A method for evaluating the running safety of a lane keeping assist driving function

Abstract

A method for evaluating the running safety of a lane keeping assist driving function provided by the present invention performs multi-source data acquisition through multiple on-vehicle cameras as data acquisition devices, constructs a forward lane detection model and a front wheel crossing line classification model. Based on the data collected by the data acquisition device through the forward lane detection model, it judges the physical relationship between the front wheels of the vehicle to be evaluated and the lane lines. Through the front wheel crossing line classification model, it judges the types of the front wheels on both sides of the vehicle to be evaluated crossing the lane lines. From the two perspectives of the physical relationship between the front wheels of the vehicle and the lane lines and the types of the front wheels on both sides of the vehicle crossing the lane lines, it simultaneously judges the driving behavior of the vehicle to be evaluated, avoiding the occurrence of misjudgment problems, ensuring that the driving behavior of the vehicle can be accurately judged, and further ensuring the accuracy of the judgment on the recognition result of the lane assist driving system to the vehicle running safety risk behavior for the vehicle to be evaluated.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent traffic control, and specifically provides a method for evaluating the operation safety of a lane keeping assist driving function. Background Art

[0002] Autonomous driving cars have become one of the development trends of the modern automotive industry, and advanced driving assistance systems have been gradually applied in mass-produced cars. Advanced Driving Assistance Systems (ADAS) are used in autonomous driving cars, which utilize various sensors installed on the vehicle (millimeter-wave radar, lidar, mono / stereo cameras, and satellite navigation) to sense the surrounding environment at any time during vehicle driving, collect data, identify, detect, and track static and dynamic objects, and combine navigation map data for system operation and analysis, so as to pre-warn the driver of possible dangers. The Lane Keeping System (LKS) is an important system function of autonomous driving cars. The lane keeping system assists in steering control to keep the vehicle driving in the center of the lane, which can prevent the vehicle from deviating from its own lane and effectively improve the active safety of the car.

[0003] However, due to the lack of effective evaluation and supervision methods for the operation safety of the lane keeping assist driving function during its operation on public roads, in practical applications, potential road traffic safety risks caused by the use of autonomous driving cars are gradually emerging. Summary of the Invention

[0004] In order to solve the problem of the lack of effective evaluation and supervision methods for the operation safety of the lane keeping assist driving function during its operation on public roads in the prior art, the present invention provides a method for evaluating the operation safety of a lane keeping assist driving function, which can evaluate the operation safety of a vehicle operating based on the lane keeping assist driving function, effectively improve the practicality of the lane keeping assist driving function, and provide technical support for the safe driving of the car.

[0005] The technical solution of the present invention is as follows: A method for evaluating the operation safety of a lane keeping assist driving function, characterized by comprising the following steps:

[0006] S1: Denote the vehicle installed with the lane assist driving system to be evaluated as the vehicle to be evaluated;

[0007] Install a data acquisition device for collecting evaluation data on the vehicle to be evaluated;

[0008] The evaluation data includes: vehicle forward lane data and vehicle side lane line data;

[0009] The data acquisition device includes: a vehicle front camera and wireless cameras on both sides of the vehicle; the vehicle forward lane data is collected based on the vehicle front camera; the lane line data on both sides of the vehicle is collected based on the wireless cameras on both sides of the vehicle;

[0010] S2: Construct a lane keeping operation safety fusion judgment model;

[0011] The lane keeping operation safety fusion judgment model includes: a forward lane detection model, a front wheel pressing line classification model, and a time synchronization module;

[0012] The input of the forward lane detection model is the vehicle forward lane data, and the outputs are lane line type, lane line position, and the distance from the inner side of the lane line to the vehicle longitudinal center line;

[0013] The input of the front wheel pressing line classification model is the lane data on both sides of the vehicle, and the output is the front wheel pressing line categories on both sides of the vehicle;

[0014] The time synchronization module is used to synchronize the time of the data collected by a total of 3 cameras, namely the vehicle front camera and the wireless cameras on both sides of the vehicle;

[0015] The lane line types include: yellow dotted line, yellow solid line, white dotted line, white solid line;

[0016] The front wheel pressing line categories on both sides of the vehicle include: left front wheel pressing the line, left front wheel not pressing the line, right front wheel pressing the line, right front wheel not pressing the line;

[0017] S3: Train the forward lane detection model and the front wheel pressing line classification model based on historical data to obtain the trained forward lane detection model and the front wheel pressing line classification model;

[0018] S4: Operate the vehicle to be evaluated, and obtain the evaluation data corresponding to the vehicle to be evaluated based on the data acquisition device, denoted as: to-be-evaluated data;

[0019] S5: Input the to-be-evaluated data into the trained forward lane detection model and the front wheel pressing line classification model, and denote the output results of the two models as: driving behavior data of the vehicle to be evaluated;

[0020] S6: Identify based on the driving behavior data of the vehicle to be evaluated. When the identification result includes any of the following results, it is determined that the identification result of the lane assist driving system of the vehicle to be evaluated for vehicle operation safety risk behaviors is unqualified;

[0021] Pressing the left white solid line: The forward lane detection model outputs that the lane line is a white solid line, the distance from the inner side of the lane line to the vehicle longitudinal center line is less than half of the vehicle width, and the front wheel pressing line classification model identifies that the left front wheel presses the line;

[0022] Press the right solid white line: The forward lane detection model outputs that the lane line is a solid white line, the distance from the inner side of the lane line to the vehicle's longitudinal center line is less than half of the vehicle width, and the front wheel line pressing classification model recognizes that the right front wheel presses the line;

[0023] Press the left solid yellow line: The forward lane detection model outputs that the lane line is a solid yellow line, the distance from the inner side of the lane line to the vehicle's longitudinal center line is less than half of the vehicle width, and the front wheel line pressing classification model recognizes that the left front wheel presses the line;

[0024] Press the right solid yellow line: The forward lane detection model outputs that the lane line is a solid yellow line, the distance from the inner side of the lane line to the vehicle's longitudinal center line is less than half of the vehicle width, and the front wheel line pressing classification model recognizes that the right front wheel presses the line;

[0025] Ride on the line: The forward lane detection model outputs that the distance from the inner side of the lane line to the vehicle's longitudinal center line is less than one-third of the vehicle width and lasts for 3 seconds.

[0026] Its further feature lies in that:

[0027] The time synchronization method implemented by the time synchronization module includes the following steps:

[0028] a1: The vehicle front camera and the wireless cameras on both sides of the vehicle are respectively communicatively connected to the industrial control host;

[0029] a2: After power-on, the time synchronization module uses the industrial control host time as a reference to calibrate the vehicle front camera and the wireless cameras on both sides of the vehicle respectively;

[0030] a3: The three cameras start working simultaneously, and record the time corresponding to the industrial control host when the work starts, denoted as: start time;

[0031] a4: The three cameras collect picture data in real time and continuously at the same frame rate, and generate time stamps for the collected pictures respectively;

[0032] The time stamp format is: camera number - frame number - shooting time;

[0033] Among them, in the time stamp, the format of the shooting time is: year - month - day - hour: minute: second.microsecond;

[0034] The frame number is the sequential number of each picture taken by each camera after power-on;

[0035] a5: Each camera transmits the collected picture data to the industrial control host in real time;

[0036] The time synchronization module respectively obtains the picture data collected by each camera in real time and obtains the frame number of each picture;

[0037] a6: Confirm the theoretical shooting time corresponding to the frame number;

[0038] The theoretical shooting time = startup time + frame number / frame rate

[0039] a7: Calculate the difference between the shooting time of each picture data and the theoretical shooting time, denoted as: calibration difference;

[0040] Compare the absolute value of the calibration difference with a preset time threshold, and mark the camera whose absolute value of the calibration difference is greater than the time threshold as: camera to be calibrated;

[0041] a8: Correct the time of the camera to be calibrated to the time of the industrial control host;

[0042] The frame rates of the vehicle front camera and the wireless cameras on both sides of the vehicle are both 30fps;

[0043] The time threshold is: 33ms;

[0044] In step a5, compare the picture frame numbers corresponding to the three cameras in the received pictures in real time. If there is a phenomenon of missing frame numbers or stopped frame numbers in the pictures of any camera, restart the three cameras simultaneously and correct the time to the time of the industrial control host at the same time;

[0045] The installation method of the data acquisition device specifically includes the following content:

[0046] The vehicle front camera is installed inside the front windshield of the vehicle, and camera parameters are calibrated and distortion correction are obtained through a calibration method;

[0047] The wireless cameras on both sides of the vehicle are installed outside the left and right sides of the vehicle and can clearly capture the left and right front wheels of the vehicle and the lane lines;

[0048] The data acquisition device is implemented based on wireless cameras and is adsorbed on the vehicle by a magnetic adsorption method or a pneumatic adsorption method;

[0049] An industrial control host also needs to be installed on the vehicle to be evaluated. The industrial control host is installed at the windshield rearview mirror, has high-performance deep learning image processing capabilities, and can also perform data communication and image algorithm processing; the industrial control host communicates with the data acquisition device wirelessly.

[0050] A method for evaluating the running safety of a lane keeping assist driving function provided by the present invention collects multi-source data through multiple on-vehicle cameras as data collection devices, constructs a forward lane detection model and a front wheel pressing line classification model. Based on the data collected by the data collection device through the forward lane detection model, it judges the physical relationship between the front wheels of the vehicle to be evaluated and the lane lines. Through the front wheel pressing line classification model, it judges the types of the front wheels on both sides of the vehicle to be evaluated pressing the lane lines. From the two perspectives of the physical relationship between the front wheels of the vehicle and the lane lines and the types of the front wheels on both sides of the vehicle pressing the lane lines, it simultaneously judges the driving behavior of the vehicle to be evaluated, avoiding the occurrence of misjudgment problems, ensuring that the driving behavior of the vehicle can be accurately judged, and further ensuring the accuracy of the judgment of the recognition result of the lane assist driving system for the vehicle running safety risk behavior; based on the judgment result of this method, the practicability of the lane keeping assist driving function can be effectively improved, providing technical guarantee for the safe driving of the vehicle. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the lane keeping operation safety fusion judgment model provided by the present invention;

[0052] Figure 2 It is a schematic diagram of the structure of the lane keeping assist driving function operation safety evaluation system provided by the present invention;

[0053] Figure 3 It is an embodiment of the vehicle forward lane data collected by the data collection device. DETAILED DESCRIPTION OF THE INVENTION

[0054] As Figure 1 shown, the present invention includes a method for evaluating the running safety of a lane keeping assist driving function, which includes the following steps.

[0055] S1: Denote the vehicle installed with the lane assist driving system to be evaluated as the vehicle to be evaluated;

[0056] Install a data collection device for collecting evaluation data on the vehicle to be evaluated, and construct an in-vehicle system for evaluating the running safety of the lane keeping assist driving function on the vehicle to be evaluated.

[0057] The in-vehicle system includes a data collection device, an industrial control host, a communication module, a battery module, and a storage module.

[0058] The evaluation data collected by the data collection device includes: vehicle forward lane data and vehicle side lane line data;

[0059] The data acquisition device includes: a vehicle front camera and two wireless cameras on both sides of the vehicle; the vehicle forward lane data is collected based on the vehicle front camera; the lane line data on both sides of the vehicle is collected based on the two wireless cameras on both sides of the vehicle. The vehicle front camera is installed inside the vehicle front windshield, and the camera parameters are calibrated and the distortion is corrected through a calibration method; the two wireless cameras on both sides of the vehicle are installed outside the left and right sides of the vehicle and can clearly capture the left and right front wheels of the vehicle and the lane lines.

[0060] When specifically implemented, as Figure 2 shown, the data acquisition device includes 1 front camera and 2 wireless cameras. The front camera is integrated with the industrial control host and is used to obtain the vehicle forward lane data; the wireless cameras are magnetically or pneumatically adsorbed on the upper side of the left / right front wheels of the vehicle and communicate with the industrial control host through communication methods such as 4G, and are used to obtain the lane line data on both sides of the vehicle. The specific installation method and installation position of the cameras can be adjusted adaptively based on the existing technology and the specific shape of the vehicle to be evaluated.

[0061] The industrial control host is installed at the windshield rearview mirror, has high-performance deep learning image processing capabilities, and can simultaneously perform data communication and image algorithm processing; in the method of the present application, the lane keeping operation safety fusion judgment model is constructed in the industrial control host, and the recognition process of the vehicle operation safety risk behavior by the lane assist driving system to be evaluated is realized based on the computing power of the industrial control host;

[0062] The communication module is integrated in the industrial control host and is used to communicate with the wireless camera;

[0063] The battery module is integrated in the industrial control host and is used to supply power to all devices in the vehicle system. The storage module is integrated in the industrial control host and is used to store the collected image data and evaluation result data.

[0064] S2: Construct a lane keeping operation safety fusion judgment model;

[0065] The lane keeping operation safety fusion judgment model includes: a forward lane detection model, a front wheel lane line crossing classification model, and a time synchronization module;

[0066] The input of the forward lane detection model is the vehicle forward lane data in the form of pictures, and the output is the lane line type, lane line position, and the distance from the inside of the lane line to the vehicle longitudinal center line;

[0067] The input of the front wheel lane line crossing classification model is the lane data on both sides of the vehicle, and the output is the lane line crossing category of the front wheels on both sides of the vehicle;

[0068] The time synchronization module is used to synchronize the data collected by a total of 3 cameras, namely the vehicle front camera and the two wireless cameras on both sides of the vehicle;

[0069] Lane line types include: yellow dashed line, yellow solid line, white dashed line, white solid line;

[0070] The categories of the front wheels of the vehicle on both sides pressing the line include: the left front wheel pressing the line, the left front wheel not pressing the line, the right front wheel pressing the line, and the right front wheel not pressing the line.

[0071] In this embodiment, as Figure 2 shown, the lane keeping operation safety fusion judgment model identifies the operation safety risk behaviors of the lane keeping function according to the output results of the forward lane detection model and the front wheel pressing line classification model. In the prior art, most of them only judge the vehicle driving behavior through the image relationship between the front wheels of the vehicle and the lane lines on a single type of picture. However, in actual work, abnormalities such as camera angle or foreign object occlusion may occur, resulting in misjudgment. Compared with the prior art, in this method, the relationship between the vehicle and the lane lines is judged from two angles at the same time. Only when the output results of both models are consistent can the vehicle driving behavior be judged, which greatly improves the accuracy of judging the vehicle driving behavior.

[0072] The time synchronization method implemented by the time synchronization module includes the following steps:

[0073] a1: The front vehicle camera and the wireless cameras on both sides of the vehicle are respectively communicatively connected to the industrial control host;

[0074] a2: After power-on, the time synchronization module uses the time of the industrial control host as a reference to calibrate the front vehicle camera and the wireless cameras on both sides of the vehicle respectively;

[0075] a3: The three cameras start working simultaneously, and record the time corresponding to the industrial control host when the three cameras start working, denoted as: start time;

[0076] a4: The three cameras collect picture data in real time at the same frame rate without interruption, and generate time stamps for the collected pictures respectively;

[0077] The time stamp format is: camera number - frame number - shooting time;

[0078] Among them, in the time stamp, the format of the shooting time is: year - month - day - hour: minute: second.microsecond;

[0079] The frame number is the sequential number of each picture taken by each camera after power-on;

[0080] a5: Each camera transmits the collected picture data to the industrial control host in real time;

[0081] The time synchronization module respectively obtains the picture data collected by each camera in real time, and obtains the frame number of each picture;

[0082] a6: Confirm the theoretical shooting time corresponding to the frame number;

[0083] The theoretical shooting time = startup time + frame number / frame rate

[0084] a7: Calculate the difference between the shooting time of each picture data and the theoretical shooting time, denoted as: calibration difference;

[0085] Compare the absolute value of the calibration difference with a preset time threshold, and mark the camera whose absolute value of the calibration difference is greater than the time threshold as: camera to be calibrated;

[0086] a8: Correct the time of the camera to be calibrated to the time of the industrial control host.

[0087] Assume: The timestamp format corresponding to the vehicle front camera is: cameraF - frameF - timeF; the timestamp format corresponding to the vehicle left wireless camera is: cameraL - frameL - timeL; the timestamp format corresponding to the vehicle right wireless camera is cameraR - frameR - timeR; the startup time is: TIME; the frame rates of the three cameras are f;

[0088] Then, the timestamps of the pictures corresponding to the vehicle front camera are in turn: cameraF - 1 - timeF1, cameraF - 2 - timeF2,....., cameraF - N - timeFN, where N is the frame number and takes positive integer values;

[0089] The timestamp corresponding to the vehicle left wireless camera is: cameraL - N - timeLN;

[0090] The timestamp format corresponding to the vehicle right wireless camera is cameraR - N - timeRN;

[0091] The time synchronization module respectively obtains the picture data collected by each camera in real time. timeF, timeL, and timeR are the times of the cameras themselves. Theoretically, they should be the same and should all be consistent with the theoretical shooting time. However, due to some unexpected situations, such as calculation errors in software or hardware errors in the clock chip itself, the timestamps of the cameras are incorrect, resulting in different shooting times for pictures with the same frame number; since the startup times of the three are the same, that is, the shooting times of the first-frame pictures are the same, so the theoretical time corresponding to the frame number is first calculated based on the startup time,

[0092] The theoretical shooting time = startup time + frame number / frame rate = TIME + N / f;

[0093] Compare timeF, timeL, and timeR with the theoretical shooting time respectively to obtain the calibration differences, and then compare the absolute values of the calibration differences with a preset time threshold to determine whether it is a reasonable error;

[0094] Among them, regarding the time threshold, it is set according to the calculation accuracy requirements of the system. In this method, for real-time shooting, it is defined that the pictures taken within one second are regarded as real-time shooting pictures. If the calibration difference of the pictures is less than the time threshold, it is considered a reasonable error and will not affect the accuracy of the subsequent calculation results, and there is no need to correct the camera.

[0095] In this embodiment, the frame rates f of the vehicle front camera and the wireless cameras on both sides of the vehicle are both 30fps; then,

[0096] The time threshold is set to: 1s / 30fps = 33ms.

[0097] At the same time, the time synchronization module compares the latest frame numbers of the three cameras received in real time. If there is a missing frame number or a stopped frame number in the pictures of any camera, it means that the camera may have a hardware failure resulting in the stop of image acquisition, or problems such as packet loss caused by an error in the data transmission module. At this time, the three cameras need to be restarted simultaneously, and the time needs to be corrected to the time of the industrial control host to ensure that the cameras can work properly.

[0098] This application uses the time synchronization module to achieve time synchronization for the three cameras, meets the real-time multi-angle image acquisition requirements, and ensures the accuracy of the calculation results based on multi-source data. Specifically, in this method, the time of the industrial control host is used as the baseline to calibrate the three cameras. When calculating the shooting time field in the timestamp for picture calculation, as long as it is ensured that the pictures taken by the three cameras are taken in the same time period, pictures taken from different angles at the same time point can be obtained, meeting the subsequent calculation requirements based on the picture recognition results; there is no need to synchronize the time of the cameras with other servers, ensuring efficient time synchronization of the cameras and reducing the system complexity.

[0099] S3: Train the forward lane detection model and the front wheel line-pressing classification model based on historical data to obtain the trained forward lane detection model and front wheel line-pressing classification model.

[0100] When training the vehicle forward lane detection model:

[0101] Collect 10,000 pieces of forward lane data of vehicles with a resolution of 1028*720, annotate the lane line pixels and types, and divide the annotated data into a training set, a test set, and a validation set according to the ratio of 7:2:1. Train the forward lane detection model to optimize its performance on the validation set. The lane line type and lane line pixels can be directly identified and output in real time through the trained forward lane detection model.

[0102] During specific implementation, the trained forward lane detection model inputs the forward lane data collected by the front camera, detects and outputs the lane line type and lane line pixels in real time, and calculates the distance from the inner side of the lane line to the longitudinal center line of the vehicle based on the internal and external parameters of the camera and the monocular camera optical geometry method.

[0103] When training the front wheel pressing line classification model:

[0104] Collect 5,000 pieces of lane data on both sides of the vehicle with a resolution of 640*480, annotate the pictures, and divide the annotated data into a training set, a test set, and a validation set according to the ratio of 7:2:1. Train the front wheel pressing line classification model to optimize its performance on the validation set. The front wheel pressing line categories on both sides of the vehicle can be directly identified and output in real time through the trained front wheel pressing line classification model.

[0105] Collect the forward lane data of the vehicle, annotate the lane line pixels and types, and divide the annotated data into a training set, a test set, and a validation set according to the ratio of 7:2:1. Train the forward lane detection model to optimize its performance on the validation set. The lane line type and lane line pixels can be directly identified and output in real time through the trained forward lane detection model.

[0106] S4: Run the vehicle to be evaluated, obtain the evaluation data corresponding to the vehicle to be evaluated based on the data acquisition device, and record it as: the data to be evaluated;

[0107] As Figure 3 shown in the embodiment based on the pictures collected by the wireless camera, it indicates that the left front wheel of the vehicle presses the solid yellow line.

[0108] S5: Input the data to be evaluated into the trained forward lane detection model and the front wheel pressing line classification model, and record the output results of the two models as: the driving behavior data of the vehicle to be evaluated;

[0109] S6: Identify based on the driving behavior data of the vehicle to be evaluated. When the identification result includes any of the following results, it is determined that the identification result of the lane assist driving system of the vehicle to be evaluated for the vehicle operation safety risk behavior is unqualified;

[0110] Press the left solid white line: The forward lane detection model outputs that the lane line is a solid white line, the distance from the inner side of the lane line to the longitudinal center line of the vehicle is less than half of the vehicle width, and the front wheel line pressing classification model identifies that the left front wheel presses the line.

[0111] Press the right solid white line: The forward lane detection model outputs that the lane line is a solid white line, the distance from the inner side of the lane line to the longitudinal center line of the vehicle is less than half of the vehicle width, and the front wheel line pressing classification model identifies that the right front wheel presses the line.

[0112] Press the left solid yellow line: The forward lane detection model outputs that the lane line is a solid yellow line, the distance from the inner side of the lane line to the longitudinal center line of the vehicle is less than half of the vehicle width, and the front wheel line pressing classification model identifies that the left front wheel presses the line.

[0113] Press the right solid yellow line: The forward lane detection model outputs that the lane line is a solid yellow line, the distance from the inner side of the lane line to the longitudinal center line of the vehicle is less than half of the vehicle width, and the front wheel line pressing classification model identifies that the right front wheel presses the line.

[0114] Riding on the line: The forward lane detection model outputs that the distance from the inner side of the lane line to the longitudinal center line of the vehicle is less than one-third of the vehicle width and lasts for 3 seconds.

[0115] Because one of the important functions of the lane keeping assist driving function is to control the vehicle to keep driving within the lane lines, thus preventing collision accidents with vehicles in adjacent lanes. If the vehicle keeps riding on the line, it indicates that there are problems with the performance of the lane keeping function of the tested system. Therefore, in addition to detecting behaviors such as line pressing, it is also necessary to detect "riding on the line" to measure the ability of the lane keeping assist driving function to prevent collisions with vehicles in adjacent lanes.

[0116] In specific implementation, once the output of the forward lane detection model shows that the distance from the inner side of the lane line to the longitudinal center line of the vehicle is less than one-third of the vehicle width, then continuously count the output results corresponding to all pictures within 3s. If all the outputs of the forward lane detection model within 3 seconds have the same problem, it is determined as riding on the line.

[0117] In this embodiment, the camera frame rate is 30fps, that is, 30 pictures per second. If 90 consecutive pictures are all judged that the distance from the inner side of the lane line output by the forward lane detection model to the longitudinal center line of the vehicle is less than one-third of the vehicle width, it is determined that it has lasted for 3 seconds and riding on the line has occurred.

[0118] Based on the vehicle-mounted system provided by the embodiments of the present application, the lane keeping assist driving function operation safety assessment method is run. By collecting the vehicle forward lane data and the lane line data on both sides of the vehicle and inputting them into the industrial control host, the operation safety risk behaviors are identified based on the lane keeping operation safety fusion judgment model, and the collected data and evaluation results are stored in the industrial control host.

[0119] The lane keeping assist driving function operation safety assessment method and in-vehicle system provided by the present invention collect multi-source data through multiple in-vehicle cameras, construct a lane keeping operation safety fusion judgment model, and realize the judgment of lane keeping function operation safety risk behaviors such as pressing the left white solid line, pressing the right white solid line, pressing the left yellow solid line, pressing the right yellow solid line, and straddling the line. Then, a rapid, comprehensive, automated, and intelligent evaluation is carried out on the lane assist driving system to be evaluated when driving on public roads, providing technical support for the safe driving of assisted driving vehicles.

Claims

1. A method for evaluating the running safety of a lane keeping assist driving function, characterized in that, It includes the following steps: S1: Denote the vehicle installed with the lane assist driving system to be evaluated as the vehicle to be evaluated; Install a data acquisition device for collecting evaluation data on the vehicle to be evaluated; The evaluation data includes: vehicle forward lane data and vehicle side lane line data; The data acquisition device includes: a vehicle front camera and wireless cameras on both sides of the vehicle; the vehicle forward lane data is collected based on the vehicle front camera; the vehicle side lane line data is collected based on the wireless cameras on both sides of the vehicle; S2: Construct a lane keeping operation safety fusion judgment model; The lane keeping operation safety fusion judgment model includes: a forward lane detection model, a front wheel pressing line classification model, and a time synchronization module; The input of the forward lane detection model is the vehicle forward lane data, and the output is the lane line type, lane line position, and the distance from the inner side of the lane line to the vehicle longitudinal center line; The input of the front wheel pressing line classification model is the vehicle side lane data, and the output is the front wheel pressing line categories on both sides of the vehicle; The time synchronization module is used to synchronize the time of the data collected by a total of 3 cameras, namely the vehicle front camera and the wireless cameras on both sides of the vehicle; The lane line types include: yellow dotted line, yellow solid line, white dotted line, white solid line; The front wheel pressing line categories on both sides of the vehicle include: left front wheel pressing the line, left front wheel not pressing the line, right front wheel pressing the line, right front wheel not pressing the line; S3: Train the forward lane detection model and the front wheel pressing line classification model based on historical data to obtain the trained forward lane detection model and front wheel pressing line classification model; S4: Run the vehicle to be evaluated, and obtain the evaluation data corresponding to the vehicle to be evaluated based on the data acquisition device, denoted as: to-be-evaluated data; S5: Input the to-be-evaluated data into the trained forward lane detection model and front wheel pressing line classification model, and denote the output results of the two models as: driving behavior data of the vehicle to be evaluated; S6: Identify based on the driving behavior data of the vehicle to be evaluated. When the identification result includes any of the following results, it is determined that the identification result of the lane assist driving system to be evaluated for vehicle operation safety risk behavior is unqualified; Pressing the left white solid line: The forward lane detection model outputs that the lane line is a white solid line, the distance from the inner side of the lane line to the vehicle longitudinal center line is less than half of the vehicle width, and the front wheel pressing line classification model identifies that the left front wheel presses the line; Pressing the right white solid line: The forward lane detection model outputs that the lane line is a white solid line, the distance from the inner side of the lane line to the vehicle longitudinal center line is less than half of the vehicle width, and the front wheel pressing line classification model identifies that the right front wheel presses the line; Pressing the left yellow solid line: The forward lane detection model outputs that the lane line is a yellow solid line, the distance from the inner side of the lane line to the vehicle longitudinal center line is less than half of the vehicle width, and the front wheel pressing line classification model identifies that the left front wheel presses the line; Pressing the right yellow solid line: The forward lane detection model outputs that the lane line is a yellow solid line, the distance from the inner side of the lane line to the vehicle longitudinal center line is less than half of the vehicle width, and the front wheel pressing line classification model identifies that the right front wheel presses the line; Straddling driving: The distance from the inner side of the lane line output by the forward lane detection model to the longitudinal center line of the vehicle is less than one-third of the vehicle width and lasts for 3 seconds.

2. The method for evaluating the operation safety of a lane keeping assist driving function according to claim 1, wherein: An industrial control host is installed on the vehicle to be evaluated, and the industrial control host communicates with the data acquisition device wirelessly.

3. The method for evaluating the running safety of a lane keeping assist driving function according to claim 2, characterized in that: The time synchronization module implements the time synchronization method including the following steps: a1: The vehicle front camera and the wireless cameras on both sides of the vehicle are respectively communicatively connected to the industrial control host; a2: After power-on, the time synchronization module uses the industrial control host time as a reference to calibrate the vehicle front camera and the wireless cameras on both sides of the vehicle respectively; a3: The three cameras start working simultaneously, and record the time corresponding to the industrial control host when the work starts, denoted as: start time; a4: The three cameras collect picture data in real time at the same frame rate without interruption, and generate time stamps for the collected pictures respectively; The time stamp format is: camera number - frame number - shooting time; Among them, in the time stamp, the format of the shooting time is: year - month - day - hour: minute: second.microsecond; The frame number is the sequential number of each picture taken by each camera after power-on; a5: Each camera transmits the collected picture data to the industrial control host in real time; The time synchronization module respectively obtains the picture data collected by each camera in real time and obtains the frame number of each picture; a6: Confirm the theoretical shooting time corresponding to the frame number; The theoretical shooting time = start time + frame number / frame rate a7: Calculate the difference between the shooting time of each picture data and the theoretical shooting time, denoted as: calibration difference; Compare the absolute value of the calibration difference with a preset time threshold, and mark the camera with the absolute value of the calibration difference greater than the time threshold as: camera to be calibrated; a8: Correct the time of the camera to be calibrated to the time of the industrial control host.

4. The method for evaluating the operation safety of a lane keeping assist driving function according to claim 3, characterized in that: The frame rates of the vehicle front camera and the wireless cameras on both sides of the vehicle are both 30fps.

5. The method for evaluating the operation safety of a lane keeping assist driving function according to claim 4, wherein: The time threshold is: 33ms.

6. The method for evaluating the operation safety of a lane keeping assist driving function according to claim 3, characterized in that: In step a5, compare the picture frame numbers corresponding to the three cameras in the received pictures in real time. If there is a phenomenon of missing frame numbers or stopped frame numbers in the pictures of any camera, restart the three cameras simultaneously and correct the time to the time of the industrial control host at the same time.

7. The method for evaluating the running safety of a lane keeping assist driving function according to claim 1, characterized in that: The installation method of the data acquisition device specifically includes the following contents: The vehicle front camera is installed inside the vehicle front windshield, and camera parameters are calibrated and distortion correction is performed through a calibration method; The wireless cameras on both sides of the vehicle are installed outside the left and right sides of the vehicle and can clearly capture the left and right front wheels of the vehicle and the lane lines.

8. The method for evaluating the running safety of a lane keeping assist driving function according to claim 1, characterized in that: The data acquisition device is implemented based on wireless cameras and is magnetically or pneumatically adsorbed on the vehicle.

Images