ADAS intelligent driving real vehicle road test calibration method

Abstract

The invention discloses an ADAS intelligent driving real vehicle road test calibration method, which comprises the following steps: acquiring the driving speed of a driving vehicle in real time, and obtaining the environment information of the driving vehicle; fusing the environment information to obtain a driving three-dimensional environment model; identifying a driving three-dimensional environment model identification calibration quantity; planning a short-distance driving path, and obtaining intelligent driving behavior data and actual data; and comparing the actual data of the driving vehicle with the intelligent driving behavior data, and repeating the steps until the driving vehicle arrives at the destination. Relates to the technical field of automatic driving, and solves the technical problem that the follow-up automatic driving experience is affected due to the fact that calibration quantity recognition is not accurate enough. According to the method, the actual operation of driving the vehicle is compared with the intelligent driving behavior data, the correct condition of the calibration quantity is modified or solidified according to the comparison result, the calibration quantity efficiency can be guaranteed, meanwhile, the accuracy of the calibration quantity is further improved, and the data related to the calibration quantity is continuously adjusted and optimized.

Description

Technical Field

[0001] The present invention belongs to the field of autonomous driving, and specifically is a method for calibrating an ADAS intelligent driving real vehicle road test. Background Art

[0002] With the rapid development of science and technology, intelligent driving technology is gradually becoming an important development direction of the automotive industry. In the current traffic environment, people's demand for traffic safety, travel efficiency and driving comfort is growing. In the research and development of intelligent driving vehicles, road test calibration is an indispensable part. Through accurate measurement, modeling and verification of the actual road environment, road test calibration can provide basic data support for the algorithm development and optimization of the intelligent driving system, help improve the perception ability, decision-making level and safety of the intelligent driving system, and ensure its stability and reliability in various complex traffic scenarios. Traditional driving methods rely on the driver's experience and judgment, but human reaction speed and attention are limited, and they are easily affected by factors such as fatigue and emotions, leading to traffic accidents. The most accurate intelligent driving method at present is the combination of vision and radar. When determining intelligent driving behavior, this method usually uses data collected by vision or radar solutions and combines the calibration quantity in the database to determine the driving path.

[0003] Currently, all high-precision maps are calibrated by pre-calibrating fixed calibration quantities in the map. However, the workload of calibrating the calibration quantities on the driving path is extremely heavy. Although image recognition can reduce the workload of staff, the recognition of the calibration quantities through image calibration is not accurate enough due to other interferences such as occlusion by other vehicles. Incorrect calibration of the calibration quantities will affect the subsequent autonomous driving experience. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes an ADAS intelligent driving real vehicle road test calibration method, which is used to solve the technical problem that the identification of the calibration quantity is not accurate enough and affects the subsequent autonomous driving experience. The present invention solves the above problem by comparing the actual operation of the driving vehicle with the intelligent driving behavior data, and modifying or solidifying the correct situation of the calibration quantity according to the comparison result.

[0005] To achieve the above object, a first aspect of the present invention provides an ADAS intelligent driving real vehicle road test calibration method, comprising the following steps:

[0006] S1: collect the driving speed of the driving vehicle in real time, and obtain environmental information of the driving vehicle's surrounding environment according to the driving speed;

[0007] S2: Fusing the sub-data in the environmental information to obtain a 3D driving environment model; identifying calibration quantities based on several 3D driving environment models; wherein the calibration quantities include fixed obstacles, dynamic obstacles, and traffic signs, specifically traffic signs, lane lines, vehicles, pedestrians, vehicles, and other objects;

[0008] S3: Use the algorithm platform to plan short-distance driving routes, and obtain the intelligent driving behavior data of the driving vehicle based on the short-distance driving routes and calibration quantities;

[0009] S4: Control the driving vehicle to drive according to the intelligent driving behavior data on a short-distance driving route and under the premise of safety, obtain actual data, and judge whether the actual data of the driving vehicle is consistent with the intelligent driving behavior data; if yes, solidify the calibration quantity; if not, adjust the calibration quantity, and input the adjusted calibration quantity into the algorithm platform for optimization processing;

[0010] S5: Repeat steps S1-S5 until the vehicle is driven to the destination.

[0011] By identifying the three-dimensional driving environment model, the calibration quantity can be preliminarily calibrated, and then the intelligent driving behavior data can be obtained based on the collected environmental information. By comparing the data with the actual intelligent driving behavior data, the type of the calibration quantity can be accurately determined. This application reflects the correct situation of the calibration quantity based on the actual operation of the driving vehicle. When the actual data of the driving vehicle conflicts with or differs from the intelligent driving behavior data, it means that there is an error in the calibration quantity. At this time, the calibration quantity near the driving vehicle is adjusted, so as to more realistically reflect whether the calibration quantity meets the restrictions of the calibration quantity.

[0012] Preferably, the driving vehicle is equipped with a plurality of sensors to obtain environmental information; wherein the environmental information includes a plurality of environmental images, a plurality of three-dimensional point cloud data, and distance data and relative speeds between the driving vehicle and a plurality of objects.

[0013] Preferably, the fusing of sub-data in the environmental information includes:

[0014] Using SLAM technology to process the preprocessed environmental images, the three-dimensional point cloud data, the relative speeds and the corresponding distance data to obtain a three-dimensional model;

[0015] Associating objects in the environment image with points in the three-dimensional model based on feature matching technology to obtain a three-dimensional model after image association;

[0016] The driving image recognition model is used to assign labels to a number of objects in the three-dimensional model after the image is associated, so as to obtain a driving three-dimensional environment model including the position and outline of the environment information.

[0017] Preferably, the identification of calibration quantities based on a plurality of driving three-dimensional environment models includes:

[0018] The environmental images and three-dimensional point cloud data corresponding to several environmental sub-models in the three-dimensional driving environment model are input into the driving signal recognition model to obtain the calibration quantity type; wherein the driving signal recognition model is obtained based on the neural network model.

[0019] Preferably, the method of planning a short-distance driving route using an algorithm platform includes:

[0020] The environment information is integrated with the built-in map data using an algorithm platform to obtain three-dimensional map data;

[0021] Mark the current position of the driving vehicle as a source node S, select a target position from the three-dimensional map data and mark it as a terminal node T, and mark the intersections of all routes from S to T as several intermediate nodes;

[0022] Connect two directly reachable nodes under the constraints of a certain number of calibration quantities to form a certain number of edges, and connect the certain edges end to end in sequence to form a graph network; wherein the nodes include a source node S, a terminal node T, and a certain number of intermediate nodes;

[0023] The Dijkstra algorithm is used to traverse the graph network to obtain the best driving path from S to T.

[0024] Obtaining short-distance driving routes in this way meets the actual needs of autonomous driving.

[0025] Preferably, the step of obtaining the intelligent driving behavior data of the driving vehicle according to the short-distance driving route and the calibration amount includes:

[0026] The type of the quantity marked on the initial short-distance path is obtained, and the short-distance driving path is corrected based on the type of the quantity marked on the initial short-distance path to obtain the intelligent driving behavior data.

[0027] It should be noted that the method of correcting the initial short-distance path based on the type of calibration quantity is based on the correction database, and the path is corrected under safety premises during the correction. When there is no corresponding correction method for the calibration quantity or there is a danger, emergency braking is performed; the correction database stores driving methods for several calibration quantity types, and the correction database is set by technical experts in this field, and decisions such as emergency braking are made quickly to ensure the safety of the driving vehicle and staff.

[0028] Preferably, the driving signal recognition model is obtained based on a neural network model, including:

[0029] Extracting three-dimensional point cloud data and image data containing a number of calibration quantities from a database, and annotating the three-dimensional point cloud data and image data corresponding to the calibration quantity type; integrating the annotated three-dimensional point cloud data, image data and calibration quantity type into a number of training data and test data;

[0030] A driving signal recognition model is constructed based on a recurrent neural network and introduces a self-attention mechanism; the constructed driving signal recognition model is trained through a number of training data, and the trained driving signal recognition model is tested through a number of test data to obtain the test accuracy; the trained driving signal recognition model is adjusted according to the test accuracy, and the input is three-dimensional point cloud data and the output is image data as a calibrated quantity type.

[0031] Preferably, the testing of the trained driving signal recognition model by using a plurality of test data includes:

[0032] According to the model accuracy calculation formula: Get the test accuracy PR;

[0033] Where TP is the number of sample data detected correctly, FP is the number of sample data detected incorrectly, λi is the ratio of the calibration quantity corresponding to the i-th correctly detected sample data in the image data to the proportion threshold, and λi≤1, and n=FP.

[0034] It should be noted that the proportion threshold needs to be set according to the type of calibration quantity. When the calculated value of λi is greater than 1, the value of λi is 1.

[0035] In this way, when calculating and checking the accuracy, the calibration amount occupies a relatively low proportion in the image, which may lead to misjudgment. This can improve the accuracy of the acquired driving signal recognition model in identifying the calibration amount type.

[0036] Preferably, the determining whether the actual driving data of the vehicle meets the intelligent driving requirements includes:

[0037] Extract the intelligent driving behavior data and actual data, and compare the actual behavior data with the intelligent driving behavior data; when the actual behavior data and the sub-data in the intelligent driving behavior data are within the threshold interval, the intelligent driving behavior data meets the intelligent driving requirements; when the actual behavior data and the sub-data in the intelligent driving behavior data are outside the threshold interval, the intelligent driving behavior data does not meet the intelligent driving requirements.

[0038] Compared with the prior art, the beneficial effects of the present invention are as follows: according to the environmental information of the driving car environment, the system integrates the environmental information collected by multiple sensors and constructs a model. A good environmental model construction helps to improve the decision-making ability and response speed of the intelligent driving system, improve the overall driving safety, reliability and accuracy of environmental perception, and realize comprehensive environmental modeling. It can improve the accuracy of subsequent calibration, and compare the actual operation of the driving vehicle with the intelligent driving behavior data, and modify or solidify the correct situation of the calibration according to the comparison result. In this way, the calibration efficiency can be guaranteed while further improving the accuracy of the calibration, so as to continuously adjust and optimize the data related to the calibration, avoid changes in the calibration of the lane driving direction and other types of calibration, and avoid intelligent driving misjudgment, thereby improving the intelligent driving behavior to be more accurate in line with expectations, and improving the reliability of the system. By continuously collecting actual operation data, adjusting parameters and optimizing algorithms, rich experience and knowledge are accumulated, the improvement of the intelligent driving system is accelerated, and the safety of intelligent driving is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] 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.

[0040] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0041] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than 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.

[0042] See also Figure 1 The first aspect of the present invention provides an ADAS intelligent driving real vehicle road test calibration method, including the following steps:

[0043] S1: collect the driving speed of the driving vehicle in real time, and obtain environmental information of the driving vehicle's surrounding environment according to the driving speed;

[0044] S2: Fusing the sub-data in the environmental information to obtain a 3D driving environment model; identifying calibration quantities based on several 3D driving environment models; wherein the calibration quantities include fixed obstacles, dynamic obstacles, and traffic signs, specifically traffic signs, lane lines, vehicles, pedestrians, vehicles, and other objects;

[0045] A number of sensors are installed on the driving vehicle to obtain environmental information; the environmental information includes a number of environmental images, a number of three-dimensional point cloud data, and distance data and relative speeds between the driving vehicle and a number of objects. Specifically, the environmental image is obtained through a camera, the three-dimensional point cloud data and the distance data between the driving vehicle and a number of objects are obtained through a lidar sensor or a millimeter-wave radar sensor, and the relative speed between the driving vehicle and a number of objects can be obtained through the distance data and a set cycle time.

[0046] By collecting images, distance information, speed and other data in real time, the driving environment around the vehicle can be fully understood from multiple angles. The images provide rich visual information, including road conditions, traffic signs, the appearance of other vehicles and pedestrians, etc.; distance information enables the system to accurately perceive the relative position of surrounding objects and the vehicle, providing key data for avoiding collisions; speed information reflects the driving status of the vehicle, which helps to plan reasonable driving strategies. The multi-dimensional data collection method provides a more comprehensive and accurate environmental cognition for the intelligent driving system, enabling the system to better cope with various complex driving scenarios. Through the analysis of the environmental model, the system can find the best driving path, avoid obstacles, comply with traffic rules, and take into account factors such as driving efficiency and comfort. Data fusion and environmental model construction make path planning more accurate and efficient, and improve the driving performance of the intelligent driving system.

[0047] In this embodiment, the calibration quantity is identified based on several driving three-dimensional environment models, including:

[0048] Inputting environmental images and three-dimensional point cloud data corresponding to several environmental sub-models in the three-dimensional driving environment model into a driving signal recognition model to obtain a calibration quantity type; wherein the driving signal recognition model is obtained based on a neural network model;

[0049] Specifically, the driving signal recognition model is obtained based on a neural network model, including:

[0050] Extracting three-dimensional point cloud data and image data containing a number of calibration quantities from a database, and annotating the three-dimensional point cloud data and image data corresponding to the calibration quantity type; integrating the annotated three-dimensional point cloud data, image data and calibration quantity type into a number of training data and test data;

[0051] A driving signal recognition model is constructed based on a recurrent neural network and a self-attention mechanism is introduced; the constructed driving signal recognition model is trained through a number of training data, and the trained driving signal recognition model is tested through a number of test data to obtain the test accuracy; the trained driving signal recognition model is adjusted according to the test accuracy, and the input is three-dimensional point cloud data and the output is image data as a calibration type;

[0052] In order to further improve the accuracy of the recognition in this embodiment, the trained driving signal recognition model is tested by using several test data, including:

[0053] According to the model accuracy calculation formula: Get the test accuracy PR;

[0054] Where TP is the number of sample data detected correctly, FP is the number of sample data detected incorrectly, λi is the ratio of the calibration quantity corresponding to the i-th correctly detected sample data in the image data to the proportion threshold, and λi≤1, and n=FP.

[0055] It should be noted that the proportion threshold needs to be set according to the type of calibration quantity. When the calculated value of λi is greater than 1, the value of λi is 1.

[0056] S3: Use the algorithm platform to plan short-distance driving routes, and obtain the intelligent driving behavior data of the driving vehicle based on the short-distance driving routes and calibration quantities;

[0057] In this step, the algorithm platform is used to plan short-distance driving routes, including:

[0058] The environment information is integrated with the built-in map data using an algorithm platform to obtain three-dimensional map data;

[0059] Mark the current position of the driving vehicle as a source node S, select a target position from the three-dimensional map data and mark it as a terminal node T, and mark the intersections of all routes from S to T as several intermediate nodes;

[0060] Connect two directly reachable nodes under the constraints of a certain number of calibration quantities to form a certain number of edges, and connect the certain edges end to end in sequence to form a graph network; wherein the nodes include a source node S, a terminal node T, and a certain number of intermediate nodes;

[0061] The Dijkstra algorithm is used to traverse the graph network to obtain the best driving path from S to T.

[0062] Obtaining short-distance driving routes in this way meets the actual needs of autonomous driving.

[0063] Furthermore, the intelligent driving behavior data of the driving vehicle is obtained based on the short-distance driving route and the calibration quantity, including:

[0064] The type of the quantity marked on the initial short-distance path is obtained, and the short-distance driving path is corrected based on the type of the quantity marked on the initial short-distance path to obtain the intelligent driving behavior data.

[0065] S4: Control the driving vehicle to drive according to the intelligent driving behavior data on a short-distance driving route and under the premise of safety, obtain actual data, and judge whether the actual data of the driving vehicle is consistent with the intelligent driving behavior data; if yes, solidify the calibration quantity; if not, adjust the calibration quantity, and input the adjusted calibration quantity into the algorithm platform for optimization processing;

[0066] S5: Repeat steps S1-S5 until the vehicle is driven to the destination.

[0067] In this embodiment, the collected actual data is compared with the expected behavior of the intelligent driving behavior data. The intelligent driving behavior data is an ideal behavior mode determined according to the goal of intelligent driving and various traffic rules, safety standards, etc., for example, a reasonable speed range under specific road conditions, a safe following distance, and correct lane keeping behavior. Based on the comparison results, it is evaluated whether the actual intelligent driving behavior of the vehicle meets the requirements of intelligent driving;

[0068] According to the performance requirements and safety standards of intelligent driving, a threshold range for each type of calibration quantity is set. This range is an acceptable range, which is used to judge whether the intelligent driving behavior is within a reasonable range. For example, intelligent driving sets a specific speed range as a threshold range on speed-limited sections.

[0069] When the vehicle is in intelligent driving operation, the actual data is monitored in real time. If the actual data is within the set threshold range, it means that the current intelligent driving behavior is relatively ideal. At this time, the current calibration quantity and the parameters corresponding to the calibration quantity are recorded and solidified. If the behavior data is not within the set threshold range, the reason for the difference needs to be analyzed;

[0070] An exemplary explanation of this parameter is: when the calibration amount is an obstacle, the parameter is the distance and the driving speed of the vehicle.

[0071] The reasons for the difference may come from many aspects. For example, sensor failure or inaccuracy leads to data deviation, the algorithm is not adaptable enough to specific scenarios, and environmental changes (such as bad weather, road construction, etc.) affect intelligent driving behavior. Through in-depth analysis of the data and observation of the vehicle operating environment, the specific reasons that cause the behavior data to exceed the threshold range are determined. According to the cause of the difference, the automatic optimization algorithm is used, and the parameters corresponding to the adjusted calibration are re-entered into the algorithm platform for optimization. The algorithm platform will recalculate and adjust the intelligent driving behavior according to the new parameters to make the vehicle behavior closer to the expected behavior, and then collect and compare data again. The evaluation is repeated until the intelligent driving behavior meets the requirements and its calibration is solidified.

[0072] In another embodiment, it also includes S6: continuously monitoring the status of the vehicle and the surrounding environment, making dynamic adjustments, and adjusting the control strategy according to the sensor data and the actual status of the vehicle to cope with emergencies or environmental changes until the vehicle is driven to the destination.

[0073] Specifically, the vehicle status and surrounding environment are continuously monitored, and dynamic adjustments are made. The control strategy is adjusted according to the sensor data and the actual status of the vehicle, including:

[0074] S6.1. Continuously monitor the vehicle status and surrounding environment, and adjust the intelligent driving control strategy in real time according to emergencies;

[0075] S6.2. Continuously optimize the control strategy and parameters corresponding to the calibration quantity through system monitoring data and actual driving feedback, and iteratively update the system to maintain the best intelligent driving performance.

[0076] Utilize various sensors installed on the vehicle to obtain information about the surrounding environment, including road conditions, traffic flow, and obstacles, and identify possible emergencies. For example, if rainy or foggy weather occurs during intelligent driving, once an emergency is identified, the system will immediately adjust the intelligent driving control strategy according to the situation. For example, when rainy or foggy weather comes, the sensor recognition rate decreases. By adjusting and slowing down the intelligent driving speed, the safety of the vehicle is ensured and the impact on passengers is minimized. The system continuously collects sensor data and actual vehicle driving feedback. The sensor data provides real-time information about the vehicle status and surrounding environment, while the actual driving feedback can include passenger feelings, vehicle performance, etc. By analyzing these data, the system can understand the effectiveness and shortcomings of the current control strategy. Based on the data analysis results, the system continuously optimizes the control strategy. The process of continuous optimization and updating is an iterative process, providing a more intelligent, efficient, and safe driving experience.

[0077] Continuously optimizing the control strategy can improve the vehicle's driving performance and efficiency. By adjusting speed control parameters, steering sensitivity, etc., the vehicle can be driven more smoothly and efficiently, reducing energy consumption and emissions. At the same time, the optimized control strategy can also improve the vehicle's response speed and maneuverability, and enhance driving pleasure and comfort.

[0078] By continuously optimizing system parameters and iteratively updating the system, new technologies and algorithms can be continuously introduced to improve the intelligence level of the system. When encountering emergencies, the system can quickly make decisions such as emergency braking, steering or acceleration based on real-time data in the environmental model to ensure the safety of the vehicle.

[0079] Part of the data in the above formula is calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0080] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for calibrating an ADAS intelligent driving real vehicle road test, characterized in that: The following steps are involved: S1: collect the driving speed of the driving vehicle in real time, and obtain environmental information of the driving vehicle's surrounding environment according to the driving speed; S2: Fusing the sub-data in the environmental information to obtain a three-dimensional driving environment model; identifying calibration quantities based on several three-dimensional driving environment models; wherein the calibration quantities include fixed obstacles, dynamic obstacles and traffic signs; S3: Plan short-distance driving routes using the algorithm platform, and obtain intelligent driving behavior data of the driving vehicle based on the short-distance driving routes and calibration quantities; control the driving vehicle to drive on the short-distance driving routes and under the premise of safety according to the intelligent driving behavior data to obtain actual data; S4: Determine whether the actual data of the driving vehicle is consistent with the intelligent driving behavior data; if yes, solidify the calibration quantity; if no, adjust the calibration quantity; S5: Repeat steps S1-S5 until the vehicle is driven to the destination.

2. The ADAS intelligent driving real vehicle road test calibration method according to claim 1, characterized in that: The driving vehicle is equipped with a plurality of sensors to obtain environmental information; wherein the environmental information includes a plurality of environmental images, a plurality of three-dimensional point cloud data, and distance data and relative speed between the driving vehicle and a plurality of objects.

3. The ADAS intelligent driving real vehicle road test calibration method according to claim 2, characterized in that: The fusing of sub-data in the environmental information includes: Using SLAM technology to process the preprocessed environmental images, the three-dimensional point cloud data, the relative speeds and the corresponding distance data to obtain a three-dimensional model; Associating objects in the environment image with points in the three-dimensional model based on feature matching technology to obtain a three-dimensional model after image association; The driving image recognition model is used to assign labels to a number of objects in the three-dimensional model after the image is associated, so as to obtain a driving three-dimensional environment model including the position and outline of the environment information.

4. The ADAS intelligent driving real vehicle road test calibration method according to claim 2, characterized in that: The identification of calibration quantities based on a plurality of driving three-dimensional environment models includes: The environmental images and three-dimensional point cloud data corresponding to several environmental sub-models in the three-dimensional driving environment model are input into the driving signal recognition model to obtain the calibration quantity type; wherein the driving signal recognition model is obtained based on the neural network model.

5. The ADAS intelligent driving real vehicle road test calibration method according to claim 1, characterized in that: The method of planning a short-distance driving route by using an algorithm platform includes: The environment information is integrated with the built-in map data using an algorithm platform to obtain three-dimensional map data; Mark the current position of the driving vehicle as a source node S, select a target position from the three-dimensional map data and mark it as a terminal node T, and mark the intersections of all routes from S to T as several intermediate nodes; Connect two directly reachable nodes under the constraints of a number of calibration quantities to form a number of edges, and connect the edges end to end in sequence to form a graph network; wherein the nodes include a source node S, a terminal node T, and a number of intermediate nodes; The Dijkstra algorithm is used to traverse the graph network to obtain the best driving path from S to T.

6. The ADAS intelligent driving real vehicle road test calibration method according to claim 2, characterized in that: The driving signal recognition model is obtained based on a neural network model, including: Extracting three-dimensional point cloud data and image data containing a number of calibration quantities from a database, and annotating the three-dimensional point cloud data and image data corresponding to the calibration quantity type; integrating the annotated three-dimensional point cloud data, image data and calibration quantity type into a number of training data and test data; A driving signal recognition model is constructed based on a recurrent neural network and introduces a self-attention mechanism; the constructed driving signal recognition model is trained through a number of training data, and the trained driving signal recognition model is tested through a number of test data to obtain the test accuracy; the trained driving signal recognition model is adjusted according to the test accuracy, and the input is three-dimensional point cloud data and the output is image data as a calibrated quantity type.

7. The ADAS intelligent driving real vehicle road test calibration method according to claim 1, characterized in that: The testing of the trained driving signal recognition model by using a plurality of test data includes: According to the model accuracy calculation formula: Get the inspection accuracy PR; Where TP is the number of sample data detected correctly, FP is the number of sample data detected incorrectly, λi is the ratio of the calibration quantity corresponding to the i-th correctly detected sample data in the image data to the proportion threshold, and λi≤1, and n=FP.

8. The ADAS intelligent driving real vehicle road test calibration method according to claim 1, characterized in that: The determining whether the actual data of the driving vehicle meets the intelligent driving requirements includes: Extract the intelligent driving behavior data and actual data, and compare the actual behavior data with the intelligent driving behavior data; when the actual behavior data and the sub-data in the intelligent driving behavior data are within the threshold interval, the intelligent driving behavior data meets the intelligent driving requirements; when the actual behavior data and the sub-data in the intelligent driving behavior data are outside the threshold interval, the intelligent driving behavior data does not meet the intelligent driving requirements.

Images