Vehicle driving behavior assessment methods, devices, equipment and storage media

Abstract

This invention relates to the field of autonomous driving technology and discloses a method, apparatus, device, and storage medium for evaluating vehicle driving behavior. The method includes: acquiring road test data of an autonomous vehicle experiencing a traffic light recognition anomaly at a traffic light intersection, and labeling the road test data to obtain labeled data; constructing a simulation environment of the traffic light intersection and the traffic light recognition anomaly based on the labeled data, and conducting driving simulation tests to obtain simulation results; and evaluating the driving behavior of the autonomous vehicle at the traffic light intersection based on the simulation results to obtain an evaluation result of the driving behavior. This solves the technical problem in the prior art of being unable to accurately evaluate the rationality of the driving behavior of an autonomous vehicle when passing through a traffic light intersection.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a method, apparatus, device, and storage medium for evaluating vehicle driving behavior. Background Technology

[0002] Traffic lights are an important part of modern transportation and one of the scenarios where autonomous driving systems interact most frequently. In this scenario, factors such as obstructions from other vehicles, glare from traffic lights, and blurred vision due to rain can prevent vehicles from accurately receiving the traffic light status, thus affecting the autonomous driving system's judgment and increasing the likelihood of accidents caused by the system violating traffic regulations.

[0003] In scenarios such as obstruction by a vehicle ahead, glare from traffic lights, or obscured vision by rain, the autonomous driving system must be able to react appropriately based on the historical state of the traffic lights, the current state of the unobstructed traffic lights, and the status of other vehicles on the road. This is one of the key objectives to be achieved before autonomous driving can be commercially deployed.

[0004] Therefore, in order to achieve rapid iteration of autonomous driving systems in traffic light scenarios, how to evaluate the rationality of the driving behavior of autonomous vehicles when passing through traffic light intersections has become a technical problem that needs to be solved by those skilled in the art. Summary of the Invention

[0005] The main objective of this invention is to solve the technical problem in the prior art that it is impossible to accurately assess the rationality of the driving behavior of autonomous vehicles when passing through traffic light intersections.

[0006] The first aspect of this invention provides a method for evaluating vehicle driving behavior, comprising: acquiring road test data of an autonomous vehicle when a traffic light recognition anomaly occurs at a traffic light intersection, and annotating the road test data to obtain annotated data; constructing a simulation environment of a traffic light intersection and a traffic light recognition anomaly based on the annotated data, and conducting driving simulation tests to obtain simulation results; and evaluating the driving behavior of the autonomous vehicle at the traffic light intersection based on the simulation results to obtain an evaluation result of the driving behavior.

[0007] Optionally, in a first implementation of the first aspect of the present invention, the step of acquiring road test data when an autonomous vehicle experiences a traffic light recognition anomaly at a traffic light intersection, and labeling the road test data to obtain labeled data, includes: acquiring multi-frame environmental image data collected by the autonomous vehicle as it passes through the road intersection; performing scene recognition on the multi-frame environmental image data based on preset scene labels to determine whether the target scene corresponding to the multi-frame environmental image data is a traffic light intersection recognition anomaly; if so, acquiring driving data of the autonomous vehicle in the target scene; saving the driving data to obtain road test data, and labeling the road test data based on the scene labels to obtain labeled data.

[0008] Optionally, in a second implementation of the first aspect of the present invention, the step of annotating the road test data based on the scene tags to obtain annotated data includes: classifying the road test data into scenarios based on the scene tags to obtain multiple road test scenarios corresponding to the road test data; and annotating each road test scenario in the road test data with the corresponding scene tags to obtain annotated data.

[0009] Optionally, in a third implementation of the first aspect of the present invention, the step of constructing a simulation environment of a traffic light intersection and a traffic light recognition anomaly based on the labeled data, and conducting driving simulation tests to obtain simulation results includes: extracting corresponding scene fragment data from the road test data based on the scene labels; constructing a road test environment for the autonomous vehicle based on the scene fragment data to obtain a simulation environment; and conducting driving tests based on the simulation environment to obtain simulation results.

[0010] Optionally, in a fourth implementation of the first aspect of the present invention, the construction of the road test environment for the autonomous vehicle based on the scene fragment data to obtain a simulation environment includes: extracting semantic data of the autonomous vehicle corresponding to the current scene and trajectory state data of surrounding vehicles from the scene fragment data; extracting first data corresponding to the dynamic elements from the semantic data of the current scene based on the dynamic elements in the scene tags to obtain dynamic scene data; extracting second data corresponding to the static elements from the trajectory state data of surrounding vehicles based on the static elements in the scene tags to obtain real-time background data; and concatenating the dynamic scene data and the real-time background data to construct a simulation environment of a traffic light intersection where a traffic light recognition anomaly occurs.

[0011] Optionally, in a fifth implementation of the first aspect of the present invention, the step of conducting driving tests based on the simulation environment to obtain simulation results includes: acquiring road data corresponding to the scene segment data; determining multiple target simulation scenarios in the simulation environment based on the road data; controlling the autonomous vehicle to drive in the multiple target simulation scenarios to obtain multiple scenario comfort scores and multiple scenario continuous passage scores of the autonomous vehicle; and determining the simulation results of the autonomous vehicle based on the multiple scenario comfort scores and the multiple scenario continuous passage scores.

[0012] Optionally, in a sixth implementation of the first aspect of the present invention, the step of evaluating the driving behavior of the autonomous vehicle at the traffic light intersection based on the simulation results to obtain an evaluation result of the driving behavior includes: determining the driving pass rate corresponding to the abnormal driving behavior of the autonomous vehicle at the traffic light intersection based on the simulation results; determining the weight corresponding to the abnormal driving behavior; and evaluating the driving behavior of the autonomous vehicle based on the driving pass rate and the weight to obtain an evaluation result of the driving behavior.

[0013] A second aspect of the present invention provides a vehicle driving behavior evaluation device, comprising: an annotation module, configured to acquire road test data of an autonomous vehicle at a traffic light intersection when a traffic light recognition anomaly occurs, and to annotate the road test data to obtain annotated data; a simulation module, configured to construct a simulation environment of a traffic light intersection and a traffic light recognition anomaly based on the annotated data, and to perform driving simulation tests to obtain simulation results; and an evaluation module, configured to evaluate the driving behavior of the autonomous vehicle at the traffic light intersection based on the simulation results to obtain an evaluation result of the driving behavior.

[0014] Optionally, in a first implementation of the second aspect of the present invention, the annotation module includes: an acquisition unit, configured to acquire multiple frames of environmental image data collected by the autonomous vehicle when passing through a road intersection; a judgment unit, configured to perform scene recognition on the multiple frames of environmental image data based on preset scene labels, and determine whether the target scene corresponding to the multiple frames of environmental image data is a traffic light intersection recognition anomaly; an acquisition unit, configured to acquire driving data of the autonomous vehicle in the target scene if so; and an annotation unit, configured to save the driving data to obtain road test data, and annotate the road test data based on the scene labels to obtain annotated data.

[0015] Optionally, in a second implementation of the second aspect of the present invention, the annotation unit is specifically used for: classifying the road test data based on the scene labels to obtain multiple road test scenes corresponding to the road test data; and annotating each road test scene in the road test data with the corresponding scene labels to obtain annotated data.

[0016] Optionally, in a third implementation of the second aspect of the present invention, the simulation module includes: an extraction unit, configured to extract corresponding scene fragment data from the road test data based on the scene annotations; a construction unit, configured to construct the road test environment of the autonomous vehicle based on the scene fragment data to obtain a simulation environment; and a simulation unit, configured to perform driving tests based on the simulation environment to obtain simulation results.

[0017] Optionally, in a fourth implementation of the second aspect of the present invention, the construction unit is specifically used to: extract semantic data of the current scene corresponding to the autonomous vehicle and trajectory state data of surrounding vehicles from the scene fragment data; extract first data corresponding to the dynamic elements from the semantic data of the current scene based on the dynamic elements in the scene tags to obtain dynamic scene data; extract second data corresponding to the static elements from the trajectory state data of surrounding vehicles based on the static elements in the scene tags to obtain real-time background data; and concatenate the dynamic scene data and the real-time background data to construct a simulation environment of a traffic light intersection where a traffic light recognition anomaly occurs.

[0018] Optionally, in a fifth implementation of the second aspect of the present invention, the simulation unit is specifically used for: acquiring road data corresponding to the scene fragment data; determining multiple target simulation scenarios in the simulation environment based on the road data; controlling the autonomous vehicle to drive in the multiple target simulation scenarios to obtain multiple scenario comfort scores and multiple scenario continuous passage scores of the autonomous vehicle; and determining the simulation result of the autonomous vehicle based on the multiple scenario comfort scores and the multiple scenario continuous passage scores.

[0019] Optionally, in a sixth implementation of the second aspect of the present invention, the evaluation module is specifically used to: determine the driving pass rate corresponding to the abnormal driving behavior of the autonomous vehicle at the traffic light intersection based on the simulation results; determine the weight corresponding to the abnormal driving behavior; and evaluate the driving behavior of the autonomous vehicle based on the driving pass rate and the weight to obtain the evaluation result of the driving behavior.

[0020] A third aspect of the present invention provides a vehicle driving behavior assessment device, comprising: a memory and at least one processor, wherein the memory stores instructions, and the memory and the at least one processor are interconnected via a circuit;

[0021] The at least one processor invokes the instructions in the memory to cause the vehicle driving behavior assessment device to perform the various steps of the vehicle driving behavior assessment method described above.

[0022] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the steps of the above-described vehicle driving behavior assessment method.

[0023] The technical solution provided by this invention involves acquiring road test data of an autonomous vehicle encountering a traffic light recognition anomaly at a traffic light intersection, and annotating the road test data to obtain annotated data. Based on the annotated data, a simulation environment of a traffic light intersection with a traffic light recognition anomaly is constructed, and driving simulation tests are conducted to obtain simulation results. Based on the simulation results, the driving behavior of the autonomous vehicle at the traffic light intersection is evaluated to obtain an evaluation result of the driving behavior. This solves the technical problem in existing technologies where the rationality of the driving behavior of autonomous vehicles when passing through traffic light intersections cannot be accurately evaluated. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the first embodiment of the vehicle driving behavior assessment method provided by the present invention;

[0025] Figure 2 A schematic diagram of a second embodiment of the vehicle driving behavior assessment method provided by the present invention;

[0026] Figure 3 A schematic diagram of the third embodiment of the vehicle driving behavior assessment method provided by the present invention;

[0027] Figure 4 A schematic diagram of the first embodiment of the vehicle driving behavior assessment device provided by the present invention;

[0028] Figure 5 This is a schematic diagram of a second embodiment of the vehicle driving behavior assessment device provided by the present invention;

[0029] Figure 6 This is a schematic diagram of an embodiment of the vehicle driving behavior assessment device provided by the present invention. Detailed Implementation

[0030] This invention provides a method, apparatus, device, and storage medium for evaluating vehicle driving behavior. The technical solution of this invention first acquires road test data of an autonomous vehicle experiencing traffic light recognition anomalies at a traffic light intersection, and then labels the road test data to obtain labeled data. Based on the labeled data, a simulation environment of a traffic light intersection experiencing traffic light recognition anomalies is constructed, and driving simulation tests are conducted to obtain simulation results. Based on the simulation results, the driving behavior of the autonomous vehicle at the traffic light intersection is evaluated to obtain the driving behavior evaluation result. This solves the technical problem in the prior art of being unable to accurately evaluate the rationality of the driving behavior of an autonomous vehicle when passing through a traffic light intersection.

[0031] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0032] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 The first embodiment of the vehicle driving behavior assessment method in this invention includes:

[0033] 101. Obtain road test data when an autonomous vehicle encounters a traffic light recognition anomaly at a traffic light intersection, and annotate the road test data to obtain annotated data;

[0034] In this embodiment, when the vehicle enters a traffic light intersection, if the safety operator discovers abnormal behavior in the autonomous vehicle, such as sudden braking, running a red light, or failing to start when the light is green, the operator will report the abnormal behavior through the vehicle's driving system. The data acquisition device will record the detection information of the surrounding environment, road, obstacles, traffic lights, and the vehicle's own status throughout the entire process (i.e., within a certain period of time before and after the event), thereby obtaining road test data.

[0035] Specifically, the recorded data can reproduce road test scenarios in a simulated environment. Furthermore, the collected data is tagged with different scene labels based on different preset scenarios, mainly including the following categories: 1. Traffic light obstruction; 2. Rain blur; 3. Traffic light reflection; 4. Misidentification of vehicle taillights, etc. In addition to broad categories, more detailed labels are applied based on the status of surrounding vehicles, whether the autonomous vehicle starts from the first row, and the type of object obstructing the traffic light.

[0036] In this embodiment, the collected road test data is manually labeled to indicate the behavioral states that the autonomous vehicle should be in at different time periods. These are mainly divided into two categories: 1. Green light: Smooth passage; 2. Red light: Stable stop before the stop line. The labeling principle is to infer the correct state of the traffic lights when they malfunction (such as obstruction, glare, or rain blurring) by using information such as the traffic light status in the video data and the surrounding traffic flow, and to provide the time intervals during which the autonomous vehicle should be in a smooth passage or stable stop state.

[0037] 102. Construct a simulation environment for a traffic light intersection with traffic light recognition anomalies based on labeled data, and conduct driving simulation tests to obtain simulation results;

[0038] In this embodiment, the labeled scene is replayed in a simulation environment, and the autonomous driving software is launched in the simulation environment. Based on the positioning and perception information in the replayed data, the autonomous driving system outputs control commands to control the autonomous vehicle to drive in the simulation environment.

[0039] In this embodiment, corresponding scene fragment data is extracted from road test data based on scene annotation; the road test environment of autonomous vehicle is constructed using scene fragment data to obtain a simulation environment; further, the autonomous vehicle is placed in the simulation environment for simulation testing, i.e., autonomous driving testing, to obtain the simulation results of the autonomous vehicle in the simulation environment.

[0040] Specifically, during the simulation, the autonomous driving system calculates the current behavioral decision in real time based on the detected status of traffic lights and information such as surrounding traffic flow and lanes, and adjusts the driving state of the autonomous vehicle according to the decision.

[0041] 103. Based on the simulation results, evaluate the driving behavior of autonomous vehicles at traffic light intersections and obtain the evaluation results of the driving behavior.

[0042] In this embodiment, the driving behavior of the autonomous vehicle at a traffic light intersection is evaluated based on the simulation results of the autonomous vehicle in the simulation environment. The specific evaluation criteria are as follows: First, during the marked red light phase (when the vehicle comes to a stable stop before the preset road sign): the system will continuously monitor the internal state of the autonomous driving system. When the system analyzes and predicts that the current frame is a red light state, a virtual fence is placed behind the preset road sign, such as the stop line. The attribute of the virtual fence is "traffic light", and the decision given by the system is to stop behind the fence.

[0043] Specifically, by monitoring these indicators, it is possible to accurately determine whether the autonomous driving system of an autonomous vehicle has given a decision to stop at a red light. For the corresponding marked time period, the autonomous driving system is required to continuously give a decision to stop at a red light. If so, "Mark Evaluation Criterion 1" is marked as pass; otherwise, it is marked as fail. 2. Manually marked green light (smooth passage) phase: Comfort: It is required that no uncomfortable events caused by the fence in Evaluation Criterion 1 occur during this marked time period.

[0044] In this embodiment, the specific implementation method is to calculate the somatosensory value of each video frame corresponding to the video data collected by the autonomous vehicle during driving in real time. When the somatosensory value is lower than a preset threshold, it is checked whether the "fence" in evaluation criterion 1 has occurred within a preset time period. If so, the corresponding evaluation criterion 2 "comfort mark" is marked as "fail"; if the corresponding evaluation criterion 2 "comfort mark" is marked as "pass" throughout the entire driving process of the autonomous vehicle, that is, continuous passage. Here, "continuous passage" means that the autonomous vehicle is in a state of continuous passage within a preset time period.

[0045] Specifically, it is calculated whether there are times when the vehicle speed is less than a specified threshold within the preset time period; if there are times when the vehicle speed is less than the specified threshold within the preset time period, then the autonomous vehicle is marked as failing in the evaluation standard 2 "accessibility" of the driving process; otherwise, the autonomous vehicle is marked as passing in the evaluation standard 2 "accessibility" of the driving process.

[0046] Furthermore, if all the above evaluation criteria are pass, the scenario is marked as pass; if any criterion is fail, the scenario is marked as fail, which is the final pass criterion for the scenario.

[0047] In this embodiment, by comparing the overall pass rate of different autonomous driving versions and the pass rate of sub-scenes under different scene labels, the overall performance trend of different autonomous driving versions in traffic light intersection scenarios and in preset specific scenarios can be accurately reflected. Specifically, by collecting problem data of autonomous driving systems corresponding to autonomous vehicles at traffic light intersections, a dataset is established. The expected vehicle behavior is analyzed and predicted using preset labeling methods. The vehicle behavior is mainly divided into two states: passing and waiting at the intersection. The driving state of the autonomous vehicle is evaluated using this vehicle behavior as a standard. The driving state includes whether the current driving state of the autonomous vehicle is reasonable and whether any traffic rule violations have occurred.

[0048] The evaluation system proposed in this invention can achieve coverage of various scenarios through continuous iteration of the dataset used for evaluation. Furthermore, when comparing different versions of autonomous driving software, the same evaluation dataset can be used to eliminate biases in comparison results caused by different test scenarios. Finally, regression is prevented by tracking repaired scenarios.

[0049] In this embodiment of the invention, road test data of an autonomous vehicle encountering a traffic light recognition anomaly at a traffic light intersection is acquired and annotated to obtain labeled data. A simulation environment simulating a traffic light intersection and the traffic light recognition anomaly is constructed based on the labeled data, and driving simulation tests are conducted to obtain simulation results. The driving behavior of the autonomous vehicle at the traffic light intersection is evaluated based on the simulation results to obtain an evaluation result. This solves the technical problem in the prior art of being unable to accurately evaluate the rationality of the vehicle's driving behavior when passing through a traffic light intersection.

[0050] Please see Figure 2 The second embodiment of the vehicle driving behavior assessment method in this invention includes:

[0051] 201. Acquire multi-frame environmental image data collected by autonomous vehicles when passing through road intersections;

[0052] In this embodiment, the autonomous vehicle's data acquisition device acquires multi-frame environmental image data when the autonomous vehicle passes through a road intersection. The data acquisition device includes at least: an image acquisition device, a lidar, a millimeter-wave radar, an IMU, a GPS, a vehicle speed signal acquisition device, a steering wheel angle signal acquisition device, a brake pedal opening degree acquisition device, and an accelerator pedal opening degree acquisition device.

[0053] Furthermore, the data acquisition device adopts multi-source input synchronization technology to ensure the synchronization of various driving data and realize the synchronous acquisition of multiple driving data.

[0054] 202. Based on preset scene labels, perform scene recognition on multi-frame environmental image data and determine whether the target scene corresponding to the multi-frame environmental image data is a traffic light intersection recognition anomaly.

[0055] In this embodiment, scene recognition is performed on multiple frames of environmental image data based on preset scene labels to obtain recognition results. The recognition results include: the scene category corresponding to the driving data, the location information, category information, and ID information of the target object in the driving data.

[0056] Furthermore, based on the recognition results, planning and localization algorithms are used to obtain vehicle planning and localization results, respectively. Then, based on the vehicle planning and localization results, a control algorithm is used to obtain vehicle control results. The aforementioned scene recognition algorithms include: perception algorithm, planning algorithm, control algorithm, and localization algorithm.

[0057] In this embodiment, image environment data is identified based on preset scene labels, such as whether traffic lights are identified in the current scene, whether traffic lights are partially obscured, and whether there are stationary vehicles around the current autonomous vehicle. Based on this, it is determined whether the target scene corresponding to the multi-frame environmental image data is an abnormal traffic light intersection recognition.

[0058] 203. If so, then collect driving data of the autonomous vehicle in the target scenario;

[0059] In this embodiment, the driving data is driving data of predefined scenario categories: During implementation, the perception algorithm classifies the scenarios corresponding to the driving data. If the scenario corresponding to the driving data is a predefined scenario category, it conforms to the preset data collection rules. It should be noted that the above-mentioned predefined scenario categories include scenarios such as a vehicle suddenly braking, a vehicle running a red light, a green light in front of the vehicle but surrounding vehicles being stationary, and a vehicle having no obstacles around it but not proceeding on a green light. This embodiment of the invention does not impose specific restrictions on the above-mentioned predefined scenario categories.

[0060] 204. Save the driving data to obtain road test data. Classify the road test data according to the scene tags to obtain multiple road test scenes corresponding to the road test data.

[0061] In this embodiment, driving data is saved to obtain road test data, and the road test data is filtered to remove road test data that does not meet the requirements.

[0062] Furthermore, based on preset scene tags, such as whether there are traffic lights in the current scene, whether there are obstacles around the autonomous vehicle in the current scene, whether there are obstructions in the current scene, whether it is raining in the current environment, whether there is rain obscuring traffic lights in the current environment, and whether there is a green light in the current environment but vehicles are not moving, etc., a scene tag is determined for the current scene based on the keywords corresponding to each scene. The road test data is then classified into scenes based on the scene tags to obtain multiple road test scenes related to the road test data.

[0063] 205. Use the corresponding scene labels to annotate each road test scene in the road test data to obtain the labeled data;

[0064] In this embodiment, the aforementioned road test data is labeled according to the corresponding scene tags to obtain labeled data. Data labeling is the process of processing unprocessed data such as speech, images, text, and video into machine-recognizable information. The main types of data labeling are image labeling, speech labeling, 3D point cloud labeling, and text labeling. Specifically, this embodiment primarily uses image labeling, which involves processing unprocessed image data to convert it into machine-recognizable information, which is then fed into artificial intelligence algorithms and models for use.

[0065] In this embodiment, data labeling is achieved by attaching labels to provide the machine system with a large number of learning samples. The data that the machine needs to recognize and distinguish is labeled, and then the computer continuously learns the characteristics of these data, ultimately enabling the computer to recognize them autonomously.

[0066] 206. Construct a simulation environment for a traffic light intersection with traffic light recognition anomalies based on labeled data, and conduct driving simulation tests to obtain simulation results;

[0067] 207. Based on simulation results, determine the driving pass rate corresponding to abnormal driving behavior of autonomous vehicles at traffic light intersections;

[0068] In this embodiment, abnormal driving behaviors include: sudden braking, running red lights, and failing to start when obstructing traffic during green lights.

[0069] Based on simulation results, the pass rate corresponding to abnormal driving behaviors of autonomous vehicles at traffic light intersections is determined. The abnormal driving behaviors of the target autonomous vehicles can be randomly sampled from a vehicle-to-everything (V2X) data database, or obtained from driving recorders installed on multiple target autonomous vehicles. This application does not specifically limit the method of obtaining abnormal driving behaviors; those skilled in the art can determine it according to actual needs. Based on the abnormal driving behaviors, the pass rate corresponding to various abnormal driving behaviors is determined.

[0070] 208. Determine the weights corresponding to abnormal driving behaviors, and evaluate the driving behavior of autonomous vehicles based on the driving pass rate and weights to obtain the evaluation results of driving behavior.

[0071] In this embodiment, the weights are weight coefficients corresponding to various abnormal driving behaviors, representing the impact of various abnormal driving behavior data on the evaluation of abnormal driving behaviors. The weights can be determined based on the abnormal driving behavior rate and the target abnormal driving behavior rate. Specifically, the weight can be determined by calculating the entropy value corresponding to a certain target abnormal driving behavior rate, and then determining the weight corresponding to that abnormal driving behavior based on the entropy value. This process can be repeated to obtain the weights corresponding to various abnormal driving behaviors. Of course, other methods can also be used to determine the weights; no specific calculation method for the weights corresponding to various abnormal driving behaviors is limited here.

[0072] Specifically, the driving behavior of autonomous vehicles is evaluated based on the target abnormal driving behavior rate and weights corresponding to various abnormal driving behaviors, resulting in an evaluation result. This evaluation result characterizes the safety of the target autonomous vehicle's driving behavior. The evaluation result can be considered as the driving behavior safety entropy. Based on the obtained target abnormal driving behavior rate and weights corresponding to various abnormal driving behaviors of the target autonomous vehicle, a formula for calculating the safety entropy can be determined. This formula yields the safety entropy value, which is then used as the evaluation result characterizing the autonomous vehicle's driving behavior.

[0073] Step 206 in this embodiment is similar to step 103 in the first embodiment, and will not be described again here.

[0074] In this embodiment of the invention, road test data of an autonomous vehicle encountering a traffic light recognition anomaly at a traffic light intersection is acquired and annotated to obtain labeled data. A simulation environment simulating a traffic light intersection and the traffic light recognition anomaly is constructed based on the labeled data, and driving simulation tests are conducted to obtain simulation results. The driving behavior of the autonomous vehicle at the traffic light intersection is evaluated based on the simulation results to obtain an evaluation result. This solves the technical problem in the prior art of being unable to accurately evaluate the rationality of the vehicle's driving behavior when passing through a traffic light intersection.

[0075] Please see Figure 3 The third embodiment of the vehicle driving behavior assessment method in this invention includes:

[0076] 301. Obtain road test data when an autonomous vehicle experiences a traffic light recognition anomaly at a traffic light intersection, and annotate the road test data to obtain annotated data;

[0077] 302. Based on scene annotation, extract corresponding scene fragment data from the road test data;

[0078] In this embodiment, by annotating the scene, the real scene of interaction with the autonomous vehicle is extracted from the road test data, and the scene segment data corresponding to the scene segment is performed on the preset scene based on the scene data in the real scene.

[0079] Specifically, by segmenting the preset scene, scene fragment data of multiple simulation scenes are obtained; further, multiple simulation scenes are constructed based on the scene fragment data, and each target simulation scene includes information such as targets, vehicles, road networks and traffic lights, thereby increasing the richness and realism of each target simulation scene.

[0080] 303. Extract semantic data of the autonomous vehicle corresponding to the current scene and trajectory status data of surrounding vehicles from scene fragment data;

[0081] In this embodiment, the generated simulation environment is ultimately used to control the driving of autonomous vehicles. Therefore, the generated simulation environment should be converted into test scene data from the viewpoint of the autonomous vehicle being tested, so that the autonomous vehicle being tested can perform autonomous driving based on the semantic data of the current scene corresponding to the autonomous vehicle and the trajectory state data of surrounding vehicles extracted from the scene fragment data.

[0082] 304. Based on the dynamic elements in the scene tags, extract the first data corresponding to the dynamic elements from the semantic data of the current scene to obtain dynamic scene data;

[0083] In this embodiment, the location of the autonomous vehicle when receiving test scenario data is first analyzed. For example, based on the location of the autonomous vehicle, its driving speed, driving acceleration, and the time taken for data processing and transmission when the autonomous vehicle or roadside perception device collects road environment perception data, the location of the autonomous vehicle when receiving test scenario data is predicted.

[0084] Based on the dynamic elements in the scene labels, the first data corresponding to the dynamic elements is extracted from the semantic data of the current scene to obtain dynamic scene data. The position mapping relationship is determined by using the position of the autonomous vehicle when receiving test scene data and the position of the autonomous vehicle or roadside perception device when collecting road environment perception data. Then, based on the position mapping relationship, the simulation environment is converted into dynamic scene data of the viewpoint when the autonomous vehicle receives test scene data.

[0085] 305. Based on the static elements in the scene labels, extract the second data corresponding to the static elements from the trajectory status data of surrounding vehicles to obtain real-time background data;

[0086] In this embodiment, the second data corresponding to the static elements extracted from the trajectory state data of surrounding vehicles based on the static elements in the scene label is fused to obtain both geometric frame data and detailed road environment data at the location, such as texture data, light wave and electromagnetic wave reflection characteristic data, etc., thus obtaining real-time background data in the driving scene.

[0087] 306. By splicing dynamic scene data and real-time background data, a simulation environment is constructed for a traffic light intersection where traffic light recognition anomalies occur.

[0088] In this embodiment, dynamic scene data and real-time background data are spliced ​​together to construct a simulation environment of a traffic light intersection where traffic light recognition anomalies occur.

[0089] The so-called scenario is the process of comprehensive interaction between a vehicle and other vehicles, facilities, environment, roads and other elements in the traffic environment. Various scenario elements are combined to form a collection of different scenarios. Autonomous driving test scenarios are essentially a refinement of all relevant scenario elements.

[0090] Specifically, for autonomous driving testing, whether in a simulation environment or in a pre-set real-world test site, the most commonly used method is scenario-based functional testing. Its advantage lies in overcoming the limitations of mileage-based testing methods, playing a significant role in improving system development efficiency and product deployment efficiency. The collected dynamic scene data and real-time background data are stitched together to construct a simulation environment of a traffic light intersection where traffic light recognition anomalies occur. Driving simulations of autonomous vehicles are then conducted within this simulation environment.

[0091] 307. Obtain road data corresponding to scene fragment data, and determine multiple target simulation scenes in the simulation environment based on the road data;

[0092] In this embodiment, to simplify the repetitive management process from road testing to simulation and save management costs, it is no longer necessary to generate simulation sets based on road test data. Instead, road data can be directly labeled to construct a road database. This road database includes multiple sets of test data, which include labeled road data and the corresponding autonomous driving simulation scenarios.

[0093] Accordingly, since the road database stores a large amount of road data with labeled autonomous driving scenario information, after obtaining the simulation regression test request, the road data corresponding to the target autonomous driving scenario information can be obtained from the road database according to the simulation regression test request.

[0094] In this embodiment, a high-precision semantic map is generated based on the road data, and multiple target simulation scenarios in the simulation environment are determined based on the high-precision semantic map.

[0095] 308. Control the autonomous vehicle to drive in multiple target simulation scenarios, and obtain the autonomous vehicle's comfort score and continuous passage score in multiple scenarios.

[0096] In this embodiment, the autonomous vehicle is controlled to drive in multiple target simulation scenarios. The server performs autonomous driving simulation tests on each of the multiple target simulation scenarios using a preset autonomous driving algorithm, obtaining multiple initial simulation test results. The server determines the autonomous driving simulation test result based on these initial simulation test results. The preset autonomous driving algorithm includes a preset scenario safety scoring algorithm, a preset scenario comfort scoring algorithm, and / or a preset driver position difference algorithm. Furthermore, multiple scenario comfort scores and multiple scenario continuous passage scores for the autonomous vehicle are obtained. Based on these scenario comfort scores and multiple scenario continuous passage scores, the performance of the corresponding autonomous driving system for the current autonomous vehicle is determined to improve the safety of the autonomous vehicle's driving state.

[0097] 309. Based on comfort scores and continuous passage scores in multiple scenarios, determine the simulation results of autonomous vehicles;

[0098] In this embodiment, it should be noted that the autonomous driving simulation test results are used to indicate the scene continuous passage score, scene comfort score, and positional difference between the target and the driver. Furthermore, based on multiple scene comfort scores and multiple scene continuous passage scores, the simulation result of the autonomous vehicle is determined. For example, if the scene continuous passage score is 3 points, the scene comfort score is 5 points, and the positional difference between the target and the driver is 0.01 meters, then the server determines that the scene safety level is low and the scene comfort level is low. Further, based on multiple scene comfort scores and multiple scene continuous passage scores, the simulation result of the autonomous vehicle is determined.

[0099] 310. Based on the simulation results, evaluate the driving behavior of autonomous vehicles at traffic light intersections and obtain the evaluation results of the driving behavior.

[0100] Steps 301 and 310 in this embodiment are similar to steps 101 and 103 in the first embodiment, and will not be repeated here.

[0101] In this embodiment of the invention, road test data of an autonomous vehicle encountering a traffic light recognition anomaly at a traffic light intersection is acquired and annotated to obtain labeled data. A simulation environment simulating a traffic light intersection and the traffic light recognition anomaly is constructed based on the labeled data, and driving simulation tests are conducted to obtain simulation results. The driving behavior of the autonomous vehicle at the traffic light intersection is evaluated based on the simulation results to obtain an evaluation result. This solves the technical problem in the prior art of being unable to accurately evaluate the rationality of the vehicle's driving behavior when passing through a traffic light intersection.

[0102] The vehicle driving behavior assessment method in the embodiments of the present invention has been described above. The vehicle driving behavior assessment device in the embodiments of the present invention will be described below. Please refer to [link / reference]. Figure 4 The first embodiment of the vehicle driving behavior assessment device in this invention includes:

[0103] The annotation module 401 is used to acquire road test data when the autonomous vehicle encounters a traffic light recognition anomaly at a traffic light intersection, and to annotate the road test data to obtain annotated data;

[0104] The simulation module 402 is used to construct a simulation environment of a traffic light intersection and a traffic light recognition anomaly based on the labeled data, and to conduct driving simulation tests to obtain simulation results.

[0105] The evaluation module 403 is used to evaluate the driving behavior of the autonomous vehicle at the traffic light intersection based on the simulation results, and obtain the evaluation result of the driving behavior.

[0106] In this embodiment of the invention, road test data of an autonomous vehicle encountering a traffic light recognition anomaly at a traffic light intersection is acquired and annotated to obtain labeled data. A simulation environment simulating a traffic light intersection and the traffic light recognition anomaly is constructed based on the labeled data, and driving simulation tests are conducted to obtain simulation results. The driving behavior of the autonomous vehicle at the traffic light intersection is evaluated based on the simulation results to obtain an evaluation result. This solves the technical problem in the prior art of being unable to accurately evaluate the rationality of the vehicle's driving behavior when passing through a traffic light intersection.

[0107] Please see Figure 5 A second embodiment of the vehicle driving behavior assessment device in this invention specifically includes:

[0108] The annotation module 401 is used to acquire road test data when the autonomous vehicle encounters a traffic light recognition anomaly at a traffic light intersection, and to annotate the road test data to obtain annotated data;

[0109] The simulation module 402 is used to construct a simulation environment of a traffic light intersection and a traffic light recognition anomaly based on the labeled data, and to conduct driving simulation tests to obtain simulation results.

[0110] The evaluation module 403 is used to evaluate the driving behavior of the autonomous vehicle at the traffic light intersection based on the simulation results, and obtain the evaluation result of the driving behavior.

[0111] In this embodiment, the annotation module 401 includes:

[0112] The acquisition unit 4011 is used to acquire multiple frames of environmental image data collected by the autonomous vehicle when it passes through a road intersection.

[0113] The judgment unit 4012 is used to perform scene recognition on the multi-frame environmental image data based on preset scene labels, and to determine whether the target scene corresponding to the multi-frame environmental image data is a traffic light intersection recognition anomaly.

[0114] The acquisition unit 4013 is used to acquire driving data of the autonomous vehicle in the target scenario if the condition is met.

[0115] The annotation unit 4014 is used to save the driving data to obtain road test data, and to annotate the road test data based on the scene label to obtain annotated data.

[0116] In this embodiment, the annotation unit 4014 is specifically used for:

[0117] Based on the scene labels, the road test data is classified into multiple road test scenes corresponding to the road test data.

[0118] The road test scenarios in the road test data are labeled using the corresponding scene tags to obtain labeled data.

[0119] In this embodiment, the simulation module 402 includes:

[0120] Extraction unit 4021 is used to extract corresponding scene fragment data from the road test data based on the scene annotation;

[0121] The construction unit 4022 is used to construct the road test environment of the autonomous vehicle based on the scene fragment data to obtain the simulation environment;

[0122] The simulation unit 4023 is used to perform driving tests based on the simulation environment and obtain simulation results.

[0123] In this embodiment, the building unit 4022 is specifically used for:

[0124] Extract the semantic data of the autonomous vehicle corresponding to the current scene and the trajectory status data of surrounding vehicles from the scene fragment data;

[0125] Based on the dynamic elements in the scene tags, the first data corresponding to the dynamic elements is extracted from the semantic data of the current scene to obtain dynamic scene data;

[0126] Based on the static elements in the scene label, the second data corresponding to the static elements is extracted from the trajectory status data of the surrounding vehicles to obtain real-time background data;

[0127] The dynamic scene data and the real-time background data are spliced ​​together to construct a simulation environment of a traffic light intersection where a traffic light recognition anomaly occurs.

[0128] In this embodiment, the simulation unit 4023 is specifically used for:

[0129] Obtain the road data corresponding to the scene segment data;

[0130] Based on the road data, multiple target simulation scenarios are determined in the simulation environment;

[0131] The autonomous vehicle is controlled to drive in multiple target simulation scenarios to obtain multiple scenario comfort scores and multiple scenario continuous passage scores for the autonomous vehicle;

[0132] The simulation results of the autonomous vehicle are determined based on the comfort scores and continuous passage scores of the multiple scenarios.

[0133] In this embodiment, the evaluation module 403 is specifically used for:

[0134] Based on the simulation results, the driving pass rate corresponding to the abnormal driving behavior of the autonomous vehicle at the traffic light intersection is determined;

[0135] Determine the weights corresponding to the abnormal driving behaviors;

[0136] Based on the driving pass rate and the weight, the driving behavior of the autonomous vehicle is evaluated to obtain the evaluation result of the driving behavior.

[0137] In this embodiment of the invention, road test data of an autonomous vehicle encountering a traffic light recognition anomaly at a traffic light intersection is acquired and labeled to obtain labeled data. A simulation environment simulating a traffic light intersection and the traffic light recognition anomaly is constructed based on the labeled data, and driving simulation tests are conducted to obtain simulation results. The driving behavior of the autonomous vehicle at the traffic light intersection is evaluated based on the simulation results to obtain an evaluation result. This solves the technical problem in the prior art of being unable to accurately evaluate the rationality of the driving behavior of an autonomous vehicle when passing through a traffic light intersection.

[0138] above Figure 4 and Figure 5 The vehicle driving behavior assessment device in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The vehicle driving behavior assessment device in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0139] Figure 6 This is a schematic diagram of the structure of a vehicle driving behavior assessment device 600 provided in an embodiment of the present invention. The vehicle driving behavior assessment device 600 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 610 (e.g., one or more processors) and a memory 620, and one or more storage media 630 (e.g., one or more mass storage devices) for storing application programs 633 or data 632. The memory 620 and storage media 630 can be temporary or persistent storage. The program stored in the storage media 630 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the vehicle driving behavior assessment device 600. Furthermore, the processor 610 may be configured to communicate with the storage media 630 and execute the series of instruction operations in the storage media 630 on the vehicle driving behavior assessment device 600 to implement the steps of the vehicle driving behavior assessment method provided in the above-described method embodiments.

[0140] The vehicle driving behavior assessment device 600 may also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input / output interfaces 660, and / or one or more operating systems 631, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 6 The illustrated vehicle driving behavior assessment device structure does not constitute a limitation on the vehicle driving behavior assessment device provided in this application. It may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0141] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the above-described vehicle driving behavior assessment method.

[0142] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0143] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0144] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for assessing vehicle driving behavior, characterized in that, The vehicle driving behavior assessment method includes: Acquire road test data when an autonomous vehicle experiences a traffic light recognition anomaly at a traffic light intersection, and annotate the road test data to obtain annotated data; Based on the labeled data, a simulation environment was constructed for a traffic light intersection where a traffic light recognition anomaly occurred, and a driving simulation test was conducted to obtain the simulation results. Based on the simulation results, the driving behavior of the autonomous vehicle at the traffic light intersection is evaluated to obtain the evaluation result of the driving behavior; The evaluation of the autonomous vehicle's driving behavior at the traffic light intersection based on the simulation results, to obtain the evaluation result of the driving behavior, includes: Based on the simulation results, the driving pass rate corresponding to the abnormal driving behavior of the autonomous vehicle at the traffic light intersection is determined; Determine the weights corresponding to the abnormal driving behaviors; Based on the driving pass rate and the weight, the driving behavior of the autonomous vehicle is evaluated to obtain the evaluation result of the driving behavior; The method further includes: collecting problem data of autonomous driving systems corresponding to autonomous vehicles at traffic light intersections, establishing a dataset, and analyzing and predicting expected vehicle behavior through preset annotations. The vehicle behavior is divided into a passing state and an intersection waiting state. The driving state of the autonomous vehicle is evaluated based on the vehicle behavior. The driving state includes whether the current driving state of the autonomous vehicle is reasonable and whether any traffic rule violations have occurred.

2. The vehicle driving behavior assessment method according to claim 1, characterized in that, The process involves acquiring road test data when an autonomous vehicle experiences a traffic light recognition anomaly at a traffic light intersection, and then labeling the road test data to obtain labeled data, including: Acquire multi-frame environmental image data collected by autonomous vehicles when they pass through road intersections; Based on preset scene labels, scene recognition is performed on the multi-frame environmental image data to determine whether the target scene corresponding to the multi-frame environmental image data is a traffic light intersection recognition anomaly. If so, then the driving data of the autonomous vehicle in the target scenario will be collected; The driving data is saved to obtain road test data, and the road test data is labeled based on the scene tags to obtain labeled data.

3. The vehicle driving behavior assessment method according to claim 2, characterized in that, The step of labeling the road test data based on the scene labels to obtain labeled data includes: Based on the scene labels, the road test data is classified into multiple road test scenes corresponding to the road test data. The road test scenarios in the road test data are labeled using the corresponding scene tags to obtain labeled data.

4. The vehicle driving behavior assessment method according to claim 3, characterized in that, The simulation environment, which constructs a traffic light intersection based on the labeled data and simulates a traffic light recognition anomaly, is then used for driving simulation testing to obtain simulation results, including: Based on the scene annotations, extract the corresponding scene fragment data from the road test data; Based on the scene fragment data, the road test environment of the autonomous vehicle is constructed to obtain the simulation environment; Driving tests were conducted based on the simulation environment, and simulation results were obtained.

5. The vehicle driving behavior assessment method according to claim 4, characterized in that, The construction of the road test environment for the autonomous vehicle based on the scene fragment data, to obtain the simulation environment, includes: Extract the semantic data of the autonomous vehicle corresponding to the current scene and the trajectory status data of surrounding vehicles from the scene fragment data; Based on the dynamic elements in the scene tags, the first data corresponding to the dynamic elements is extracted from the semantic data of the current scene to obtain dynamic scene data; Based on the static elements in the scene label, the second data corresponding to the static elements is extracted from the trajectory status data of the surrounding vehicles to obtain real-time background data; The dynamic scene data and the real-time background data are spliced ​​together to construct a simulation environment of a traffic light intersection where a traffic light recognition anomaly occurs.

6. The vehicle driving behavior assessment method according to claim 4, characterized in that, The driving test based on the simulation environment, and the resulting simulation results, include: Obtain the road data corresponding to the scene segment data; Based on the road data, multiple target simulation scenarios are determined in the simulation environment; The autonomous vehicle is controlled to drive in multiple target simulation scenarios to obtain multiple scenario comfort scores and multiple scenario continuous passage scores for the autonomous vehicle; The simulation results of the autonomous vehicle are determined based on the comfort scores and continuous passage scores of the multiple scenarios.

7. A vehicle driving behavior assessment device, characterized in that, The vehicle driving behavior assessment device includes: The annotation module is used to acquire road test data when an autonomous vehicle encounters a traffic light recognition anomaly at a traffic light intersection, and to annotate the road test data to obtain annotated data; The simulation module is used to construct a simulation environment of a traffic light intersection and a traffic light recognition anomaly based on the labeled data, and to conduct driving simulation tests to obtain simulation results. An evaluation module is used to evaluate the driving behavior of the autonomous vehicle at the traffic light intersection based on the simulation results, and obtain the evaluation result of the driving behavior; The evaluation module is specifically used to: determine the driving pass rate corresponding to the abnormal driving behavior of the autonomous vehicle at the traffic light intersection based on the simulation results; determine the weight corresponding to the abnormal driving behavior; and evaluate the driving behavior of the autonomous vehicle based on the driving pass rate and the weight to obtain the evaluation result of the driving behavior. The device is also used to: collect problem data of autonomous driving systems corresponding to autonomous vehicles at traffic light intersections, establish a dataset, analyze and predict expected vehicle behavior through preset labeling, wherein the vehicle behavior is divided into a passing state and an intersection waiting state, and use the vehicle behavior as a standard to evaluate the driving state of the autonomous vehicle, wherein the driving state includes whether the current driving state of the autonomous vehicle is reasonable and whether it has violated traffic rules.

8. A vehicle driving behavior assessment device, characterized in that, The vehicle driving behavior assessment device includes: a memory and at least one processor, wherein the memory stores instructions, and the memory and the at least one processor are interconnected via a line; The at least one processor invokes the instructions in the memory to cause the vehicle driving behavior assessment device to perform the various steps of the vehicle driving behavior assessment method as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by the processor, it implements the various steps of the vehicle driving behavior evaluation method as described in any one of claims 1-6.

Images