Driving scene closed-loop test method and device based on three-dimensional reconstruction technology, equipment and medium

Abstract

The invention discloses a driving scene closed-loop test method and device based on a three-dimensional reconstruction technology, equipment and a medium, and relates to the technical field of vehicle simulation testing, and the method comprises the steps: obtaining a test case, extracting scene related information according to the test case, obtaining road scene data, carrying out the construction based on the road scene data, and obtaining a driving scene closed-loop test result. The method comprises the steps of obtaining a reconstructed driving scene, editing the reconstructed driving scene according to a test case to obtain generalized road scene data, and performing closed-loop test on the generalized road scene data to obtain a test result. The scene information is extracted through the test case, the driving scene is constructed, edited and reconstructed, closed-loop test is carried out after road scene data is generalized, and finally the test result is obtained, so that the accuracy and integrity of the simulation driving scene are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle simulation testing, and in particular to a driving scene closed-loop testing method, device, equipment and medium based on three-dimensional reconstruction technology. Background Art

[0002] In recent years, end-to-end intelligent driving technology has gradually become the mainstream solution in the field of intelligent driving. Compared with traditional rule-based intelligent driving solutions, end-to-end intelligent driving solutions have the characteristics of strong generalization ability, simplified system architecture, and excellent robustness. The end-to-end intelligent driving algorithm directly goes from sensor input to planning result output or control result output through a unified neural network, but due to its higher degree of opacity, it puts higher requirements on simulation testing.

[0003] However, the existing driving simulation is to initialize the driving simulation processing of the vehicle model in a preset simulation scenario to generate vehicle information, wherein the vehicle information represents the driving information of the vehicle model in the preset simulation scenario; perform data conversion processing on the vehicle information to generate message data corresponding to the vehicle information; perform logical calculation processing on the message data, output control instructions, and simulate motion control of the vehicle model based on the control instructions.

[0004] Existing solutions use traditional simulation scene software to build simulation scenes. The generated scenes are quite different from real scenes and cannot meet the simulation test requirements of end-to-end intelligent driving algorithms. In addition, end-to-end intelligent driving algorithms require a large number of rich simulation test scenes, which cannot be met by existing methods. Therefore, there is an urgent need for a closed-loop testing method for driving scenes based on 3D reconstruction technology to improve the accuracy and completeness of simulated driving scenes. Summary of the invention

[0005] The main purpose of this application is to provide a closed-loop testing method, device, equipment and medium for driving scenes based on three-dimensional reconstruction technology, aiming to solve the technical problem of how to improve the accuracy and completeness of simulated driving scenes.

[0006] To achieve the above objectives, the present application proposes a closed-loop testing method for driving scenarios based on three-dimensional reconstruction technology, the method comprising:

[0007] Get test cases;

[0008] Extracting scene-related information according to the test case to obtain road scene data;

[0009] Constructing based on the road scene data to obtain a reconstructed driving scene;

[0010] Editing the reconstructed driving scene according to the test case to obtain generalized road scene data;

[0011] A closed-loop test is performed on the generalized road scene data to obtain a test result.

[0012] In one embodiment, the step of extracting scene-related information according to the test case to obtain road scene data includes:

[0013] Extracting scene-related information according to the test case to obtain initial road scene data;

[0014] Segmenting the initial road scene data into a plurality of data segments in chronological order;

[0015] Analyze the multiple data segments to determine features corresponding to the multiple data segments;

[0016] Corresponding features are added to the plurality of data segments to obtain road scene data, wherein the features include date, time, weather, location, driving action and road condition.

[0017] In one embodiment, the step of constructing based on the road scene data to obtain a reconstructed driving scene includes:

[0018] Analyzing the road scene data by a preset deep learning model to obtain geometric shape and texture information;

[0019] generating a three-dimensional radiation field according to the geometric shape and the texture information;

[0020] The three-dimensional radiation field is rendered using an image according to a three-dimensional reconstruction technique to obtain a two-dimensional image;

[0021] Comparing the image in the road scene data with the two-dimensional image to obtain a comparison result;

[0022] Based on the comparison result, a reconstructed driving scene of any viewing angle is generated by presetting a three-dimensional reconstruction model.

[0023] In one embodiment, the step of editing the reconstructed driving scene according to the test case to obtain generalized road scene data includes:

[0024] Get test requirements based on test cases;

[0025] Analyze the test case according to a preset large language model to obtain a dense language description;

[0026] Retrieve according to the dense language description to obtain similar scenes;

[0027] Separating the dynamic foreground from the static background of the similar scene by using three-dimensional reconstruction technology to obtain the dynamic background part;

[0028] Build a scene asset library;

[0029] According to the test requirements, the scene asset library is inserted into the dynamic background part for editing to obtain generalized road scene data.

[0030] In one embodiment, the step of performing a closed-loop test on the generalized road scene data to obtain a test result includes:

[0031] The generalized road scene data is simulated and tested through an intelligent driving algorithm to obtain data change records and performance indicators;

[0032] According to the data change record and the performance index, determining whether the generalized road scene data meets the preset simulation requirements;

[0033] When the generalized road scene data meets the preset simulation requirements, a test result is generated.

[0034] In one embodiment, before the step of subjecting the generalized road scene data to a simulation test through an intelligent driving algorithm to obtain data change records and performance indicators, the step includes:

[0035] Evaluate the intelligent driving algorithm and obtain the evaluation results;

[0036] When the evaluation result shows that the intelligent driving algorithm does not meet the preset performance requirements, adjusting the parameters or structure of the intelligent driving algorithm for optimization until the intelligent driving algorithm meets the preset performance requirements;

[0037] When the intelligent driving algorithm meets the preset performance requirements, the step of performing simulation testing on the generalized road scene data through the intelligent driving algorithm to obtain data change records and performance indicators is continued.

[0038] In one embodiment, after the step of performing a closed-loop test on the generalized road scene data to obtain a test result, the following steps are included:

[0039] Obtain generalized road scene data that failed the test;

[0040] Modifying the weather, day and night conditions, traffic participants and behaviors of the traffic participants in the road scene data to obtain updated road scene data;

[0041] The updated road scene data is tested.

[0042] In addition, to achieve the above purpose, the present application also proposes a driving scene closed-loop test device based on three-dimensional reconstruction technology, and the driving scene closed-loop test device based on three-dimensional reconstruction technology includes:

[0043] Acquisition module, used to obtain test cases;

[0044] An analysis module is used to extract scene-related information according to the test case to obtain road scene data;

[0045] A construction module, used to construct based on the road scene data to obtain a reconstructed driving scene;

[0046] A processing module, used for editing the reconstructed driving scene according to the test case to obtain generalized road scene data;

[0047] The result module is used to perform a closed-loop test on the generalized road scene data to obtain a test result.

[0048] In addition, to achieve the above-mentioned purpose, the present application also proposes a medium, which is a computer-readable medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the driving scene closed-loop test method based on three-dimensional reconstruction technology as described above are implemented.

[0049] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the driving scene closed-loop testing method based on three-dimensional reconstruction technology as described above.

[0050] This application obtains test cases, extracts scene-related information according to the test cases, obtains road scene data, constructs based on the road scene data, obtains reconstructed driving scenes, edits the reconstructed driving scenes according to the test cases, obtains generalized road scene data, performs closed-loop testing on the generalized road scene data, and obtains test results. By extracting scene information through test cases, constructing and editing reconstructed driving scenes, performing closed-loop testing after generalizing road scene data, and finally obtaining test results, the accuracy and completeness of the simulated driving scenes are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0052] Figure 1This is a flowchart of the first embodiment of the closed-loop testing method for driving scenes based on three-dimensional reconstruction technology of the present application;

[0053] Figure 2 This is a NeRF principle diagram of the first embodiment of the driving scene closed-loop testing method based on three-dimensional reconstruction technology of the present application;

[0054] Figure 3 This is a flow chart of a second embodiment of a closed-loop testing method for a driving scene based on three-dimensional reconstruction technology of the present application;

[0055] Figure 4 This is a flowchart of the third embodiment of the closed-loop testing method for driving scenes based on three-dimensional reconstruction technology of the present application;

[0056] Figure 5 This is a schematic diagram of the module structure of a driving scene closed-loop test device based on three-dimensional reconstruction technology in an embodiment of the present application;

[0057] Figure 6 This is a schematic diagram of the device structure of the hardware operating environment involved in the closed-loop testing method for driving scenarios based on three-dimensional reconstruction technology in an embodiment of the present application.

[0058] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0059] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0060] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0061] The main solution of the embodiment of the present application is: by obtaining test cases, extracting scene-related information according to the test cases to obtain road scene data, constructing based on the road scene data to obtain a reconstructed driving scene, editing the reconstructed driving scene according to the test cases to obtain generalized road scene data, performing closed-loop testing on the generalized road scene data, and obtaining test results.

[0062] Based on this, the embodiment of the present application provides a closed-loop testing method for driving scenes based on three-dimensional reconstruction technology. Figure 1 , Figure 1 This is a flow chart of the first embodiment of the closed-loop testing method for driving scenarios based on three-dimensional reconstruction technology of the present application.

[0063] In this embodiment, the driving scene closed-loop test method based on 3D reconstruction technology includes steps S10 to S50:

[0064] Step S10, obtaining a test case.

[0065] It should be noted that test cases usually include detailed descriptions of specific driving scenarios, such as traffic conditions, weather conditions, road types, etc.

[0066] Step S20, extracting scene-related information according to the test case to obtain road scene data.

[0067] It should be noted that extracting scene-related information based on test cases is a key step in intelligent driving simulation testing, which involves extracting key information from test cases, such as traffic conditions, weather conditions, road types, etc., to build an accurate simulated road scene. This process requires careful analysis of the use cases to ensure that all relevant details are captured and used for scene construction. The extracted data will be used to generate simulated road scenes, including elements such as vehicles, pedestrians, traffic signs, and signal lights. In order to enhance the realism and complexity of the scene, it can be further expanded, for example, by simulating different lighting conditions, weather changes (such as rain, snow, fog), and different time periods (such as morning and evening rush hours, nighttime). In addition, the degree of road wear, slippery road conditions, and temporary traffic control measures can also be considered.

[0068] Step S30: constructing based on the road scene data to obtain a reconstructed driving scene.

[0069] It should be understood that, first, the road scene data is analyzed by a preset deep learning model to obtain geometric shape and texture information. Specifically, using the object detection algorithm, important traffic elements such as vehicles, pedestrians, traffic signs, and road conditions can be identified and located from the image or video stream of the road scene. Then, based on these marked objects, convolutional neural networks (CNNs) or other feature extraction methods are further used to capture the geometric shape of the object and its surface characteristics, including color, texture, and other visual attributes.

[0070] Further, according to the geometric shape and the texture information, a three-dimensional radiation field is generated, and the three-dimensional radiation field is rendered using an image according to a three-dimensional reconstruction technique to obtain a two-dimensional image. Specifically, according to the geometric shape and the texture information, generating a three-dimensional radiation field involves using a Neural Radiance Fields (NeRF) technique. NeRF is capable of learning continuous three-dimensional scene representations and allowing the synthesis of high-quality new views from any perspective. Specifically, given a set of multi-view images and corresponding camera pose parameters as input, NeRF trains a multi-layer perceptron (MLP) that can predict the color value and volume density of each spatial position, thereby constructing a complete three-dimensional radiation field. This process not only takes into account the appearance of static objects, but also can handle dynamically changing factors, such as changes in lighting conditions or the influence of moving objects. In order to ensure that the generated three-dimensional radiation field is accurate enough, a large number of training samples and efficient optimization strategies are usually required to minimize the difference between the rendered image and the original observation. Next, according to the three-dimensional reconstruction technique, the three-dimensional radiation field is rendered using an image to obtain a two-dimensional image. This step relies on the volume rendering equation, that is, the color and density at the sampling point are integrated along the ray direction to simulate the attenuation effect during light propagation. In practice, in order to speed up rendering without sacrificing quality, researchers have proposed a variety of improvements, such as using a layered sampling scheme or combining feature grids to guide the distribution of NeRF point clouds. In addition, additional constraints, such as smoothness priors or physical consistency assumptions, can be introduced to improve the realism of the final output image.

[0071] Furthermore, the image in the road scene data is compared with the two-dimensional image to obtain a comparison result, and then based on the comparison result, a preset three-dimensional reconstruction model is used to generate a reconstructed driving scene of any perspective. Figure 2 The schematic diagram of NeRF principle shown in the figure, specifically, the image in the road scene data is compared with the two-dimensional image to obtain a comparison result, which is intended to evaluate whether the generated three-dimensional radiation field faithfully reproduces the original scene. To this end, a series of quantitative indicators, such as peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), etc., can be designed to measure the similarity between the two. At the same time, the subjective evaluation of human observers is used to supplement the shortcomings of objective measurement. Once the existing deviation is determined, the model parameters can be adjusted and retrained until a satisfactory matching effect is achieved. On this basis, based on the comparison results, a reconstructed driving scene of any perspective is generated by presetting a three-dimensional reconstruction model. This means that users can choose different perspectives to view the entire scene according to their needs, whether it is a bird's-eye view or a first-person perspective, they can get a highly realistic experience.

[0072] Step S40: Edit the reconstructed driving scene according to the test case to obtain generalized road scene data.

[0073] It should be noted that, first, the test requirements are obtained according to the test cases, and then the test cases are analyzed according to the preset large language model to obtain dense language descriptions, and then the dense language descriptions are retrieved to obtain similar scenarios. Specifically, the process of obtaining test requirements according to the test cases is the basis for ensuring that the software or system meets the expected functional and performance requirements. This process usually includes understanding business requirements, user requirements, and specific functional requirements to determine the direction and content of the test. For complex systems such as autonomous driving, this means analyzing the behavior of the vehicle in different driving scenarios in detail to ensure that it meets safety standards and technical specifications. In order to ensure the comprehensiveness and accuracy of the test, the tester also needs to identify the various modules of the system and their interaction methods, and clarify which parts need special attention. Once the test requirements are clarified, the next step is to use the preset large language model (LLM) to analyze the test cases to obtain dense language descriptions. The large language model can automatically generate detailed test case descriptions based on a small number of prompt words provided. These descriptions not only cover the test steps and expected results, but may also include boundary conditions and other special considerations. For example, when dealing with a test case involving the response speed of intelligent traffic lights, LLM can generate a detailed text describing how to set up the test environment, the specific process of executing the test, and what the expected results are. Subsequently, based on the dense language description obtained, similar test scenarios in history are found through retrieval technology. This step can be achieved by building a vector search-based system that uses embedding technology to convert text into numerical representations and then finds the closest matches in a large document database. For example, when faced with a new urban road test, similar road condition test cases in other cities that have been processed before will be retrieved. This practice not only helps save time, but also draws on past lessons learned to improve the effectiveness and efficiency of new tests.

[0074] Furthermore, the dynamic foreground and static background are separated for similar scenes using 3D reconstruction technology to obtain the dynamic background part, and then the scene asset library is constructed. According to the test requirements, the scene asset library is inserted into the dynamic background part for editing to obtain the generalized road scene data. Specifically, the separation of dynamic foreground and static background for similar scenes using 3D reconstruction technology depends on 3D reconstruction algorithms and technologies, such as methods based on RGB-D cameras or neural rendering technologies such as NeRF. Through these methods, accurate depth information can be extracted from multi-view images, and fixed objects in the scene such as buildings, road facilities and other static elements, as well as pedestrians, vehicles and other objects that may move can be distinguished. In order to ensure the accuracy of the separation results, semantic segmentation and instance segmentation techniques are usually combined to identify and mark different types of objects. In addition, the TSDF (truncated signed distance function) model is also widely used in such tasks. It can effectively represent the surface of objects and help accurately define which parts belong to the foreground and which belong to the background. The purpose of building a scene asset library is to accumulate a series of various elements that can be used for simulation, including but not limited to different weather conditions, time periods, traffic flow, lighting changes, special events, etc. This library not only contains predefined 3D models, but also covers various material maps, lighting effects and physical property settings, aiming to provide a comprehensive and flexible foundation to support the creation of diverse and realistic test scenarios. For example, when dealing with the performance of night driving, appropriate lighting effects can be selected from the asset library and applied to the current scene; when simulating driving in bad weather conditions, the corresponding rain, snow and fog effects can be called. It is worth noting that with the development of AI generated content (AIGC) technology, new assets can now be automatically generated with the help of generative adversarial networks (GANs), style transfer and other means, further enriching the available resources. Finally, inserting the scene asset library into the dynamic background part for editing according to the test requirements means customizing the scene configuration for specific test goals. This involves a series of operations such as selecting a suitable background template, adding necessary foreground objects, and setting reasonable interaction rules. For example, when designing a lane change test on a highway, not only the lane change operation under normal circumstances should be considered, but also abnormal situations such as emergency braking and vehicles that suddenly cut in should be considered. In addition, random factors, such as the sudden appearance of pedestrians or animals, can also be introduced to test the ability of the autonomous driving system to cope with emergencies. For urban street scenes, in addition to the basic road layout, we also need to pay attention to the characteristics of crowded areas such as traffic lights, zebra crossings, and bus stops to ensure that the test is as close to the actual situation as possible. On this basis, NeRF technology can also be used to generate object models with highly realistic physical properties and complex geometric structures to enhance the realism of the scene. The generalized road scene data finally obtained not only reflects the complexity of the real world, but also maintains a certain degree of controllability, which is convenient for subsequent analysis and evaluation.

[0075] Step S50, performing a closed-loop test on the generalized road scene data to obtain a test result.

[0076] It should be noted that in the process of simulating and testing the generalized road scene data through the intelligent driving algorithm, it is first necessary to ensure that the scene data has been accurately converted into a format suitable for the simulation platform and can reflect various traffic conditions and environmental changes in the real world. In the simulation environment, the autonomous driving algorithm will make corresponding decisions based on the input scene data, such as path planning, speed adjustment, obstacle avoidance and other operations. As the test progresses, the data change records of vehicle behavior are continuously collected, including but not limited to position information, speed curve, acceleration change, steering angle, and relative distance to other traffic participants. Once sufficient test data is obtained, the next step is to evaluate whether these data change records and performance indicators meet the requirements based on the preset standards. This step involves statistical analysis of a large number of data points to determine whether they fall within the predefined range. If all key indicators meet the preset simulation requirements, it is considered that the set of generalized road scene data has successfully passed the simulation test. At this time, a test report is generated to record the phenomena observed and the results obtained during the entire test process.

[0077] Further, after step S50, it also includes: obtaining the generalized road scene data that failed the test, modifying the weather, day and night conditions, traffic participants and traffic participant behaviors of the road scene data, obtaining updated road scene data, and testing the updated road scene data. Specifically, when it is found that the simulation test under certain specific conditions fails to achieve the expected results, it is first necessary to analyze these failure cases in detail to identify the key factors that lead to the failure of the test, such as weather conditions, day and night changes, types of traffic participants and their behavior patterns, etc. In view of the above problems, the updated scene data can be created by adjusting the environmental parameters in the road scene. For example, if the rainy conditions in the original scene affect the performance of the sensor, you can try to reduce the rainfall intensity or change to sunny weather and retest; for the perception error caused by insufficient light at night, the scene can be changed to daytime or simulated street lights can be added to improve visibility. In addition, more types of traffic participants (such as pedestrians, bicycles, motorcycles) and complex behavior patterns (such as sudden lane changes, emergency braking) can be introduced to comprehensively evaluate the coping ability of the autonomous driving system. After completing the modification of the road scene data, the next step is to conduct a new round of simulation tests on the updated scene. This not only includes repeating previous test cases, but also adding some new challenging conditions to ensure that the system can operate stably under a wider range of conditions. At this stage, it is important to record detailed test logs, including but not limited to vehicle trajectories, speed changes, decision-making processes, interactions with other objects, etc., for subsequent analysis. Based on the test results obtained with the updated road scene data, the effectiveness of the algorithm improvement is further verified. If the new version of the data can make the system perform significantly better than the previous state, and all key performance indicators meet the preset standards, the adjustments made are successful. At this point, a detailed report should be prepared to summarize the problems encountered and solutions during the test, and at the same time, suggestions for future research directions should be made.

[0078] This embodiment provides a closed-loop test method for driving scenes based on three-dimensional reconstruction technology, which obtains test cases, extracts scene-related information according to the test cases, obtains road scene data, constructs based on the road scene data, obtains reconstructed driving scenes, edits the reconstructed driving scenes according to the test cases, obtains generalized road scene data, performs closed-loop testing on the generalized road scene data, and obtains test results. By extracting scene information through test cases, constructing and editing reconstructed driving scenes, performing closed-loop testing after generalizing road scene data, and finally obtaining test results, the accuracy and completeness of the simulated driving scenes are improved.

[0079] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can refer to the above introduction, and will not be repeated later. Figure 3 The driving scene closed-loop test method based on three-dimensional reconstruction technology, step S20, further includes steps S201 to S204:

[0080] Step S201, obtaining initial road scene data according to a test case.

[0081] It should be noted that when creating the initial road scene data, you first need to build a basic framework, which includes static elements such as the geometry of the road (such as straight roads, curves, slopes), lane line information (one-way / two-way, solid / dashed lines), and traffic signs (speed limit signs, traffic lights). At the same time, in order to simulate the real traffic environment, dynamic elements, namely traffic participants and their behavior patterns, need to be added. For example, different types of vehicles (cars, trucks, buses), pedestrians, cyclists, etc. are added to the simulation, and their behavior logic is set, such as regular driving, turning, stopping and waiting, suddenly crossing the road, etc., and then the initial road scene data is generated based on the above basic framework according to the predefined test cases.

[0082] Step S202: dividing the initial road scene data into multiple data segments in chronological order.

[0083] It should be noted that when dividing the data segments, it is necessary to set a reasonable time interval according to the specific test requirements. For example, for general urban road driving, every 5 seconds or 10 seconds can be selected as a segment; on the highway, a longer time span, such as 30 seconds or 1 minute, may be required to reflect the stable driving conditions over a longer distance. In addition, for some special events (such as traffic light changes, pedestrians crossing the road), a shorter time window, such as 1-2 seconds, should be used to accurately record the dynamics of these moments. Once the time interval is determined, the next step is to cut the original data stream. This step requires the use of efficient algorithms to ensure data integrity while avoiding human errors introduced by segmentation. Specifically, each data segment should contain complete sensor readings (such as camera image frames, radar echoes), vehicle status information (speed, acceleration, steering angle) and environmental parameters (weather conditions, light intensity). In order to facilitate subsequent analysis, associations should also be established between the segments to ensure that they can be correctly sorted and reproduce the entire driving process. After the data segments are divided, in-depth research can be carried out on each segment. For example, the accuracy of the perception system can be evaluated by comparing the position changes of objects in adjacent segments; or the response speed and rationality of the decision-making module in the face of emergencies can be analyzed. More importantly, this fine-grained data processing method makes it easier for researchers to find the key factors that may cause system failure and propose improvement measures accordingly.

[0084] Step S203: Analyze the multiple data segments to determine the features corresponding to the multiple data segments.

[0085] It should be noted that for each data segment, various perceptual information needs to be extracted and quantified. For example, camera images and lidar point clouds provide visual and spatial descriptions of the surrounding environment, from which the position and motion status of objects such as road boundaries, traffic signs, other vehicles and pedestrians can be identified. In addition, millimeter-wave radar signals help monitor targets at longer distances and detect objects under low visibility conditions. Combined with the vehicle's own sensor data (such as speed, acceleration, and steering angle), we can build a detailed dynamic scene model. Next, based on the above perceptual information, the behavioral characteristics of the decision module are further analyzed. This includes whether the choice of path planning is reasonable, whether the timing of lane change is appropriate, and whether the following distance is safe. In particular, in the face of complex or emergency situations, such as sudden obstacles in front or abnormal movements of nearby vehicles, it has the ability to respond quickly and take appropriate risk avoidance measures. By evaluating these decision behaviors, potential risk points can be discovered, and algorithm parameters can be adjusted or new logical rules can be introduced in a targeted manner. In addition to static and dynamic features, it is also necessary to pay attention to changes in environmental factors in each data segment. Factors such as weather conditions (sunny, rainy, snowy), lighting conditions (daytime, nighttime), and traffic flow can significantly affect the perception and decision-making efficiency of the autonomous driving system. Therefore, the role of these external variables should be fully considered during the analysis process to ensure that the system can operate stably under various conditions. For example, driving at night will reduce the effectiveness of some sensors due to insufficient light. At this time, the perception effect can be improved by adding simulated street lights or optimizing image enhancement algorithms.

[0086] Step S204, adding corresponding features to the multiple data segments to obtain road scene data, wherein the features include date, time, weather, location, driving action and road condition.

[0087] It should be noted that the date and time are crucial to understanding traffic patterns during specific time periods. For example, during the morning and evening rush hours on weekdays, urban roads usually have higher vehicle density and faster driving speeds; on weekends or holidays, different traffic flow characteristics may be presented. In addition, the difference in light between night and day can also affect the working efficiency of sensors, so special attention needs to be paid. Secondly, weather conditions are an extremely important factor. Most sensors can work normally in clear weather, while rainy, snowy or foggy days may cause the performance of vision and radar systems to deteriorate. By simulating various weather conditions, the system's adaptability to different climates can be tested and improved. For example, rain may make the road surface slippery and increase braking distance; when heavy snow covers the ground, lane line recognition becomes more difficult. In addition, location information can help us understand the infrastructure layout and traffic rules in a specific area. Road designs in different cities may vary significantly, and some areas may have special traffic management measures or speed limits. Accurately marking locations helps ensure that the test scenario is as close to the actual use environment as possible and can be optimized specifically for the challenges of specific regions. Next, driving actions refer to the specific operations taken by the vehicle at a certain moment, such as acceleration, deceleration, steering, parking, etc. These actions reflect the driver's response to changes in the surrounding environment and are also an important basis for measuring the rationality of the decision-making of the autonomous driving system. By analyzing a series of continuous actions, it is possible to evaluate whether the path planning is smooth, whether the obstacle avoidance strategy is effective, and the overall driving safety. Finally, road conditions cover many aspects from flatness to obstacle distribution. Good road conditions are conducive to improving driving comfort, but in reality, it is inevitable to encounter bad road sections such as potholes and construction areas. In this case, the autonomous driving system has sufficient robustness to deal with emergencies. For example, when an obstacle is detected ahead, a decision to slow down or detour is made in a timely manner.

[0088] Create more realistic and diverse scenarios for testing by adding detailed characterization information to multiple pieces of data.

[0089] This embodiment constructs detailed road scene data by dividing the initial road scene data by time, analyzing it, and adding features, thereby enhancing the authenticity and diversity of the test.

[0090] Based on the first embodiment of the present application, in the third embodiment of the present application, the same or similar contents as those in the first embodiment can be referred to the above description, and will not be described in detail later. Figure 4 The driving scene closed-loop test method based on three-dimensional reconstruction technology, step S50, further includes steps S501 to S503:

[0091] Step S501, the generalized road scene data is simulated and tested through an intelligent driving algorithm to obtain data change records and performance indicators.

[0092] It should be noted that after the intelligent driving algorithm under test is deployed in the simulation environment, various operations during the vehicle driving process, such as acceleration, deceleration, and steering, are simulated according to the input road scene data. At the same time, it will also record the information collected by all sensors (cameras, lidars, millimeter-wave radars, etc.) in real time, as well as the interaction details between the vehicle and the surrounding environment. For example, when encountering pedestrians suddenly crossing the road or the vehicle in front brakes suddenly, whether the correct avoidance decision can be made in time; whether the accuracy will be affected under harsh conditions such as rain or snow. As the test progresses, a large number of data change records will be generated, which reflect the state changes of the vehicle at various moments during the entire driving process. Specifically, the data change records include position information (the trajectory of the vehicle's change relative to the initial position), speed curve (the change of the vehicle's speed over time), acceleration change (behavior characteristics of vehicle acceleration and deceleration), steering angle (angle change when the vehicle turns), and the distance of other objects (such as the relative distance change with the vehicle in front, pedestrians or other obstacles). In addition, changes in the internal state of the vehicle, such as engine speed, fuel consumption and other indicators, are also included. By analyzing these data change records, a series of important performance indicators can be extracted to measure the performance of the intelligent driving algorithm. Common performance indicators include, but are not limited to: perception accuracy (target recognition rate and classification accuracy), decision rationality (smoothness of path planning, rationality of lane change timing), actuator response speed (brake reaction time, steering control accuracy), system stability (ability to run without failure for a long time), and computing resource utilization (CPU / GPU usage, memory usage and other non-functional indicators). In particular, in the face of complex or emergency situations, such as sudden obstacles in front or abnormal movements of nearby vehicles, the system should have the ability to respond quickly and take appropriate risk avoidance measures. Based on the above data analysis results, if the test results under certain specific conditions show that the algorithm has deficiencies, it is necessary to further adjust the parameters or improve the model structure to improve its adaptability and robustness. For example, under low light conditions at night, the ability to recognize objects can be improved by enhancing the image processing algorithm; for complex urban intersections, more advanced prediction models need to be introduced to optimize the decision-making process.

[0093] Furthermore, before step S501, it also includes: evaluating the intelligent driving algorithm to obtain an evaluation result; when the evaluation result is that the intelligent driving algorithm does not meet the preset performance requirements, adjusting the parameters or structure of the intelligent driving algorithm for optimization until the intelligent driving algorithm meets the preset performance requirements, and when the intelligent driving algorithm meets the preset performance requirements, continuing to perform simulation tests on the generalized road scene data through the intelligent driving algorithm to obtain data changes and performance indicators. Specifically, before performing the simulation test, an intelligent driving algorithm is constructed to determine whether the intelligent driving algorithm meets the preset performance requirements. If not, it is necessary to adjust the algorithm parameters or structure for optimization. Finally, when the algorithm finally meets the requirements, continue to use the generalized road scene data for more simulation tests to further verify its stability and reliability.

[0094] Step S502: judging whether the generalized road scene data meets the preset simulation requirements according to the data change records and the performance indicators.

[0095] It should be noted that the above process involves not only a detailed analysis of the test results, but also a comprehensive evaluation of the algorithm performance to ensure that the intelligent driving system can operate safely and reliably under various conditions. First, it is necessary to compare the differences between the actual data change records and the preset simulation requirements. For example, in terms of perception accuracy, if the target recognition rate or classification accuracy fails to meet the predetermined standards, it indicates that the current road scene data may not be sufficient to fully cover all potential situations. At this time, it should be checked whether there are certain types of objects or environmental features that are not fully considered, which in turn affects the detection ability of the system. For the rationality of decision-making, the smoothness of path planning and the rationality of lane change timing are important considerations. If it is found that the performance of the vehicle in complex intersections or high-density traffic flows is not as expected, it means that the logical scenario parameters need to be adjusted or new training samples need to be introduced to better simulate real-world driving scenarios. Secondly, the response speed of the actuator is also a key point. When encountering an emergency, such as an obstacle suddenly appearing in front or an unexpected action by a nearby vehicle, measures should be taken quickly and accurately. By analyzing specific values ​​such as braking reaction time and steering control accuracy, it is possible to evaluate whether the vehicle control system is sensitive enough and adjust the relevant parameter settings accordingly. In addition, we need to pay attention to the stability of the system under long-term operation, that is, whether it can maintain consistent performance without failure during continuous work. Any abnormal fluctuations may be a signal of potential problems, prompting developers to further investigate and solve these problems.

[0096] Furthermore, in order to ensure the effectiveness of simulation testing, the scenario library used must cover sufficient diversity. This means not only including cases under normal driving conditions, but also paying special attention to those rare but dangerous situations, such as extreme weather events, traffic accident scenes, etc. At the same time, with the advancement of technology and changes in application scenarios, it is also necessary to regularly update and improve the scenario library. This not only keeps up with the latest industry developments, but also provides a solid foundation for future innovations.

[0097] Furthermore, from the perspective of accident data drive, using the existing big data of road traffic accidents to analyze and extract characteristic elements suitable for autonomous driving tests can effectively improve the authenticity and coverage of test scenarios. For example, combined with the combined test method, a set of test cases for highway two-lane lane change logic scenarios is generated based on the data of real road traffic accidents that occurred in my country in the NAIS database. This method can not only increase the number of test scenarios, but also ensure that each scenario has a high application value, thereby better serving the safety testing and evaluation of autonomous driving.

[0098] Step S503: When the generalized road scene data meets the preset simulation requirements, a test result is generated.

[0099] It should be noted that the test result report usually includes multiple aspects: the overview briefly introduces the purpose of the test, the type of scenario library used and its characteristics, and the main technologies and tools used; the performance indicator section lists all preset performance standards and compares the data change records obtained from the actual test; the case analysis selects several representative test examples for detailed analysis, explaining the system's operating procedures and effect evaluation in these scenarios; the problems and suggestions summarize all the problems found in the test and give corresponding improvement suggestions; the conclusion is based on the above content to draw a final evaluation of the current version of the intelligent driving system, clearly pointing out its advantages, while not avoiding the existing limitations. For some functional modules that have successfully passed the test, it can be considered to gradually advance to the field road test stage in order to obtain more feedback information in a real environment.

[0100] This embodiment uses the generalized road scene data to perform simulation tests through the intelligent driving algorithm, analyzes the obtained data changes and performance indicators to determine whether they meet the preset requirements. If they do, the test results are generated, which improves the adaptability and reliability of the test.

[0101] This application also provides a driving scene closed-loop test device based on 3D reconstruction technology, please refer to Figure 5 , the device comprises:

[0102] The acquisition module 10 is used to acquire test cases.

[0103] The analysis module 20 is used to extract scene-related information according to the test case to obtain road scene data.

[0104] The construction module 30 is used to construct based on the road scene data to obtain a reconstructed driving scene.

[0105] The processing module 40 is used to edit the reconstructed driving scene according to the test case to obtain generalized road scene data.

[0106] The result module 50 is used to perform a closed-loop test on the generalized road scene data to obtain a test result.

[0107] The driving scene closed-loop test device based on three-dimensional reconstruction technology provided by the present application adopts the driving scene closed-loop test method based on three-dimensional reconstruction technology in the above-mentioned embodiment, which can solve the technical problem of how to improve the accuracy and completeness of the simulated driving scene. Compared with the prior art, the beneficial effects of the driving scene closed-loop test device based on three-dimensional reconstruction technology provided by the present application are the same as the beneficial effects of the driving scene closed-loop test method based on three-dimensional reconstruction technology provided by the above-mentioned embodiment, and the other technical features in the driving scene closed-loop test device based on three-dimensional reconstruction technology are the same as the features disclosed in the above-mentioned embodiment method, which will not be repeated here.

[0108] In one embodiment, the analysis module 20 is also used to obtain initial road scene data according to the test case; divide the initial road scene data into multiple data segments in chronological order; analyze the multiple data segments to determine the features corresponding to the multiple data segments; add corresponding features to the multiple data segments to obtain road scene data, and the features include date, time, weather, location, driving action and road conditions.

[0109] In one embodiment, the construction module 30 is also used to analyze the road scene data through a preset deep learning model to obtain geometric shape and texture information; generate a three-dimensional radiation field based on the geometric shape and texture information; render the three-dimensional radiation field using an image according to the three-dimensional reconstruction technology to obtain a two-dimensional image; compare the image in the road scene data with the two-dimensional image to obtain a comparison result; based on the comparison result, generate a reconstructed driving scene from any perspective through a preset three-dimensional reconstruction model.

[0110] In one embodiment, the processing module 40 is also used to obtain test requirements based on test cases; analyze the test cases according to a preset large language model to obtain a dense language description; retrieve according to the dense language description to obtain similar scenes; use three-dimensional reconstruction technology to separate the dynamic foreground and static background of similar scenes to obtain a dynamic background part; build a scene asset library; insert the scene asset library into the dynamic background part for editing according to the test requirements to obtain generalized road scene data.

[0111] In one embodiment, the result module 50 is also used to perform simulation testing on the generalized road scene data through an intelligent driving algorithm to obtain data changes and performance indicators; based on the data changes and performance indicators, determine whether the generalized road scene data meets the preset simulation requirements; when the generalized road scene data meets the preset simulation requirements, generate a test result.

[0112] In one embodiment, the result module 50 is also used to evaluate the intelligent driving algorithm to obtain an evaluation result; when the evaluation result is that the intelligent driving algorithm does not meet the preset performance requirements, the parameters or structure of the intelligent driving algorithm are adjusted for optimization until the intelligent driving algorithm meets the preset performance requirements; when the intelligent driving algorithm meets the preset performance requirements, the generalized road scene data is simulated and tested through the intelligent driving algorithm to obtain data changes and performance indicators.

[0113] In one embodiment, the result module 50 is also used to obtain generalized road scene data that fails the test; modify the weather, day and night conditions, traffic participants and traffic participant behaviors of the road scene data to obtain updated road scene data; and test the updated road scene data.

[0114] The present application provides a driving scene closed-loop test device based on three-dimensional reconstruction technology, and the driving scene closed-loop test device based on three-dimensional reconstruction technology includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the driving scene closed-loop test method based on three-dimensional reconstruction technology in the above-mentioned embodiment one.

[0115] Reference below Figure 6 , which shows a schematic diagram of the structure of a driving scene closed-loop test device based on three-dimensional reconstruction technology suitable for implementing the embodiment of the present application. The driving scene closed-loop test device based on three-dimensional reconstruction technology in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 6The driving scene closed-loop test equipment based on three-dimensional reconstruction technology shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0116] like Figure 6 As shown, the driving scene closed-loop test equipment based on three-dimensional reconstruction technology may include a processing device 1001 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM: Read Only Memory) 1002 or the program loaded from the storage device 1003 to the random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the driving scene closed-loop test equipment based on three-dimensional reconstruction technology are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other via a bus 1005. The input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow the driving scene closed-loop test device based on the three-dimensional reconstruction technology to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a driving scene closed-loop test device based on the three-dimensional reconstruction technology with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or provided alternatively.

[0117] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0118] The driving scene closed-loop test equipment based on three-dimensional reconstruction technology provided by the present application adopts the driving scene closed-loop test method based on three-dimensional reconstruction technology in the above-mentioned embodiment, which can solve the technical problem of how to improve the accuracy and completeness of the simulated driving scene. Compared with the prior art, the beneficial effects of the driving scene closed-loop test equipment based on three-dimensional reconstruction technology provided by the present application are the same as the beneficial effects of the driving scene closed-loop test method based on three-dimensional reconstruction technology provided by the above-mentioned embodiment, and the other technical features in the driving scene closed-loop test equipment based on three-dimensional reconstruction technology are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0119] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0120] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0121] The present application provides a computer-readable medium having computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the driving scene closed-loop testing method based on three-dimensional reconstruction technology in the above-mentioned embodiment.

[0122] The computer-readable medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0123] The above-mentioned computer-readable medium may be included in the driving scene closed-loop test device based on the three-dimensional reconstruction technology; or it may exist independently without being assembled into the driving scene closed-loop test device based on the three-dimensional reconstruction technology.

[0124] The computer-readable medium carries one or more programs. When the one or more programs are executed by the driving scene closed-loop test device based on the three-dimensional reconstruction technology, the driving scene closed-loop test device based on the three-dimensional reconstruction technology can be written in one or more programming languages ​​or a combination thereof to perform the computer program code of the present application. The programming language includes an object-oriented programming language, such as Java, Smalltalk, C++, and a conventional procedural programming language, such as "C" language or a similar programming language. The program code can be executed completely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0125] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0126] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0127] The readable medium provided in the present application is a computer-readable medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned closed-loop test method for driving scenes based on three-dimensional reconstruction technology, and can solve the technical problem of how to improve the accuracy and completeness of simulated driving scenes. Compared with the prior art, the beneficial effects of the computer-readable medium provided in the present application are the same as the beneficial effects of the closed-loop test method for driving scenes based on three-dimensional reconstruction technology provided in the above-mentioned embodiments, and will not be elaborated here.

[0128] The present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned driving scene closed-loop testing method based on three-dimensional reconstruction technology.

[0129] The computer program product provided by the present application can solve the technical problem of how to improve the accuracy and completeness of the simulated driving scene. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the driving scene closed-loop test method based on three-dimensional reconstruction technology provided by the above embodiment, and will not be repeated here.

[0130] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A closed-loop testing method for driving scenarios based on three-dimensional reconstruction technology, characterized in that: The method comprises: Get test cases; Extracting scene-related information according to the test case to obtain road scene data; Constructing based on the road scene data to obtain a reconstructed driving scene; Editing the reconstructed driving scene according to the test case to obtain generalized road scene data; A closed-loop test is performed on the generalized road scene data to obtain a test result.

2. The method according to claim 1, characterized in that The step of extracting scene-related information according to the test case to obtain road scene data includes: Extracting scene-related information according to the test case to obtain initial road scene data; Segmenting the initial road scene data into a plurality of data segments in chronological order; Analyze the multiple data segments to determine features corresponding to the multiple data segments; Corresponding features are added to the plurality of data segments to obtain road scene data, wherein the features include date, time, weather, location, driving action and road condition.

3. The method according to claim 1, characterized in that The step of constructing based on the road scene data to obtain a reconstructed driving scene includes: Analyzing the road scene data by a preset deep learning model to obtain geometric shape and texture information; generating a three-dimensional radiation field according to the geometric shape and the texture information; The three-dimensional radiation field is rendered using an image according to a three-dimensional reconstruction technique to obtain a two-dimensional image; Comparing the image in the road scene data with the two-dimensional image to obtain a comparison result; Based on the comparison result, a reconstructed driving scene of any viewing angle is generated by presetting a three-dimensional reconstruction model.

4. The method according to claim 1, characterized in that The step of editing the reconstructed driving scene according to the test case to obtain generalized road scene data comprises: Get test requirements based on test cases; Analyze the test case according to a preset large language model to obtain a dense language description; Retrieve according to the dense language description to obtain similar scenes; Separating the dynamic foreground from the static background of the similar scene by using three-dimensional reconstruction technology to obtain the dynamic background part; Build a scene asset library; According to the test requirements, the scene asset library is inserted into the dynamic background part for editing to obtain generalized road scene data.

5. The method according to claim 1, characterized in that The step of performing a closed-loop test on the generalized road scene data to obtain a test result comprises: The generalized road scene data is simulated and tested through an intelligent driving algorithm to obtain data change records and performance indicators; According to the data change record and the performance index, determining whether the generalized road scene data meets the preset simulation requirements; When the generalized road scene data meets the preset simulation requirements, a test result is generated.

6. The method according to claim 5, characterized in that Before the step of performing simulation test on the generalized road scene data through the intelligent driving algorithm to obtain data change records and performance indicators, the method includes: Evaluate the intelligent driving algorithm and obtain the evaluation results; When the evaluation result shows that the intelligent driving algorithm does not meet the preset performance requirements, adjusting the parameters or structure of the intelligent driving algorithm for optimization until the intelligent driving algorithm meets the preset performance requirements; When the intelligent driving algorithm meets the preset performance requirements, the step of performing simulation testing on the generalized road scene data through the intelligent driving algorithm to obtain data change records and performance indicators is continued.

7. The method according to claim 1, characterized in that After the step of performing a closed-loop test on the generalized road scene data to obtain a test result, the method further comprises: Obtain generalized road scene data that failed the test; Modifying the weather, day and night conditions, traffic participants and behaviors of the traffic participants in the road scene data to obtain updated road scene data; The updated road scene data is tested.

8. A driving scene closed-loop test device based on three-dimensional reconstruction technology, characterized in that: The device comprises: Acquisition module, used to obtain test cases; An analysis module is used to extract scene-related information according to the test case to obtain road scene data; A construction module, used to construct based on the road scene data to obtain a reconstructed driving scene; A processing module, used for editing the reconstructed driving scene according to the test case to obtain generalized road scene data; The result module is used to perform a closed-loop test on the generalized road scene data to obtain a test result.

9. A driving scene closed-loop test device based on three-dimensional reconstruction technology, characterized in that: The device includes: a memory, a processor, and a driving scene closed-loop test program based on three-dimensional reconstruction technology stored in the memory and running on the processor, wherein the driving scene closed-loop test program based on three-dimensional reconstruction technology is configured to implement the steps of the driving scene closed-loop test method based on three-dimensional reconstruction technology as described in any one of claims 1-7.

10. A storage medium, characterized in that: The storage medium stores a driving scene closed-loop test program based on three-dimensional reconstruction technology. When the driving scene closed-loop test program based on three-dimensional reconstruction technology is executed by the processor, the steps of the driving scene closed-loop test method based on three-dimensional reconstruction technology as described in any one of claims 1-7 are implemented.

Images