Automatic driving model data processing method and device based on vehicle and road cloud integrated architecture, automatic driving model self-evolution learning system and equipment and medium

Abstract

The embodiment of the invention discloses an automatic driving model data processing method and device based on a vehicle-road cloud integrated architecture, an automatic driving model self-evolution learning system, equipment and a medium, and the data processing method comprises the steps: receiving road side data and vehicle side sensing data; extracting high-value scene fragment data from the roadside data; reconstructing the 3D static road scene based on the vehicle end sensing data, and identifying and extracting a plurality of dynamic traffic participants from the vehicle end data; combining the plurality of dynamic traffic participants, the reconstructed 3D static road scene and the high-value scene fragment data to generate a road traffic scene; for each target dynamic traffic participant in the road traffic scene, road traffic data under multiple visual angles are obtained by switching the visual angles of the target dynamic traffic participants, and the road traffic data are used for training an automatic driving model. By adopting the technical scheme, the problem of insufficient completeness of model training data is solved.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of autonomous driving technology, and specifically, to a method and apparatus for processing autonomous driving model data based on a vehicle-road-cloud integrated architecture, an autonomous driving model self-evolution learning system, equipment, and medium. Background Art

[0002] With the vigorous development of a new round of technological revolution and industrial transformation, the deep integration of automobiles with artificial intelligence, information communication, large models and other technologies has made intelligence and networking an important direction for the development of the automobile industry. The "Guidelines for the Construction of the National Internet of Vehicles Industry Standard System (Intelligent Connected Vehicles)" released in 2023 proposes that by 2030, a comprehensive intelligent connected vehicle standard system will be formed to support the coordinated development of single-vehicle intelligence and networking empowerment.

[0003] At present, with the advancement of pilot cities for vehicle-road-cloud integration, the construction of intelligent roadside infrastructure has been carried out on a large scale, and various operation and maintenance companies have accumulated a large amount of roadside data. Roadside equipment can obtain traffic flow data of different models, different seasons, different light / temperatures and holidays from an all-round perspective with a long period, high frequency and wide field of view. Due to the lack of high-value scene recognition and extraction tools, these data have low data utilization value and cause huge waste of resources. On the other hand, major companies and other new forces in car manufacturing have entered the development of autonomous driving models in order to occupy the high ground of technology and application. However, the chimney-style development model of various car companies has led to isolated model training data and incomplete types, which cannot meet the completeness of training data. The trained models have poor generalization ability and robustness, and are difficult to adapt to the complex and changeable real traffic environment. Summary of the invention

[0004] Embodiments of the present invention provide a method and apparatus for processing autonomous driving model data under a vehicle-road-cloud integrated architecture, an autonomous driving model self-evolutionary learning system, equipment, and medium, to solve the problems of low utilization of roadside data and insufficient completeness of model training data.

[0005] In a first aspect, the present invention provides a method for processing autonomous driving model data based on a vehicle-road-cloud integrated architecture, wherein the vehicle-road-cloud integrated architecture includes a data collection vehicle, a roadside sensing device, and a cloud. The processing method provided by an embodiment of the present invention is applied to the cloud, and the processing method includes:

[0006] Receiving roadside data uploaded by roadside sensing equipment, and receiving vehicle-side sensing data uploaded by data collection vehicles, wherein the roadside data includes traffic flow data;

[0007] Extracting high-value scene segment data from the roadside data, wherein the high-value scene segment data includes driving trajectory information of multiple dynamic traffic participants in different scenarios, and the driving trajectory information includes location information and speed information of trajectory points;

[0008] Reconstruct 3D static road scenes based on vehicle-side perception data, and identify and extract multiple dynamic traffic participants from vehicle-side perception data;

[0009] Combine multiple dynamic traffic participants, reconstructed 3D static road scenes, and high-value scene fragment data to generate road traffic scenes;

[0010] For each target dynamic traffic participant in the road traffic scene, road traffic data from multiple perspectives are obtained by switching the perspectives of each target dynamic traffic participant, and the road traffic data is used to train the autonomous driving model.

[0011] Optionally, the method provided in this embodiment further includes:

[0012] Test the trained autonomous driving model and deploy the tested autonomous driving model to the intelligent connected vehicle.

[0013] Optionally, extract high-value scene segment data from roadside data, including:

[0014] Determine the intersection section corresponding to the roadside data, and convert the high-precision map data corresponding to the intersection section into local map data that meets the training data format requirements;

[0015] Extract the driving trajectory information of each dynamic traffic participant within a set time period from the roadside data, and search in the local map according to the starting point location information of each dynamic traffic participant to obtain the candidate driving trajectory of each dynamic traffic participant, and smooth the candidate driving trajectory;

[0016] Based on the candidate driving trajectory after smoothing and combined with the local map data, the behavior characteristic information of each dynamic traffic participant is determined, wherein the behavior characteristics include lane affiliation, collision characteristics, distance between trajectory points, distance between target contours, and traffic speed and traffic volume of each lane;

[0017] According to the behavior characteristic information of each dynamic traffic participant, the target behavior of each dynamic traffic participant is determined, and a corresponding behavior label is added to each target behavior, and the behavior label is stored in a scene label library, wherein the scene label library also includes multiple types of scene labels, and each type of scene label includes multiple levels of sub-labels;

[0018] Filter out target behavior labels and various scene labels that meet the model training requirements from the scene label library, combine the filtered labels, and use the combined labels and their corresponding target behavior data as high-value scene fragment data.

[0019] Optionally, the high-precision map data corresponding to the intersection section is converted into local map data that meets the training data format requirements, including:

[0020] Convert the map format of the high-precision map data corresponding to the intersection section into the geographic data format ShapeFile;

[0021] Extracting road elements from the map data converted from the map format, the road elements including lane center lines, lane lines, road lines and intersection areas;

[0022] The extracted road elements are converted so that the data form of the converted road elements meets the requirements of the training data, wherein the conversion includes converting the extracted road elements from the geocentric coordinate system WGS84 to the universal transverse Mercator coordinate system UTM, converting the enumeration type, and sparse processing of data points;

[0023] According to the road elements after data format conversion and the topological connection relationship between the road elements, local map data that meets the training data format requirements are obtained.

[0024] Optionally, combine the filtered tags, including:

[0025] The screened tags are combined according to preset tag combination rules, wherein the preset tag combination rules include any one or more of the following: logically combining different tags according to the relationship between different tags, sorting the tags in chronological order, assigning different weights to different tags, and adding different priorities to different tags.

[0026] In a second aspect, an embodiment of the present invention further provides a processing device for autonomous driving model data based on a vehicle-road-cloud integrated architecture, the processing device comprising:

[0027] A data receiving module is configured to receive roadside data uploaded by a roadside sensing device and vehicle-side sensing data uploaded by a data collection vehicle, wherein the roadside data is traffic flow data;

[0028] A high-value scene segment data extraction module is configured to extract high-value scene segment data from the roadside data, wherein the high-value scene segment data includes driving trajectory information of multiple dynamic traffic participants in different scenarios, and the driving trajectory includes location information and speed information of the trajectory point;

[0029] A static road scene reconstruction module is configured to reconstruct a 3D static road scene based on the vehicle-side perception data, and identify and extract multiple dynamic traffic participants from the vehicle-side perception data;

[0030] A road traffic scene generation module is configured to combine multiple dynamic traffic participants, reconstructed 3D static road scenes, and high-value scene fragment data to generate road traffic scene data;

[0031] The training data generation module is configured to obtain road traffic data from multiple perspectives for each target dynamic traffic participant in the road traffic scene by switching the perspective of each target dynamic traffic participant. The road traffic data is used to train the autonomous driving model.

[0032] Optionally, the device provided by the embodiment of the present invention further includes:

[0033] The model deployment module is configured to test the trained autonomous driving model and deploy the tested autonomous driving model to the intelligent connected vehicle.

[0034] Optional, high-value scene segment data extraction module, including:

[0035] A local map data conversion unit is configured to determine the intersection section corresponding to the roadside data, and convert the high-precision map data corresponding to the intersection section into local map data that meets the training data format requirements;

[0036] A driving trajectory determination unit is configured to extract driving trajectory information of each dynamic traffic participant within a set time period from the roadside data, and search in the local map according to the starting point position information of each dynamic traffic participant to obtain candidate driving trajectories of each dynamic traffic participant, and perform smoothing on the candidate driving trajectories;

[0037] A characteristic information determination unit is configured to determine the behavior characteristic information of each dynamic traffic participant based on the candidate driving trajectory after smoothing and in combination with the local map data, wherein the behavior characteristics include lane affiliation, collision characteristics, distances between trajectory points, distances between target contours, and traffic speed and traffic volume of each lane;

[0038] The target behavior determination unit is configured to determine the target behavior of each dynamic traffic participant according to the behavior characteristic information of each dynamic traffic participant, add a corresponding behavior label to each target behavior, and store the behavior label in a scene label library, wherein the scene label library also includes multiple types of scene labels, and each type of scene label includes multiple levels of sub-labels;

[0039] The label combination unit is configured to filter out target behavior labels and various scene labels that meet the model training requirements from the scene label library, combine the filtered labels, and use the combined labels and their corresponding target behavior data as high-value scene fragment data.

[0040] Optionally, the local map data conversion unit is specifically configured as follows:

[0041] Convert the map format of the high-precision map data corresponding to the intersection section into a geographic data format ShapeFile;

[0042] Extracting road elements from the map data after the map format conversion, wherein the road elements include lane center lines, lane lines, road lines and intersection areas;

[0043] The extracted road elements are converted so that the data form of the converted road elements meets the requirements of the training data, wherein the conversion includes converting the extracted road elements from the geocentric coordinate system WGS84 to the universal transverse Mercator coordinate system UTM, converting the enumeration type, and sparse processing of data points;

[0044] According to the road elements after data format conversion and the topological connection relationship between the road elements, local map data that meets the training data format requirements are obtained.

[0045] Optionally, the label combination unit is specifically configured as follows:

[0046] Filter out target behavior labels and various scene labels that meet the model training requirements from the scene label library;

[0047] Combining the screened tags according to a preset tag combination rule, wherein the preset tag combination rule includes any one or more of the following: logically combining different tags according to the relationship between different tags, sorting the tags in chronological order, assigning different weights to different tags, and adding different priorities to different tags;

[0048] The combined labels and their corresponding target behavior data are used as high-value scene fragment data.

[0049] In a third aspect, an embodiment of the present invention further provides a self-evolutionary learning system for a large dynamic driving model based on a vehicle-road-cloud integrated architecture, the learning system comprising:

[0050] Data collection vehicle, used to collect vehicle-side perception data and upload the vehicle-side perception data to the cloud control basic platform. The vehicle-side perception data includes surrounding environment information and traffic participant status information;

[0051] Roadside sensing equipment is used to collect roadside data and upload it to the cloud control basic platform. The roadside data includes traffic flow data;

[0052] The cloud control basic platform is used to pre-process the vehicle-side perception data and roadside data, and upload the pre-processed vehicle-side perception data and pre-processed roadside data to the cloud;

[0053] Cloud for:

[0054] Receiving pre-processed vehicle-side sensing data, and receiving pre-processed roadside data;

[0055] Extracting high-value scene segment data from roadside data, wherein the high-value scene segment data includes driving trajectory information of multiple dynamic traffic participants under different scene labels, and the driving trajectory information includes location information and speed information of trajectory points;

[0056] Reconstruct 3D static road scenes based on vehicle-side perception data, and identify and extract multiple dynamic traffic participants from vehicle-side perception data;

[0057] Combine multiple dynamic traffic participants, reconstructed 3D static road scenes, and high-value scene fragment data to generate road traffic scene data;

[0058] For each target dynamic traffic participant in the road traffic scene, the road traffic data under multiple perspectives are obtained by switching the perspectives of each target dynamic traffic participant, and the autonomous driving model is trained with the road traffic data;

[0059] Test the trained autonomous driving model and deploy the tested autonomous driving model to the intelligent connected vehicle.

[0060] In a fourth aspect, an embodiment of the present invention further provides a computing device, including:

[0061] A memory storing executable program code;

[0062] a processor coupled to the memory;

[0063] The processor calls the executable program code stored in the memory to execute the method for processing autonomous driving model data based on the vehicle-road-cloud integrated architecture provided by any embodiment of the present invention.

[0064] In a fifth aspect, an embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method for processing autonomous driving model data based on the vehicle-road-cloud integrated architecture provided by any embodiment of the present invention is implemented.

[0065] The technical solution provided by the embodiment of the present invention can obtain high-value scene fragments by identifying roadside data, and the obtained high-value scene fragments can be used to generate road traffic data from multiple perspectives on the vehicle side. This setting not only improves the utilization rate of roadside data, but also effectively improves the completeness and diversity of vehicle-side data. Using the obtained vehicle-side data from multiple perspectives to train the autonomous driving model can effectively improve the generalization ability and robustness of the trained model, so that it can adapt to the complex and changeable real-world traffic environment. In addition, the use of the trained autonomous driving model has no limitations and can be deployed in vehicles of various brands, which solves the problem that the training results of each car company in the related technology are only applicable to vehicles of their own brands and cannot be shared.

[0066] The innovative features of the embodiments of the present invention include:

[0067] 1. By identifying the roadside data, high-value scene fragment data can be obtained. This setting not only improves the utilization rate of the roadside data, but also increases the diversity of the training samples of the autonomous driving model. Compared with the method of only using the data collected by the vehicle as the training samples of the autonomous driving model, in the embodiment of the present invention, the use of roadside data makes the content of the training samples of the autonomous driving model richer, and the generalization ability of the trained model is effectively improved, which is one of the innovations of the embodiment of the present invention.

[0068] 2. By using high-value scene fragment data extracted from roadside data and combining it with vehicle-side perception data, road traffic data from the vehicle-side perspective can be generated. This setting effectively improves the completeness and diversity of vehicle-side data. Using the vehicle-side data from multiple perspectives to train the autonomous driving model can effectively improve the generalization ability and robustness of the trained model, so that it can adapt to the complex and changeable real traffic environment, which is one of the innovative points of the embodiment of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the embodiments of the present invention 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, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0070] Figure 1a A schematic diagram of a vehicle-road-cloud integrated data closed loop provided in Embodiment 1 of the present invention;

[0071] Figure 1b A flowchart of a method for processing autonomous driving model data based on a vehicle-road-cloud integrated architecture provided in Embodiment 1 of the present invention;

[0072] Figure 2 A schematic diagram of the structure of a large-scale self-evolutionary learning system for dynamic driving model based on a vehicle-road-cloud integrated architecture provided in the second embodiment of the present invention;

[0073] Figure 3 A structural block diagram of a device for processing autonomous driving model data based on a vehicle-road-cloud integrated architecture provided in Embodiment 3 of the present invention;

[0074] Figure 4 A schematic diagram of the structure of a computing device provided in Embodiment 4 of the present invention. DETAILED DESCRIPTION

[0075] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] It should be noted that the terms "including" and "having" and any variations thereof in the embodiments of the present invention and the accompanying drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices.

[0077] The embodiments of the present invention disclose a method, device, self-evolution learning system, equipment and medium for processing autonomous driving model data based on a vehicle-road-cloud integrated architecture. The following are detailed descriptions of each.

[0078] Embodiment 1

[0079] The method for processing the autonomous driving model data provided in the first embodiment of the present invention is implemented based on a vehicle-road-cloud integrated architecture. Figure 1a A schematic diagram of a vehicle-road-cloud integrated data closed loop provided in Embodiment 1 of the present invention, such as Figure 1a As shown, the vehicle-road-cloud integrated architecture includes roadside sensing devices (i.e. Figure 1a Roadside collection equipment in the data collection vehicle (i.e. Figure 1a Smart Driving in the cloud (including Figure 1aThe data generation tool and training deployment tool in the cloud) include roadside perception devices, such as roadside cameras and roadside radars, which can obtain perception data such as traffic participants, traffic events, and traffic flow status in the area of ​​interest. Data collection vehicles can obtain surrounding environment information, traffic participant status information, etc. through on-board devices such as on-board cameras and on-board radars. Roadside perception devices and data collection vehicles can first upload the collected data to the cloud control basic platform, which pre-processes, stores and manages the received data, and then uploads the pre-processed data to the cloud. The data generation tool in the cloud will generate an autonomous driving model training data set based on the received roadside data and vehicle-side data. Then, users can use the training deployment tool to train the autonomous driving model, and by testing and verifying the autonomous driving model, they can output a reliable large model, which can be deployed to the vehicle through OTA (Over-The-Air). Finally, roadside collection equipment and intelligent driving vehicles can collect new high-value data again and transmit it to the cloud, automatically optimize the autonomous driving model in the cloud, and then deploy the optimized autonomous driving model to the vehicle through OTA tools, thus realizing an integrated vehicle-road-cloud data closed loop.

[0080] Figure 1b This is a flowchart of a method for processing autonomous driving model data based on a vehicle-road-cloud integrated architecture provided in the first embodiment of the present invention. The method is applied to the cloud of the vehicle-road-cloud integrated architecture and can be executed by a data generation tool on the cloud. The data generation tool can be implemented in software and / or hardware. Figure 1b As shown, the method provided in this embodiment specifically includes:

[0081] S110, receiving roadside data uploaded by roadside sensing equipment, and receiving vehicle-side sensing data uploaded by data collection vehicles.

[0082] Among them, roadside data includes traffic flow data of different vehicle models, different seasons, different light / temperatures and holidays. Roadside data is collected by roadside sensing equipment, which includes roadside cameras, roadside radars and other equipment installed on the road. Roadside sensing equipment can obtain roadside data from an all-round perspective with a long period, high frequency and wide field of view, and can use single-point sensing, cross-point sensing, regional continuous sensing and other technical means to manage roadside data and improve the quality of roadside data. Vehicle-side sensing data includes traffic environment information, traffic participant status information, etc. Vehicle-side sensing data can be collected by on-board cameras, on-board radars and other equipment in data collection vehicles.

[0083] In this embodiment, in order to facilitate the processing of roadside data and vehicle-side perception data by the cloud, the roadside perception device and the data collection vehicle can upload the collected data to the cloud control basic platform respectively. The cloud control basic platform first pre-processes the roadside data and vehicle-side perception data, such as performing noise filtering, invalid data deletion, etc., and then uploads the pre-processed roadside data and vehicle-side perception data to the cloud. The roadside data and vehicle-side perception data processed by the cloud are the data pre-processed by the cloud control basic platform.

[0084] S120: Extract high-value scene segment data from the roadside data.

[0085] In this embodiment, the high-value scene segment data includes the driving trajectory information of multiple dynamic traffic participants in different scenes, wherein the dynamic traffic participants include pedestrians, vehicles, etc. The driving trajectory information of the dynamic traffic participants includes the location information and speed information of the trajectory points. Different scenes can be represented by scene tags.

[0086] In this embodiment, the high-value scene fragment data extracted from the roadside data can be used to generate training samples for the autonomous driving model. This configuration not only improves the utilization rate of the roadside data, but also increases the diversity of the training samples for the autonomous driving model. Compared with the method of using only the data collected by the vehicle as the training samples for the autonomous driving model, in this embodiment, the use of roadside data makes the content of the training samples for the autonomous driving model richer, and the generalization ability of the trained model is effectively improved.

[0087] As an optional implementation, extracting high-value scene segment data from roadside data can be achieved through the following steps A to E:

[0088] A. Determine the corresponding intersection section based on the roadside data, and convert the high-precision map data corresponding to the intersection section into local map data that meets the training data format requirements.

[0089] In this embodiment, the corresponding intersection section can be determined according to the location of the roadside sensing device, and the high-precision map data corresponding to the intersection section can be extracted. Among them, the high-precision map data is usually used in the high-precision navigation process of autonomous driving, and in the training stage of the autonomous driving model, it needs to be converted into local map data that meets the training data format requirements. Among them, the training data format requirements include map format requirements, coordinate system conversion of road elements, etc., which can be achieved through the following steps:

[0090] a1. Convert the map format of high-precision map data corresponding to the intersection section, such as OpenDrive (an open file format) and NDS (Navigation Data Standard) format, into the geographic data format (ShapeFile) required for model training.

[0091] a2. Extract road elements from the map data after the map format conversion.

[0092] Among them, road elements include lane center lines, lane lines, road lines and intersection areas.

[0093] a3. Convert the extracted road elements so that the data format of the converted road elements meets the requirements of the training data.

[0094] The data format conversion of the road elements includes: converting the extracted road elements from the geocentric coordinate system (WGS84 coordinate system) to the UTM (Universal Transverse Mercator) coordinate system, converting the enumeration type, and sparse processing of data points.

[0095] a4. According to the road elements after data format conversion and the topological connection relationship between the road elements, local map data that meets the training data format requirements are obtained.

[0096] Specifically, the topological connection relationship of each road element in the local map after data format conversion may be determined according to the topological connection relationship of each road element in the high-precision map.

[0097] In this embodiment, by adopting the above technical solution, the high-precision map data of a certain intersection section can be converted into local map data that meets the training data format requirements.

[0098] B. Extract the driving trajectory information of each dynamic traffic participant within a set time period from the roadside data, and search in the local map according to the starting point location information of each dynamic traffic participant to obtain the candidate driving trajectory of each dynamic traffic participant, and smooth the candidate driving trajectory.

[0099] Among them, the set time period can be set according to actual needs. In this embodiment, a sliding window method can be used, with several time domain lengths (for example, 10-60s) as a sliding window processing cycle to extract the driving trajectory information of all dynamic traffic participants during this period. For each dynamic traffic participant, the driving path can be searched according to the starting point position of the dynamic traffic participant in combination with the local map to obtain all possible driving path information of the dynamic traffic participant, that is, the candidate driving path. Then, the trajectory optimization algorithm, such as the FemPosSmooth algorithm (an algorithm for reference line smoothing), can be used to optimize the driving trajectory of the dynamic traffic participant to alleviate the position jump in the driving trajectory. Then, based on the processed trajectory information, a filtering algorithm (such as a Kalman filtering algorithm) can be used to optimize the speed information in the driving trajectory of the dynamic traffic participant, and finally the processed trajectory and speed information are obtained.

[0100] C. Based on the smoothed candidate driving trajectory and combined with local map data, the behavioral characteristic information of each dynamic traffic participant is determined.

[0101] Among them, the behavioral characteristics include lane affiliation, collision characteristics, distance between trajectory points, distance between target contours, and traffic speed and traffic volume in each lane. This embodiment does not specifically limit the types and quantities of behavioral characteristics of dynamic participants.

[0102] Specifically, for the behavior feature of lane affiliation, the lane the current dynamic traffic participant is currently in and the target lane to be changed to can be determined based on the driving trajectory information of the current dynamic traffic participant and combined with the local map.

[0103] For collision features, collision detection can be performed based on various collision detection algorithms to predict possible collisions between the current dynamic traffic participant and other dynamic traffic participants, or collisions that have already occurred.

[0104] For the behavior feature of trajectory distance, the Euclidean distance between different trajectory points at any time or continuous time, as well as the Euclidean distance between the contours of dynamic traffic participants can be calculated.

[0105] For the behavior characteristic of traffic speed, the average speed distribution and traffic volume of all lanes can be calculated.

[0106] D. According to the behavioral characteristic information of each dynamic traffic participant, determine the target behavior of each dynamic traffic participant, add a corresponding behavior label to each target behavior, and store each behavior label in the scene label library.

[0107] Among them, the behaviors of dynamic traffic participants include lane changing, slow driving in congestion, other vehicles cutting into the current lane (Cut-In) and detouring around obstacles, etc. This embodiment does not specifically limit the behavior types of dynamic traffic participants.

[0108] Specifically, for the behavior of lane changing, it is possible to determine whether the dynamic traffic participant wants to change lanes based on the behavioral characteristic of lane affiliation, and to calculate indicators such as the number of lane changes, lane change time, and lane change lateral speed / acceleration. Then, the collision characteristics and trajectory distance can be used to determine the situation of dangerous lane changes.

[0109] For the behavior of slowing down in congestion, the congestion slowing down scenario can be judged based on behavioral characteristics such as lane affiliation and traffic speed.

[0110] For the behavior of other vehicles cutting into the current lane, it is possible to determine whether there are other vehicles cutting into the lane where a dynamic traffic participant is currently traveling based on lane affiliation, collision characteristics and trajectory distance, and calculate TTC (Time To Collision, the estimated time when two vehicles may collide) and the shortest distance.

[0111] For the behavior of detouring around obstacles, behavioral characteristics such as lane affiliation and traffic speed can be used, and the scenario of detouring around obstacles can be determined based on the type of obstacles detected.

[0112] For the above-mentioned typical behaviors of dynamic traffic participants, in the actual model training process, it is necessary to combine the needs of users and carry out targeted parameter calibration and feature value selection in order to better meet the specific needs of users.

[0113] After determining the target behavior of dynamic traffic participants, corresponding behavior labels can be added to each target behavior data segment, and each behavior label can be stored in the scene label library for subsequent training of the autonomous driving model. Among them, the scene label library stores multiple types of labels, including lighting condition labels, weather condition labels, time period labels, visibility labels, road type labels, traffic flow status labels, and target behavior labels. Each type of label is further divided into multiple sub-category labels, such as the intensity level of lighting conditions, weather changes, data in different time periods, visibility distance classification, road type, and specific actions of target behavior.

[0114] E. Filter out target behavior labels and various scene labels that meet the model training requirements from the scene label library, combine the filtered labels, and use the combined labels and their corresponding target behavior data as high-value scene fragment data, where each type of scene label includes multiple levels of sub-labels.

[0115] In this embodiment, high-value scenarios are defined by tags, which can effectively mine high-value scenarios for autonomous driving from real data. In order to ensure the comprehensiveness and effectiveness of high-value scenarios, the operation of tags in this embodiment mainly includes two parts: tag screening and tag combination.

[0116] In the label screening stage, scene labels can be screened according to a series of predefined criteria. The predefined criteria include lighting conditions, weather conditions, time periods, visibility, road types, traffic flow status and target behavior. Each criterion is further divided into multiple subcategories, such as the intensity of lighting conditions, weather changes, data in different time periods, visibility distance classification, road types and specific actions of target behavior.

[0117] In the label combination stage, the filtered labels can be combined according to the preset label combination rules so that the combined labels meet the model training requirements. The preset label combination rules may include any one or more of the following:

[0118] (1) Logical combination, such as using logical AND, OR, NOT conditions to determine the relationship between tags;

[0119] (2) Sequence combination, for example, sorting labels according to time series, such as sorting labels according to the order of changing lanes first, then cutting into a lane, and then changing lanes again after cutting into a lane;

[0120] (3) Weight allocation, that is, assigning different weights to different labels to reflect their influence in scene recognition;

[0121] (4) Priority combination, that is, giving different priorities according to the importance of tags.

[0122] The following is a specific example of screening high-value driving risk scenarios under certain weak perceptions to illustrate the process of label combination:

[0123] ① Use logical combinations (such as logical "AND") and conditional combinations, including labels such as "light conditions (low light)", "weather conditions (rainy days)", and "time conditions (night)" to perform preliminary filtering on the data;

[0124] ② Use sequence combinations to match labels such as "target behavior (CutIn)", "target behavior (lane change)", and "target behavior (CutIn)" in chronological order to perform secondary filtering on the data;

[0125] ③ Use weight allocation to set weights from high to low for labels such as "road type (intersection area)", "visibility (low visibility)", and "traffic flow status (heavy traffic volume)", and filter the data again according to the set threshold of the weight and value, for example, extract the data with weight and value higher than the set threshold.

[0126] ④ Using the priority combination, the data is prioritized in the order that the priority of the label “largest number (target behavior (CutIn))” is higher than the priority of the label “largest number (target behavior (lane change))”.

[0127] According to the above-mentioned label combination process, after completing the combination of various labels, the data obtained is high-value scene data, that is, the driving trajectory data of dynamic traffic participants containing various scene labels and behavior labels. These data fragments are written into the training sample database for use by subsequent model training software.

[0128] It should be noted that the extraction of high-value scene data provided in this embodiment adopts the method of label screening and label combination to meet the user's demand for high-value scene extraction. This embodiment does not specifically limit the label screening rules and combination rules. In actual engineering applications, users can customize high-value scenes according to actual usage requirements, and can perform targeted label screening and combination processing based on actual definitions.

[0129] S130, reconstructing the 3D static road scene based on the vehicle-side perception data, and identifying and extracting multiple dynamic traffic participants from the vehicle-side perception data.

[0130] Among them, the vehicle-side perception data includes surrounding environment information, traffic participant status information, etc. Based on the vehicle-side perception data, the 3D static road scene in the traffic scene can be reconstructed. Among them, there are many methods for reconstructing three-dimensional scenes. For example, 3DGS (3D Gaussian Splatting) technology can be used to effectively organize the transmitted information in the form of Gaussian distribution by explicitly representing the dot matrix in 3D space, thereby realizing real-time image rendering. Alternatively, a laser beam can be used to illuminate the surface of the object, and accurate three-dimensional data can be obtained by measuring the laser reflection time or angle. Alternatively, a large amount of image data can be trained using a deep learning algorithm to learn the three-dimensional shape and appearance information of the object. This embodiment does not specifically limit the reconstruction means of the three-dimensional scene. For the reconstructed 3D static road scene, the static traffic participants contained therein, such as buildings, traffic signs, street lights and green belts, can use the target 3D annotation method to obtain accurate annotation information.

[0131] In addition, each dynamic traffic participant can also be identified from the vehicle-side perception data. In this embodiment, the information of each dynamic traffic participant, including vehicles, pedestrians, etc., can be extracted from the vehicle-side perception data for use in the subsequent generation of road traffic scenes.

[0132] S140, combining multiple dynamic traffic participants, reconstructed 3D static road scenes, and high-value scene fragment data to generate a road traffic scene.

[0133] In this embodiment, the driving trajectory information of multiple dynamic traffic participants extracted from the vehicle-side perception data can be matched with the driving trajectory information of multiple dynamic traffic participants in the high-value scene fragment data to obtain the driving trajectory information of multiple target dynamic traffic participants. Then, the driving trajectory information of the multiple target dynamic traffic participants can be matched with the reconstructed 3D static road scene, and the driving trajectory information of the target dynamic traffic participants can be embedded in the reconstructed 3D static road scene in real time according to the timestamp. In this process, the consistency of the two parts of data in time and space must be ensured.

[0134] Specifically, the reconstructed 3D static road scene data can be imported into the selected 3D modeling software or traffic simulation platform. During this process, the scale and coordinate system of the scene can be adjusted as needed to ensure that the 3D static road scene is consistent with the actual traffic conditions. Then, the driving trajectory information of each target dynamic participant can be mapped to the 3D static road scene according to the timestamp. During this process, it is necessary to ensure that the time in the 3D scene is consistent with the timestamp of the driving trajectory of the dynamic traffic participants. After the road traffic scene is obtained, it can be displayed through professional visualization tools or platforms.

[0135] S150: For each target dynamic traffic participant in the road traffic scene, road traffic data under multiple perspectives are obtained by switching the perspectives of each target dynamic traffic participant.

[0136] In this embodiment, by switching the perspectives of each target dynamic traffic participant, road traffic data from multiple perspectives can be obtained, and the obtained road traffic data can be used for the training of the autonomous driving model. After the training is completed, the trained autonomous driving model can be tested, and the testing process can be implemented through the model deployment tool. The model deployment tool can be used to perform rapid deployment verification to ensure the performance of the model on the vehicle-side hardware, and the deployment tool software supports deployment operation integration and code reuse on different hardware platforms, thereby simplifying the deployment process.

[0137] Furthermore, the tested autonomous driving models can be deployed to smart connected vehicles of various brands through OTA tools, and the autonomous driving models of vehicles of different brands can be upgraded through OAT tools.

[0138] In the related art, each car company can usually only collect road data through its own road collection vehicles, and the collected road data will not be shared with other car companies, which will cause the problem of single and incomplete training data for autonomous driving models. In addition, each car company only trains the autonomous driving models developed by itself, and the training results are not shared with each other. The generalization ability and robustness of the autonomous driving models trained with such training data are poor, and it is difficult to adapt to the complex and changeable real traffic environment. In this embodiment,

[0139] By identifying the roadside data, high-value scene fragments can be obtained, and the obtained high-value scene fragments can be used to generate road traffic data from the vehicle-side perspective. This setting not only improves the utilization rate of roadside data, but also effectively improves the completeness and diversity of vehicle-side data. Using the vehicle-side data from multiple perspectives to train the autonomous driving model can effectively improve the generalization ability and robustness of the trained model, so that it can adapt to the complex and changeable real-world traffic environment. In addition, the use of the trained autonomous driving model has no limitations and can be deployed in vehicles of various brands, which solves the problem that the training results of each car company in the relevant technology are only applicable to its own brand of vehicles and cannot be shared.

[0140] Embodiment 2

[0141] Figure 2 A structural diagram of a large-scale self-evolutionary learning system for dynamic driving model based on a vehicle-road-cloud integrated architecture provided in the second embodiment of the present invention is shown in FIG. Figure 2 As shown, the system provided in this embodiment includes a data collection vehicle 210, a roadside sensing device 220, a cloud control basic platform 230 and a cloud 240, wherein:

[0142] The data collection vehicle 210 is used to collect vehicle-side perception data and upload the vehicle-side perception data to the cloud control basic platform 230, wherein the vehicle-side perception data includes surrounding environment information and traffic participant status information.

[0143] The roadside sensing device 220 is used to collect roadside data and upload the roadside data to the cloud control basic platform 230, wherein the roadside data includes traffic flow data.

[0144] The cloud control basic platform 230 is used to pre-process the vehicle-side perception data and the roadside data, and upload the pre-processed vehicle-side perception data and the pre-processed roadside data to the cloud 240.

[0145] The cloud 240 is configured with a data generation tool, a model training tool, and a model deployment tool, wherein the data generation tool is used to:

[0146] Receiving pre-processed vehicle-side sensing data, and receiving pre-processed roadside data;

[0147] Extracting high-value scene segment data from roadside data, wherein the high-value scene segment data includes driving trajectory information of multiple dynamic traffic participants under different scene labels, and the driving trajectory includes location information and speed information of trajectory points;

[0148] Reconstructing the 3D static road scene based on the vehicle-side perception data, and extracting the multiple dynamic traffic participants from the vehicle-side perception data;

[0149] Combine multiple dynamic traffic participants, reconstructed 3D static road scenes, and high-value scene fragment data to generate road traffic scene data;

[0150] For each target dynamic traffic participant in the road traffic scene, the road traffic data under multiple perspectives are obtained by switching the perspectives of each target dynamic traffic participant.

[0151] After generating road traffic data from multiple vehicle-side perspectives, the model training tool uses the road traffic data and the user's initial data set as input to train the autonomous driving model to improve the generalization of the model. The trained autonomous driving model is quickly deployed and verified through the model deployment tool to ensure the performance of the model on the vehicle-side hardware. The autonomous driving model that has been deployed and verified is deployed to the vehicle through the OTA tool, and the vehicle-side model version can be upgraded through the OTA tool to improve the performance of the autonomous driving vehicle. When the roadside perception equipment and data collection vehicles upload the collected new data to the cloud again, the cloud can generate new training samples according to the above process, and use the new training samples to train the autonomous driving model, so that the performance of the autonomous driving model can be continuously optimized.

[0152] In this embodiment, the process of obtaining high-value scene fragments by identifying roadside data, and the process of generating road traffic data from multiple perspectives on the vehicle side using high-value scene fragments and vehicle-side perception data can be referred to the description of the above embodiment and will not be repeated here.

[0153] Embodiment 3

[0154] Figure 3 A structural block diagram of a device for processing autonomous driving model data based on a vehicle-road-cloud integrated architecture provided in Embodiment 3 of the present invention, such as Figure 3 As shown, the processing device provided in this embodiment includes: a data receiving module 310, a high-value scene segment data extraction module 320, a static road scene reconstruction module 330, a road traffic scene generation module 340 and a training data generation module 350, wherein:

[0155] The data receiving module 310 is configured to receive roadside data uploaded by a roadside sensing device and vehicle-side sensing data uploaded by a data collection vehicle, wherein the roadside data is traffic flow data;

[0156] The high-value scene segment data extraction module 320 is configured to extract high-value scene segment data from the roadside data, wherein the high-value scene segment data includes driving trajectory information of multiple dynamic traffic participants in different scenes, and the driving trajectory includes position information and speed information of the trajectory points;

[0157] A static road scene reconstruction module 330 is configured to reconstruct the 3D static road scene based on the vehicle-side perception data and extract multiple dynamic traffic participants from the vehicle-side perception data;

[0158] A road traffic scene generation module 340 is configured to combine multiple dynamic traffic participants, reconstructed static road scenes, and high-value scene segment data to generate road traffic scene data;

[0159] The training data generation module 350 is configured to obtain road traffic data from multiple perspectives for each target dynamic traffic participant in the road traffic scene by switching the perspective of each target dynamic traffic participant. The road traffic data is used to train the autonomous driving model.

[0160] Optionally, the high-value scene segment data extraction module 320 specifically includes:

[0161] A local map data conversion unit is configured to determine the intersection section corresponding to the roadside data, and convert the high-precision map data corresponding to the intersection section into local map data that meets the training data format requirements;

[0162] A driving trajectory determination unit is configured to extract driving trajectory information of each dynamic traffic participant within a set time period from the roadside data, and search in the local map according to the starting point position information of each dynamic traffic participant to obtain candidate driving trajectories of each dynamic traffic participant, and perform smoothing on the candidate driving trajectories;

[0163] A characteristic information determination unit is configured to determine the behavior characteristic information of each dynamic traffic participant based on the candidate driving trajectory after smoothing and in combination with the local map data, wherein the behavior characteristics include lane affiliation, collision characteristics, distances between trajectory points, distances between target contours, and traffic speed and traffic volume of each lane;

[0164] The target behavior determination unit is configured to determine the target behavior of each dynamic traffic participant according to the behavior characteristic information of each dynamic traffic participant, add a corresponding behavior label to each target behavior, and store the behavior label in a scene label library, wherein the scene label library also includes multiple types of scene labels, and each type of scene label includes multiple levels of sub-labels;

[0165] The label combination unit is configured to filter out target behavior labels and various scene labels that meet the model training requirements from the scene label library, combine the filtered labels, and use the combined labels and their corresponding target behavior data as high-value scene fragment data.

[0166] Optionally, the local map data conversion unit is specifically configured as follows:

[0167] Convert the map format of the high-precision map data corresponding to the intersection section into a geographic data format ShapeFile;

[0168] Extracting road elements from the map data after the map format conversion, wherein the road elements include lane center lines, lane lines, road lines and intersection areas;

[0169] The extracted road elements are converted so that the data form of the converted road elements meets the requirements of the training data, wherein the conversion includes converting the extracted road elements from the geocentric coordinate system WGS84 to the universal transverse Mercator coordinate system UTM, converting the enumeration type, and sparse processing of data points;

[0170] According to the road elements after data format conversion and the topological connection relationship between the road elements, local map data that meets the training data format requirements are obtained.

[0171] Optionally, the label combination unit is specifically configured as follows:

[0172] Filter out target behavior labels and various scene labels that meet the model training requirements from the scene label library;

[0173] Combining the screened tags according to a preset tag combination rule, wherein the preset tag combination rule includes any one or more of the following: logically combining different tags according to the relationship between different tags, sorting the tags in chronological order, assigning different weights to different tags, and adding different priorities to different tags;

[0174] The combined labels and their corresponding target behavior data serve as high-value scene fragment data.

[0175] Embodiment 4

[0176] See also Figure 4 , Figure 4 This is a schematic diagram of the structure of a computing device provided in Embodiment 4 of the present invention, where the computing device is a cloud server. Figure 4 As shown, the computing device may include:

[0177] A memory 701 storing executable program codes;

[0178] a processor 702 coupled to the memory 701;

[0179] Among them, the processor 702 calls the executable program code stored in the memory 701 to execute the processing method of the automatic driving model data based on the vehicle-road-cloud integrated architecture provided by any embodiment of the present invention.

[0180] An embodiment of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute a method for processing autonomous driving model data based on a vehicle-road-cloud integrated architecture provided by any embodiment of the present invention.

[0181] In various embodiments of the present invention, it should be understood that the size of the serial numbers of the above-mentioned processes does not mean the necessary order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0182] In the embodiments provided by the present invention, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined according to A. However, it should also be understood that determining B according to A does not mean determining B only according to A, and B can also be determined according to A and / or other information.

[0183] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0184] If the above integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-accessible memory. 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, which is stored in a memory and includes several requests for a computer device (which can be a personal computer, a server or a network device, etc., specifically a processor in a computer device) to perform some or all of the steps of the above methods of various embodiments of the present invention.

[0185] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0186] Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of an embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present invention.

[0187] Those skilled in the art can understand that the modules in the device in the embodiment can be distributed in the device in the embodiment according to the description of the embodiment, or can be changed accordingly and located in one or more devices different from the embodiment. The modules in the above embodiment can be combined into one module, or can be further divided into multiple sub-modules.

[0188] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for processing autonomous driving model data based on a vehicle-road-cloud integrated architecture, wherein: The vehicle-road-cloud integrated architecture includes a data collection vehicle, a roadside sensing device, and a cloud. The processing method is applied to the cloud, and is characterized in that the processing method includes: Receiving roadside data uploaded by the roadside sensing device, and receiving vehicle-side sensing data uploaded by the data collection vehicle, wherein the roadside data includes traffic flow data; Extracting high-value scene segment data from the roadside data, wherein the high-value scene segment data includes driving trajectory information of multiple dynamic traffic participants in different scenarios, and the driving trajectory information includes location information and speed information of trajectory points; Reconstructing the 3D static road scene based on the vehicle-side perception data, and identifying and extracting the multiple dynamic traffic participants from the vehicle-side perception data; Combining the multiple dynamic traffic participants, the reconstructed 3D static road scene, and the high-value scene segment data to generate a road traffic scene; For each target dynamic traffic participant in the road traffic scene, road traffic data under multiple perspectives are obtained by switching the perspectives of each target dynamic traffic participant, and the road traffic data is used to train the autonomous driving model.

2. The method according to claim 1, characterized in that The method further comprises: Test the trained autonomous driving model and deploy the tested autonomous driving model to the intelligent connected vehicle.

3. The method according to claim 1, characterized in that The extracting high-value scene segment data from the roadside data comprises: Determine the intersection section corresponding to the roadside data, and convert the high-precision map data corresponding to the intersection section into local map data that meets the training data format requirements; Extracting driving trajectory information of each dynamic traffic participant within a set time period from the roadside data, searching in the local map according to the starting point location information of each dynamic traffic participant, obtaining candidate driving trajectories of each dynamic traffic participant, and smoothing the candidate driving trajectories; Determine the behavior characteristic information of each dynamic traffic participant according to the candidate driving trajectory after smoothing and in combination with the local map data, wherein the behavior characteristic includes lane affiliation, collision characteristics, distance between trajectory points, distance between target contours, and traffic speed and traffic volume of each lane; According to the behavior characteristic information of each dynamic traffic participant, the target behavior of each dynamic traffic participant is determined, and a corresponding behavior label is added to each target behavior, and the behavior label is stored in a scene label library, wherein the scene label library also includes multiple types of scene labels, and each type of scene label includes multiple levels of sub-labels; Target behavior labels and various scene labels that meet the model training requirements are screened out from the scene label library, and the screened labels are combined, and the combined labels and their corresponding target behavior data are used as high-value scene fragment data.

4. The method according to claim 3, characterized in that: The step of converting the high-precision map data corresponding to the intersection section into local map data that meets the training data format requirements includes: Convert the map format of the high-precision map data corresponding to the intersection section into a geographic data format ShapeFile; Extracting road elements from the map data after the map format conversion, wherein the road elements include lane center lines, lane lines, road lines and intersection areas; The extracted road elements are converted so that the data form of the converted road elements meets the requirements of the training data, wherein the conversion includes converting the extracted road elements from the geocentric coordinate system WGS84 to the universal transverse Mercator coordinate system UTM, converting the enumeration type, and sparse processing of data points; According to the road elements after data format conversion and the topological connection relationship between the road elements, local map data that meets the training data format requirements are obtained.

5. The method according to claim 3, characterized in that: The step of combining the filtered tags includes: The screened tags are combined according to preset tag combination rules, wherein the preset tag combination rules include any one or more of the following: logically combining different tags according to the relationship between different tags, sorting the tags in chronological order, assigning different weights to different tags, and adding different priorities to different tags.

6. A processing device for autonomous driving model data based on a vehicle-road-cloud integrated architecture, characterized in that: The processing device comprises: A data receiving module is configured to receive the roadside data uploaded by the roadside sensing device and the vehicle-side sensing data uploaded by the data collection vehicle, wherein the roadside data is traffic flow data; A high-value scene segment data extraction module is configured to extract high-value scene segment data from the roadside data, wherein the high-value scene segment data includes driving trajectory information of multiple dynamic traffic participants in different scenes, and the driving trajectory information includes position information and speed information of trajectory points; A static road scene reconstruction module is configured to reconstruct the 3D static road scene based on the vehicle-side perception data, and identify and extract the multiple dynamic traffic participants from the vehicle-side perception data; A road traffic scene generation module is configured to combine the multiple dynamic traffic participants, the reconstructed 3D static road scene and the high-value scene segment data to generate road traffic scene data; The training data generation module is configured to obtain road traffic data from multiple perspectives for each target dynamic traffic participant in the road traffic scene by switching the perspective of each target dynamic traffic participant, and the road traffic data is used to train the autonomous driving model.

7. The device according to claim 6, characterized in that The high-value scene segment data extraction module specifically includes: A local map data conversion unit is configured to determine the intersection section corresponding to the roadside data, and convert the high-precision map data corresponding to the intersection section into local map data that meets the training data format requirements; a driving trajectory determination unit configured to extract driving trajectory information of each dynamic traffic participant within a set time period from the roadside data, search in the local map according to the starting point position information of each dynamic traffic participant, obtain candidate driving trajectories of each dynamic traffic participant, and perform smoothing on the candidate driving trajectories; a characteristic information determination unit configured to determine the behavior characteristic information of each dynamic traffic participant based on the candidate driving trajectory after smoothing and in combination with the local map data, wherein the behavior characteristics include lane affiliation, collision characteristics, distances between trajectory points, distances between target contours, and traffic speed and traffic volume of each lane; A target behavior determination unit is configured to determine the target behavior of each dynamic traffic participant according to the behavior characteristic information of each dynamic traffic participant, add a corresponding behavior tag to each target behavior, and store the behavior tag in a scene tag library, wherein the scene tag library also includes multiple types of scene tags, and each type of scene tag includes multiple levels of sub-tags; The label combination unit is configured to filter out target behavior labels and various scene labels that meet the model training requirements from the scene label library, combine the filtered labels, and use the combined labels and their corresponding target behavior data as high-value scene fragment data.

8. An autonomous driving model self-evolution learning system based on a vehicle-road-cloud integrated architecture, characterized in that: The learning system comprises: A data collection vehicle is used to collect vehicle-side perception data and upload the vehicle-side perception data to the cloud control basic platform. The vehicle-side perception data includes surrounding environment information and traffic participant status information; A roadside sensing device is used to collect roadside data and upload the roadside data to the cloud control basic platform, wherein the roadside data includes traffic flow data; The cloud control basic platform is used to pre-process the vehicle-side perception data and the roadside data, and upload the pre-processed vehicle-side perception data and the pre-processed roadside data to the cloud; The cloud is used for: Receiving the pre-processed vehicle-side sensing data, and receiving the pre-processed roadside data; Extracting high-value scene segment data from the roadside data, wherein the high-value scene segment data includes driving trajectory information of multiple dynamic traffic participants in different scenarios, and the driving trajectory information includes location information and speed information of trajectory points; Reconstructing the 3D static road scene based on the vehicle-side perception data, and identifying and extracting the multiple dynamic traffic participants from the vehicle-side perception data; Combining the multiple dynamic traffic participants, the reconstructed 3D static road scene, and the high-value scene fragment data to generate road traffic scene data; For each target dynamic traffic participant in the road traffic scene, road traffic data under multiple perspectives are obtained by switching the perspectives of each target dynamic traffic participant, and an autonomous driving model is trained based on the road traffic data; Test the trained autonomous driving model and deploy the tested autonomous driving model to the intelligent connected vehicle.

9. A computing device, characterized in that The computing device comprises: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method for processing autonomous driving model data based on the vehicle-road-cloud integrated architecture as described in any one of claims 1-5.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, it implements the method for processing autonomous driving model data based on a vehicle-road-cloud integrated architecture as described in any one of claims 1-5.

Images