns, acceleration parameters, and attitude changes. These characteristics collectively constitute the basis for distinguishing different targets and determining whether a target belongs to a controlled object.

[0063] Step 301 is executed by the roadside perception system (specifically its edge computing unit). The roadside perception system first acquires the collected roadside perception data (such as image data captured by high-definition cameras), and then automatically extracts the static and dynamic features of each detected target from the original image using a preset image recognition algorithm (such as a deep learning model) to form a target feature vector (containing static and dynamic feature vectors).

[0064] Step 302: Match the above features with the controlled target features contained in the test scenario configuration information.

[0065] The controlled target features included in the test scenario configuration information refer to the expected features of each controlled target determined or derived by the roadside perception system based on the test scenario configuration information (especially the list of controlled targets, including the type, initial position, and motion trajectory parameters of the controlled targets). For example, the roadside perception system can retrieve corresponding static features such as shape, size, outline, color, and texture from a pre-stored appearance model library based on the type of the controlled target (such as a vehicle or pedestrian). Simultaneously, it can determine the expected motion pattern features based on its motion trajectory parameters (such as speed, acceleration, and direction of motion). These expected features serve as standard templates for matching against the target features extracted in real time.

[0066] Step 302 is still performed by the roadside perception system. The roadside perception system matches the target feature vector extracted in step 301 with the controlled target features in the test scenario configuration information one by one. The purpose of the matching is to calculate the similarity between the target actually perceived by the roadside perception system and each preset controlled target feature. This process can use similarity measurement algorithms (such as cosine similarity, Euclidean distance, etc.) to output a quantified matching score, reflecting the closeness between the target actually perceived by the roadside perception system and the preset controlled target.

[0067] Step 303: Based on the matching results, determine whether the target is a candidate controlled target or a candidate uncontrolled target.

[0068] Candidate controlled targets are those whose characteristics, based on preliminary assessment, are relatively close to those of controlled targets in the test scenario configuration information and are likely to belong to the controlled objects within the test plan. Candidate uncontrolled targets are those whose characteristics, based on preliminary assessment, are not close to any of the controlled target characteristics and are more likely to be unrelated targets that have accidentally entered the system (such as stray personnel, wild animals, unrelated vehicles, etc.). Both types of labels are preliminary identification results, and final confirmation requires secondary verification by the cloud platform.

[0069] Step 303 is still performed by the roadside perception system. Based on the matching results obtained in step 302 (i.e., the matching degree value of each target), the roadside perception system assigns a preliminary label to the target according to preset judgment rules (e.g., if the matching degree is higher than a certain threshold, it is judged as a candidate controlled target; if it is lower than the threshold, it is judged as a candidate uncontrolled target). This judgment result assigns a temporary identity of "candidate controlled" or "candidate uncontrolled" to each target.

[0070] Through steps 301 to 303 above, the roadside perception system has completed the preliminary classification of all targets in the test site, laying the foundation for efficient fusion verification of the cloud platform.

[0071] Figure 4 A flowchart illustrating another embodiment of the target recognition method provided in this application. Figure 4 The process shown is in Figure 2 Based on the illustrated process, an exemplary implementation is described that performs secondary verification on the preliminary target recognition results using roadside perception data, vehicle-side perception data, and the current operational status data of the test scenario. For example... Figure 4 As shown, it includes the following steps: Step 401: Obtain the test scenario configuration information uploaded by the test scenario simulation module, and send the test scenario configuration information to the roadside perception system.

[0072] Step 402: Obtain the roadside perception data and preliminary target recognition results uploaded by the roadside perception system, and obtain the vehicle-side perception data uploaded by the vehicle-side control system under test; for candidate controlled targets, execute step 403; for candidate uncontrolled targets, execute step 405; for fuzzy targets, execute step 407.

[0073] The fuzzy target is a candidate controlled target or a candidate uncontrolled target whose absolute value of the difference between the corresponding feature matching degree and the preset threshold is less than the preset deviation value.

[0074] In the aforementioned embodiments, when making an initial judgment, the roadside perception system divides targets into candidate controlled targets and candidate uncontrolled targets based on a preset threshold. However, for targets whose absolute value of the difference between their feature matching degree and the preset threshold is less than a preset deviation value, simply classifying them based on the preset threshold may result in errors. Therefore, these targets are defined as fuzzy targets and are specifically verified during the secondary verification stage.

[0075] Step 403: Based on the roadside perception data and vehicle-side perception data, verify whether the candidate controlled target exists in the list of controlled targets included in the test scenario configuration information; if yes, continue to step 404; if no, determine the candidate controlled target as an uncontrolled target and jump to step 408.

[0076] In one embodiment, verifying whether a candidate controlled target exists in the list of controlled targets included in the test scenario configuration information based on roadside perception data and vehicle-side perception data includes: fusing the roadside perception data and vehicle-side perception data to obtain the fusion features of the candidate controlled target; matching the fusion features with the controlled target features included in the test scenario configuration information; and determining the candidate controlled target as a controlled target if a controlled target feature that matches the fusion features exists.

[0077] Specifically, the cloud platform acquires roadside perception data uploaded by the roadside perception system and vehicle-side perception data uploaded by the vehicle's control system. After spatiotemporally aligning the two, it uses feature-level fusion methods (such as feature stitching or weighted fusion) or decision-level fusion methods (decision-level fusion methods refer to independently performing target identification or feature matching based on the roadside perception data and vehicle-side perception data, each outputting a preliminary judgment result, and then making a comprehensive decision based on the two judgment results. Common integration methods include: weighted voting (such as roadside weight 0.6 and vehicle-side weight 0.4), logical rules (if the two are consistent, adopt the result; if they are inconsistent, trust the one with higher confidence), or probability fusion based on Bayesian estimation) to generate fused features for the target. These fused features are more comprehensive and robust than features from a single source (such as the features extracted in step 301 above).

[0078] Then, the cloud platform compares the obtained fused features with the controlled target features stored in the test scenario configuration information one by one, and calculates the similarity. Cosine similarity, Euclidean distance, and other measurement methods can be used for matching, outputting the matching score between the fused features and each controlled target feature. If the matching score between a controlled target feature and the fused feature is higher than a preset matching threshold, it indicates that the candidate controlled target is highly consistent with a controlled object in the test plan, and step 404 is executed to further verify that it is a true controlled target. If the matching scores of all templates are lower than the threshold, it indicates that the target is not in the controlled target list and will be subsequently determined as an uncontrolled target.

[0079] Step 404: Based on the roadside perception data, vehicle-side perception data, and the current operating status data of the test scenario, verify whether the motion state of the candidate controlled target is consistent with the control commands contained in the current operating status data of the test scenario; if consistent, determine the candidate controlled target as a controlled target and proceed to step 408; if inconsistent, determine the candidate controlled target as an uncontrolled target and proceed to step 408.

[0080] In step 403, candidate controlled targets in the controlled target list are verified through feature fusion and matching. However, since identity matching can only prove that the target conforms to the controlled object in terms of attributes, it cannot confirm whether its actual movement behavior strictly follows the test plan (for example, the target may deviate from the predetermined trajectory due to communication failure, control abnormality or external interference). Therefore, for such targets, it is necessary to continue to perform dynamic consistency verification of its movement state through step 404.

[0081] Specifically, step 404 is used to verify whether the motion state (i.e., actual behavior, such as position, velocity, heading angle, acceleration, etc.) of the candidate controlled target is consistent with the test plan. The cloud platform extracts the actual motion state of the target from roadside perception data and vehicle-side perception data, and at the same time obtains the corresponding theoretical control command value from the current operating state data of the test scenario, and then compares the two.

[0082] If the actual motion state is consistent with the theoretical control command (e.g., the speed deviation is less than the allowable value, and the trajectory matches), then the target is confirmed as a controlled target, and the process jumps to step 408; if they are inconsistent (e.g., the target speed deviates significantly from the command or the trajectory is abnormal), then the target is determined to be an uncontrolled target (possibly due to communication failure or target abnormality), and the process jumps to step 408.

[0083] Step 404 uses dynamic comparison to effectively identify abnormal targets whose identities match but whose behavior is out of control, thus ensuring test security.

[0084] Step 405: Based on the roadside perception data and vehicle-side perception data, verify whether the candidate uncontrolled target is located outside the test scenario planning range; if so, determine the target as an uncontrolled target and jump to step 408; if not, continue to execute step 408.

[0085] In step 402 above, for objects initially identified as candidate uncontrolled targets, it is necessary to further confirm whether they are indeed unrelated targets that have accidentally entered the area. Since the initial identification relying solely on the roadside system may result in misjudgments (for example, some controlled targets may be misjudged as uncontrolled due to changes in appearance or occlusion), their spatial location is first quickly verified in step 405.

[0086] Specifically, step 405 is used to determine whether the candidate uncontrolled target is located outside the test scenario planning range. Here, "test scenario planning range" refers to a pre-defined effective area within a closed test site used to execute the current test task, typically defined by geographic coordinate boundaries or lane lines; "outside the planning range" means the target's location coordinates exceed these boundaries, such as appearing outside the road shoulder, outside the median strip, or in an unopened area. The cloud platform obtains the target's location coordinates based on roadside perception data and vehicle-side perception data and compares them with the pre-defined test scenario planning range boundaries.

[0087] If the target is outside the planned area, it is directly determined to be an uncontrolled target (such as a person or animal that has mistakenly entered the site), and the process jumps to step 408; if the target is within the planned area, the possibility that it is a controlled target cannot be simply ruled out, and step 406 needs to be executed to perform trajectory anomaly verification.

[0088] Step 405 quickly filters obviously irrelevant targets by spatial location, reducing unnecessary subsequent calculations and improving verification efficiency.

[0089] Step 406: Based on the roadside perception data and vehicle-side perception data, verify whether the motion trajectory of the candidate uncontrolled target is abnormal; if abnormal, determine the candidate uncontrolled target as an uncontrolled target and proceed to step 408; if not abnormal, determine the candidate uncontrolled target as a controlled target and proceed to step 408.

[0090] After confirming in step 405 that the candidate uncontrolled target is within the planned range of the test scenario, the possibility that it is actually a controlled target cannot be ruled out (for example, a controlled target may exhibit unexpected behavior due to communication interruption or control anomaly, or may be misjudged as uncontrolled by the roadside system). Therefore, further dynamic analysis of its motion behavior is required in step 406.

[0091] Specifically, step 406 is used to verify whether the motion trajectory of the candidate uncontrolled target is abnormal. The motion trajectory refers to the target's position sequence across multiple consecutive frames, reflecting its movement path and dynamic behavior. Abnormal features include sudden intrusion (sudden acceleration from a standstill or low speed into the test area), irregular turning back and forth (repeatedly changing direction or hesitating), sudden speed changes (far exceeding the normal range), and sudden direction changes (unexpected turns), which are inconsistent with normal traffic behavior. The cloud platform obtains the target's position information across multiple consecutive frames based on roadside perception data and vehicle-side perception data, generates the motion trajectory, and then analyzes whether the aforementioned abnormal features exist in the trajectory.

[0092] If there is a significant anomaly in the trajectory, the target is determined to be an uncontrolled target (e.g., a malicious intrusion or an out-of-control object), and the process jumps to step 408; if the trajectory is normal (e.g., uniform linear motion, in accordance with traffic rules), it is preliminarily identified that an error may have occurred, and the target should actually be a controlled target within the test plan. Therefore, it is corrected to a controlled target, and the process jumps to step 408.

[0093] Step 406 uses behavior analysis to correct misjudgments caused by similar appearance, obstruction, or temporary anomalies in the roadside system, effectively ensuring the integrity of the test scenario and data consistency.

[0094] Step 407: Perform multi-source cross-validation based on roadside perception data and vehicle-side perception data from at least one roadside node to determine whether the fuzzy target is a controlled target or an uncontrolled target; after determination, proceed to step 408.

[0095] In one embodiment, the above-mentioned multi-source cross-validation includes: extracting a first feature of the fuzzy target from roadside perception data of at least one roadside node, and extracting a second feature of the fuzzy target from vehicle-side perception data; determining a cross-source feature consistency measure between the first feature and the second feature; and determining whether the fuzzy target is a controlled target or an uncontrolled target based on the cross-source feature consistency measure.

[0096] The first feature refers to the feature of the fuzzy target extracted from roadside perception data collected from at least one roadside node, such as static features (e.g., shape outline, color distribution, texture details, size ratio, etc.) and dynamic features (e.g., motion trajectory, speed change pattern, acceleration parameters, attitude change, etc.).

[0097] The second feature refers to the feature of the same fuzzy target extracted from the vehicle-side perception data. Its description is similar to that of the first feature, but it originates from the vehicle-side perspective.

[0098] Cross-source feature consistency metrics quantify the similarity between a first feature and a second feature, and can be calculated using algorithms such as cosine similarity, Euclidean distance, and cross-correlation. This metric reflects the degree of consistency between observations of the same target from two different sources: roadside and vehicle-side. Strong consistency indicates a high degree of feature matching between the two perspectives, suggesting a high degree of confidence in the target's identity; weak consistency may indicate occlusion, lighting differences, or identification errors.

[0099] Step 407 is used to identify the ambiguous target. First, the cloud platform extracts the first feature of the ambiguous target from roadside perception data of at least one roadside node, and simultaneously extracts the second feature of the ambiguous target from vehicle-side perception data. Next, the cloud platform calculates a cross-source feature consistency measure (e.g., cosine similarity) between the first and second features. Finally, the target type is determined based on the comparison between the consistency measure value and a preset threshold: if the cross-source feature consistency measure is greater than the preset threshold, it indicates that the roadside and vehicle-side observations of the target are highly consistent, and the target's features are relatively close to those of a controlled target, thus it is determined to be a controlled target; if the consistency measure is less than or equal to the preset threshold, it indicates that there is a significant difference in features between the two perspectives, possibly due to a misidentification by either the roadside or vehicle-side perspective, or the target itself is uncontrolled, thus it is determined to be an uncontrolled target.

[0100] This embodiment combines cross-source features from the roadside and the vehicle, leveraging the advantages of the vehicle's close-up and high-resolution capabilities to compensate for the shortcomings of long-distance recognition on the roadside, effectively improving the reliability of recognizing blurry targets.

[0101] In another embodiment, at least one roadside node includes multiple roadside nodes, and the aforementioned multi-source cross-validation includes: extracting multiple first features of the fuzzy target from the roadside perception data of the multiple roadside nodes, and extracting second features of the fuzzy target from the vehicle-side perception data; determining a cross-view consistency measure among the multiple first features, and determining a cross-source consistency measure between the multiple first features and the second feature; and determining whether the fuzzy target is a controlled target or an uncontrolled target based on the cross-view consistency measure and the cross-source consistency measure.

[0102] The cross-view consistency metric quantifies the similarity between multiple primary features, i.e., the feature consistency of the same target under different roadside camera views. The higher the metric value, the more stable the appearance of the target under different views, and the higher the credibility of the target itself; if the consistency is low, there may be occlusion, changes in lighting, or confusion regarding the target's identity.

[0103] For example, a cross-view consistency measure among multiple first features is determined as follows: First, the similarity between features extracted from each pair of different roadside nodes is calculated (e.g., using cosine similarity or Euclidean distance). Then, an aggregation method is selected based on actual needs: If average similarity is used, the arithmetic mean of the similarities of all feature pairs is taken as the cross-view consistency measure, which reflects the overall feature stability of the target across all roadside views. If minimum similarity is used, the minimum value among the similarities of all feature pairs is taken as the cross-view consistency measure, which is more sensitive to occlusion or recognition deviations at a certain viewpoint and can detect the most unfavorable viewpoint differences. A higher cross-view consistency measure indicates that the appearance features of the target are more consistent across different roadside cameras, and the credibility of the target's identity is higher.

[0104] For example, the cross-source consistency measure between multiple first features and second features is calculated as follows: First, the similarity between each first feature and second feature is calculated separately. Then, an aggregation method is selected according to actual needs: If average similarity is used, the arithmetic mean of the similarities between all first features and second features is calculated, and this value comprehensively evaluates the overall matching degree between all roadside views and vehicle-side views. If maximum similarity is used, the maximum value among all similarities is taken, and this value is used to select the roadside view that is most consistent with the vehicle-side observation, which is suitable for situations where some roadside nodes are occluded while others are normal. If weighted average similarity is used, different weights are assigned according to the confidence of each roadside node (e.g., based on the distance of the target from the camera, image clarity, detection confidence, etc.), and then the weighted average is calculated. The higher the cross-source consistency measure, the more the features of the roadside multi-view observations match the vehicle-side observations, and the greater the probability that the target is a controlled target.

[0105] The cloud platform makes a comprehensive judgment based on cross-view consistency metrics and cross-source consistency metrics. Generally, if the cross-view consistency metric is higher than the first threshold and the cross-source consistency metric is higher than the second threshold, it means that the target's features are stable across all views and consistent with the vehicle's observations, and it can be judged as a controlled target; otherwise, if either metric is lower than the corresponding threshold, it means that there is a view conflict or cross-source inconsistency, and it is judged as an uncontrolled target.

[0106] This embodiment fully leverages the global coverage advantage of the roadside system and the near-field accuracy advantage of the vehicle by integrating multi-view data from multiple roadside nodes and vehicle-side data. It is particularly suitable for complex scenarios where targets are partially obscured, have uneven lighting, or are difficult to identify at long distances, significantly improving the accuracy and robustness of identifying blurry targets.

[0107] Step 408: Determine the final target identification result and perform the corresponding test management operations.

[0108] For a detailed description of step 408, please refer to [link / reference]. Figure 5 The relevant descriptions in the illustrated embodiments will not be elaborated here.

[0109] Figure 4 The process shown involves performing differentiated verification strategies (including identity fusion verification, motion consistency verification, spatial position verification, trajectory anomaly verification, and multi-source cross-verification) on candidate controlled targets, candidate uncontrolled targets, and fuzzy targets respectively. This process performs secondary verification on the preliminary target recognition results generated by the roadside perception system, thereby effectively distinguishing between controlled and uncontrolled targets in the test scenario and improving the accuracy and robustness of the recognition.

[0110] It should be noted that, Figure 4The order of steps in the process shown is merely an illustrative example. In practical applications, other reasonable order of steps or parallel processing methods (such as combining steps 403 and 404, executing steps 405 and 406 in parallel, or skipping certain verification steps under specific conditions) all fall within the protection scope of this application, as long as the core function of multi-source data fusion and secondary verification can be achieved.

[0111] Figure 5 A flowchart illustrating another embodiment of the target recognition method provided in this application. Figure 5 The process shown is in Figure 2 Based on the illustrated process, the following steps are included: Step 501: Obtain the test scenario configuration information uploaded by the test scenario simulation module, and send the test scenario configuration information to the roadside perception system.

[0112] Step 502: Obtain the roadside perception data and preliminary target recognition results uploaded by the roadside perception system, and obtain the vehicle-side perception data uploaded by the vehicle-side control system under test; wherein, the preliminary target recognition results are determined by the roadside perception system based on the roadside perception data and test scenario configuration information, and the preliminary target recognition results include: candidate controlled targets and candidate uncontrolled targets.

[0113] Step 503: Based on the roadside perception data, vehicle-side perception data, and the current operating status data of the test scenario, perform a second verification on the preliminary target recognition results.

[0114] Step 504: Determine the final target recognition result based on the secondary verification result.

[0115] For a detailed description of steps 501 to 504, please refer to the relevant descriptions in the above embodiments, which will not be repeated here.

[0116] Step 505: Perform the test management operation corresponding to the final target identification result.

[0117] In one embodiment, the cloud platform synchronously sends the final target recognition result to the roadside perception system, the vehicle-to-the-test (V2T) control system, and the test scenario simulation module, so that each module can execute the corresponding response strategy. For example, for an object ultimately confirmed as a controlled target: the roadside perception system continuously tracks its motion state and uploads image perception data in real time to ensure the normal execution of the test scenario; the V2T control system executes predetermined test commands such as passing and avoidance based on the state of the controlled target. For an object ultimately confirmed as an uncontrolled target (such as a person who has mistakenly entered the test area): the cloud platform immediately triggers a safety warning, sends deceleration, avoidance, or stopping commands to the V2T control system, instructing the V2T vehicle to perform the corresponding operation, and simultaneously sends a warning message to the test personnel, notifying them to intervene manually; if the uncontrolled target fails to leave the test area in time, the cloud platform further instructs the test scenario simulation module to suspend the current test scenario to avoid test interruption or collision accidents.

[0118] Figure 5 The process shown is in Figure 2 Based on the process shown, by executing test management operations corresponding to the final target identification results, a closed loop from perception to decision-making is completed, ensuring the safety and continuity of unmanned testing.

[0119] Finally, this application also provides the following embodiments: During the test, the cloud platform records in real time the image recognition data of all detected targets, the secondary verification results, and any anomalies that occur during the recognition process (such as matching degree approaching the threshold, inconsistency between multiple viewpoints, and trajectory anomalies), and organizes this data into a recognition sample dataset. For example, in a complete unmanned test, the cloud platform can collect a dataset containing more than 1,000 image samples. Each sample includes the original image, extracted feature vectors, preliminary recognition labels, secondary verification labels, and the final judgment result.

[0120] After the test, the cloud platform optimized and trained the image feature extraction algorithm deployed in the roadside perception system based on the sample dataset. For example, deep learning models such as YOLOv8 were used to fine-tune the feature extraction network to better adapt it to the specific target appearance and motion patterns in the test site. At the same time, the cloud platform dynamically adjusted the preset threshold for feature matching based on the statistical distribution of matching degree and false positive cases in the sample data (for example, adjusting the 85% threshold under normal scenarios to a value more suitable for the current scenario) to achieve a better balance between accuracy and recall.

[0121] After optimization, the cloud platform updates the controlled target image feature comparison library stored in the roadside perception system, incorporating newly added or corrected controlled target feature templates into the comparison library, and distributing optimized algorithm parameters (such as network weights and matching thresholds) to the edge computing unit of the roadside perception system. As a result, subsequent unmanned testing can be conducted based on a more accurate recognition model and feature comparison library, continuously improving the accuracy and efficiency of target recognition, and better adapting to the testing needs of various vehicles with different driving assistance capabilities.

[0122] Figure 6 This is a block diagram illustrating an embodiment of a target recognition device provided in this application. Figure 6 As shown, the device includes: The scenario configuration information synchronization module 61 is used to obtain the test scenario configuration information uploaded by the test scenario simulation module and send the test scenario configuration information to the roadside perception system. The roadside data acquisition module 62 is used to acquire roadside perception data and preliminary target recognition results uploaded by the roadside perception system; wherein, the preliminary target recognition results are determined by the roadside perception system based on the roadside perception data and the test scenario configuration information, and the preliminary target recognition results include: candidate controlled targets and candidate uncontrolled targets; Vehicle-side data acquisition module 63 is used to acquire vehicle-side perception data uploaded by the vehicle-side control system under test. The secondary verification module 64 is used to perform secondary verification on the preliminary target recognition result based on the roadside perception data, the vehicle-side perception data, and the current operating status data of the test scenario. The result determination module 65 is used to determine the final target recognition result based on the secondary verification result.

[0123] In one possible implementation, the roadside perception system determines the preliminary target recognition result based on the roadside perception data and the test scenario configuration information, including: The roadside sensing system extracts the features of the target from the roadside sensing data; The features are matched with the controlled target features contained in the test scenario configuration information; Based on the matching results, the target is determined to be either a candidate controlled target or a candidate uncontrolled target.

[0124] In one possible implementation, the secondary verification module 64 performs a secondary verification on the preliminary target recognition result based on the roadside perception data, the vehicle-side perception data, and the current operating status data of the test scenario, including at least one of the following: Based on the roadside perception data and the vehicle-side perception data, verify whether the candidate controlled target exists in the list of controlled targets included in the test scenario configuration information; Based on the roadside perception data, the vehicle-side perception data, and the current operating status data of the test scenario, verify whether the motion state of the candidate controlled target is consistent with the control commands contained in the current operating status data of the test scenario. Based on the roadside perception data and the vehicle-side perception data, verify whether the candidate uncontrolled target is located outside the planned range of the test scenario. Based on the roadside perception data and the vehicle-side perception data, verify whether the motion trajectory of the candidate uncontrolled target is abnormal; For fuzzy targets, multi-source cross-validation is performed based on the roadside perception data from at least one roadside node and the vehicle-side perception data to determine whether the fuzzy target is a controlled target or an uncontrolled target; wherein, the fuzzy target is a candidate controlled target or a candidate uncontrolled target whose absolute value of the difference between the corresponding feature matching degree and a preset threshold is less than a preset deviation value.

[0125] In one possible implementation, the secondary verification module 64 verifies whether the candidate controlled target exists in the controlled target list included in the test scenario configuration information based on the roadside perception data and the vehicle-side perception data, including: The roadside perception data and the vehicle-side perception data are fused to obtain the fusion features of the candidate controlled target; The fused features are matched with the controlled target features contained in the test scenario configuration information; If a controlled target feature that matches the fusion feature exists, the candidate controlled target is determined as a controlled target.

[0126] In one possible implementation, the secondary verification module 64 verifies whether the motion trajectory of the candidate uncontrolled target is abnormal based on the roadside perception data and the vehicle-side perception data, including: Based on the roadside perception data and the vehicle-side perception data, obtain the position information of the candidate uncontrolled target in multiple consecutive frames, and generate the motion trajectory of the candidate uncontrolled target; Determine whether the motion trajectory exhibits at least one abnormal characteristic among sudden intrusion, irregular reversal, sudden speed change, or sudden direction change; If so, it is determined that the movement trajectory is abnormal.

[0127] In one possible implementation, the multi-source cross-validation includes: The first feature of the fuzzy target is extracted from the roadside perception data of the at least one roadside node, and the second feature of the fuzzy target is extracted from the vehicle-side perception data; Determine a cross-source feature consistency measure between the first feature and the second feature; Based on the cross-source feature consistency metric, the fuzzy target is determined to be either a controlled target or an uncontrolled target.

[0128] In one possible implementation, the at least one roadside node includes multiple roadside nodes; the multi-source cross-validation includes: Multiple first features of the fuzzy target are extracted from the roadside perception data of multiple roadside nodes, and second features of the fuzzy target are extracted from the vehicle-side perception data. Determine a cross-perspective consistency measure among the plurality of first features, and determine a cross-source consistency measure between the plurality of first features and the second feature; Based on the cross-perspective consistency metric and the cross-source consistency metric, the fuzzy target is determined to be either a controlled target or an uncontrolled target.

[0129] In one possible implementation, the device further includes: The test management module is used to perform test management operations corresponding to the final target identification result.

[0130] In another embodiment of this application, an electronic device is also provided, such as... Figure 7 As shown, it includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor, when executing a program stored in memory, implements the target recognition method described in any of the foregoing method embodiments.

[0131] The electronic device provided in this application embodiment enables the processor to accurately distinguish between controlled and uncontrolled targets in the test scenario by executing the program stored in the memory. This avoids test interruptions or safety accidents caused by misjudging uncontrolled targets as controlled targets, and also avoids incomplete test scenarios caused by misjudging controlled targets as uncontrolled targets, thereby significantly improving the safety, continuity and data consistency of unmanned testing.

[0132] The communication bus 1140 mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus 1140 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0133] The communication interface 1120 is used for communication between the above-mentioned electronic device and other devices.

[0134] The memory 1130 may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0135] The processor 1110 mentioned above can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0136] In another embodiment of this application, a computer-readable storage medium is provided, on which a program for a target recognition method is stored, wherein when the program for the target recognition method is executed by a processor, it implements the steps of the target recognition method described in any of the foregoing method embodiments.

[0137] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0138] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A target recognition method, characterized in that, include: Obtain the test scenario configuration information uploaded by the test scenario simulation module, and send the test scenario configuration information to the roadside perception system; The system acquires roadside perception data and preliminary target recognition results uploaded by the roadside perception system, as well as vehicle-side perception data uploaded by the vehicle-side control system under test. The preliminary target recognition results are determined by the roadside perception system based on the roadside perception data and the test scenario configuration information. These preliminary target recognition results include candidate controlled targets and candidate uncontrolled targets. Controlled targets are targets belonging to controlled objects within the test scenario plan, while uncontrolled targets are irrelevant targets that are not part of the test scenario plan and have unexpectedly entered the test scenario. The preliminary target recognition result is then verified a second time based on the roadside perception data, the vehicle-side perception data, and the current operating status data of the test scenario. Based on the results of the secondary verification, the final target recognition result is determined.

2. The method according to claim 1, characterized in that, The roadside perception system determines the preliminary target recognition result based on the roadside perception data and the test scenario configuration information, including: The roadside sensing system extracts the features of the target from the roadside sensing data; The features are matched with the controlled target features contained in the test scenario configuration information; Based on the matching results, the target is determined to be either a candidate controlled target or a candidate uncontrolled target.

3. The method according to claim 1, characterized in that, The step of performing a secondary verification on the preliminary target recognition result based on the roadside perception data, the vehicle-side perception data, and the current operating status data of the test scenario includes at least one of the following: Based on the roadside perception data and the vehicle-side perception data, verify whether the candidate controlled target exists in the list of controlled targets included in the test scenario configuration information; Based on the roadside perception data, the vehicle-side perception data, and the current operating status data of the test scenario, verify whether the motion state of the candidate controlled target is consistent with the control commands contained in the current operating status data of the test scenario. Based on the roadside perception data and the vehicle-side perception data, verify whether the candidate uncontrolled target is located outside the planned range of the test scenario. Based on the roadside perception data and the vehicle-side perception data, verify whether the motion trajectory of the candidate uncontrolled target is abnormal; For fuzzy targets, multi-source cross-validation is performed based on the roadside perception data from at least one roadside node and the vehicle-side perception data to determine whether the fuzzy target is a controlled target or an uncontrolled target; wherein, the fuzzy target is a candidate controlled target or a candidate uncontrolled target whose absolute value of the difference between the corresponding feature matching degree and a preset threshold is less than a preset deviation value. The feature matching degree is obtained by matching the features of the target extracted by the roadside perception system from the roadside perception data with the features of the controlled target contained in the test scenario configuration information, and is used to reflect the similarity between the target extracted by the roadside perception system from the roadside perception data and the controlled target contained in the test scenario configuration information.

4. The method according to claim 3, characterized in that, The step of verifying whether the candidate controlled target exists in the controlled target list included in the test scenario configuration information based on the roadside perception data and the vehicle-side perception data includes: The roadside perception data and the vehicle-side perception data are fused to obtain the fusion features of the candidate controlled target; The fused features are matched with the controlled target features contained in the test scenario configuration information; If a controlled target feature that matches the fusion feature exists, the candidate controlled target is determined as a controlled target.

5. The method according to claim 3, characterized in that, The step of verifying whether the motion trajectory of the candidate uncontrolled target is abnormal based on the roadside perception data and the vehicle-side perception data includes: Based on the roadside perception data and the vehicle-side perception data, obtain the position information of the candidate uncontrolled target in multiple consecutive frames, and generate the motion trajectory of the candidate uncontrolled target; Determine whether the motion trajectory exhibits at least one abnormal characteristic among sudden intrusion, irregular reversal, sudden speed change, or sudden direction change; If so, it is determined that the movement trajectory is abnormal.

6. The method according to claim 3, characterized in that, The multi-source cross-validation includes: The first feature of the fuzzy target is extracted from the roadside perception data of the at least one roadside node, and the second feature of the fuzzy target is extracted from the vehicle-side perception data; Determine a cross-source feature consistency measure between the first feature and the second feature; Based on the cross-source feature consistency measure, the fuzzy target is determined to be either a controlled target or an uncontrolled target.

7. The method according to claim 3, characterized in that, The at least one roadside node includes multiple roadside nodes; the multi-source cross-validation includes: Multiple first features of the fuzzy target are extracted from the roadside perception data of multiple roadside nodes, and second features of the fuzzy target are extracted from the vehicle-side perception data. Determine a cross-perspective consistency measure among the plurality of first features, and determine a cross-source consistency measure between the plurality of first features and the second feature; Based on the cross-perspective consistency metric and the cross-source consistency metric, the fuzzy target is determined to be either a controlled target or an uncontrolled target.

8. The method according to claim 1, characterized in that, Also includes: Perform test management operations corresponding to the final target identification result.

9. A target recognition device, characterized in that, include: The scenario configuration information synchronization module is used to obtain the test scenario configuration information uploaded by the test scenario simulation module and send the test scenario configuration information to the roadside perception system. The roadside data acquisition module is used to acquire roadside perception data and preliminary target recognition results uploaded by the roadside perception system. The preliminary target recognition results are determined by the roadside perception system based on the roadside perception data and the test scenario configuration information. These preliminary target recognition results include candidate controlled targets and candidate uncontrolled targets. Controlled targets are targets belonging to controlled objects within the test scenario plan, while uncontrolled targets are irrelevant targets that are not part of the test scenario plan and have accidentally entered the test scenario. The vehicle-side data acquisition module is used to acquire vehicle-side perception data uploaded by the vehicle's control system under test. The secondary verification module is used to perform secondary verification on the preliminary target recognition result based on the roadside perception data, the vehicle-side perception data, and the current operating status data of the test scenario. The result determination module is used to determine the final target recognition result based on the secondary verification result.

10. A vehicle-road-cloud integrated testing system, characterized in that, include: The test scenario simulation module is used to generate test scenario configuration information; The roadside perception system is used to collect roadside perception data and determine preliminary target recognition results based on the roadside perception data and the test scenario configuration information. The preliminary target identification results include candidate controlled targets and candidate uncontrolled targets; among which, controlled targets are targets that belong to controlled objects within the test scenario plan, and uncontrolled targets are irrelevant targets that do not belong to the test scenario plan and accidentally intrude into the test scenario; The vehicle-side control system under test is used to collect vehicle-side sensing data; The cloud platform is communicatively connected to the test scenario simulation module, the roadside perception system, and the vehicle-to-the-test control system, respectively. The cloud platform is configured as follows: Obtain the test scenario configuration information uploaded by the test scenario simulation module, and send the test scenario configuration information to the roadside perception system; Acquire the roadside perception data and the preliminary target recognition result uploaded by the roadside perception system, and acquire the vehicle-side perception data uploaded by the vehicle-side control system under test; The preliminary target recognition result is then verified a second time based on the roadside perception data, the vehicle-side perception data, and the current operating status data of the test scenario. Based on the results of the secondary verification, the final target recognition result is determined.

11. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the target recognition method according to any one of claims 1 to 8.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program for a target recognition method, which, when executed by a processor, implements the steps of the target recognition method according to any one of claims 1-8.

Images