A driving data collection and analysis system and method

Abstract

The application relates to the field of intelligent traffic, in particular to a driving data collection and analysis system and method, which comprises a global multidimensional data collection module, an intelligent fusion perception processing module, a data set and scatter module and a backtracking analysis platform; wherein the global multidimensional data collection module collects vehicle external environment, vehicle self state and driver operation state data in the vehicle running process; the intelligent fusion perception processing module performs real-time fusion, structured processing and labeling on the data, forms true value perception and road label data; the data set and scatter module converges the processed data, stores and manages the data, and distributes the data to a subjective and objective evaluation module, generates subjective and objective evaluation results through artificial label dotting and algorithm rule matching; and the backtracking analysis platform is used for backtracking and evaluating vehicle illegal or risk behaviors in combination with the subjective and objective evaluation results, and generating an analysis report. Therefore, the problems of low perception range, insufficient perception accuracy and single evaluation function in the prior art are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent transportation, in particular to a driving data collection and analysis system and method. BACKGROUND

[0002] Before officially going on the road, intelligent networked vehicles need to be verified for compliance with open road traffic rules, but existing data collection equipment cannot meet the recording requirements of traffic police for test data, and there are problems of missing key data or insufficient precision, which makes it difficult to effectively carry out testing work. The current technical solution mainly faces the following challenges: the range of environmental perception outside the vehicle body is limited and cannot be perceived in real time, existing equipment relies on a single orientation or local sensor, which is difficult to cover the 360° scene around the vehicle, and cannot meet the real-time monitoring needs in a dynamic traffic environment. The precision of the perceived data is insufficient, and the existing solution mainly uses offline fusion processing, and the data precision cannot reach the relative true value. For example, although some solutions can output traffic participant information, they cannot distinguish the type of lane, and it is difficult to support accurate judgment of test results.

[0003] In addition, the existing technology also has the defect of low equipment integration, and the wiring harness is messy and difficult to calibrate, which increases the installation and maintenance cost in actual application. More importantly, the traditional solution lacks real-time test evaluation function combining subjective and objective factors, and cannot simultaneously realize manual experience judgment and automatic rule matching, making it difficult to comprehensively verify the compliance of vehicle behavior. Therefore, there is an urgent need for a system that can realize panoramic perception, high-precision data collection, equipment simplification, and real-time evaluation of subjective and objective factors to promote the progress of traffic rule compliance testing technology. SUMMARY

[0004] The present application provides a driving data collection and analysis system and method to solve the problems of low perception range, insufficient perception precision, and single evaluation function in the prior art.

[0005] The first aspect of the present application provides a driving data collection and analysis system, comprising: a global multi-dimensional data collection module, an intelligent fusion perception processing module, a data set distribution module, and a backtracking analysis platform. The global multi-dimensional data collection module is used to collect data of the external environment of the vehicle, the state of the vehicle itself, and the state of the driver during the operation of the vehicle in real time. The intelligent fusion perception processing module is used to perform real-time fusion processing on the data, and to perform structured processing and labeling on the fusion data to form true value perception and road label data. The data set distribution module is used to gather the processed data, store and manage the data, and distribute the data to the subjective and objective evaluation module to generate subjective and objective evaluation results through manual label marking and algorithm rule matching.

[0006] The backtracking analysis platform is used to combine the subjective and objective evaluation results, perform offline backtracking and evaluation identification on the vehicle's illegal or risky behavior, and generate an analysis report.

[0007] Preferably, the global multi-dimensional data acquisition module comprises a driver operation state acquisition module, a vehicle body outside road and environment detection module, a vehicle body data acquisition module, and an integrated device, wherein the driver operation state acquisition module monitors the driver's steering wheel operation, pedal action state data in real time through an interior view camera and a high-precision inertial navigation system; the vehicle body outside road and environment detection module detects the external traffic participants and road feature data of the vehicle through a surround view camera and a main laser radar; the vehicle body data acquisition module acquires vehicle data through a vehicle body CAN bus and a combined navigation hardware facility to form a vehicle dynamic data set; and the integrated device is used to fix the sensors through a unified standard suction cup support on the roof, thereby simplifying the equipment installation and reducing the calibration difficulty.

[0008] Preferably, the intelligent fusion perception processing module comprises a data fusion perception module and a data structured processing module, wherein the data fusion perception module is used to receive sensor data, process noise reduction through a Gaussian filtering algorithm, and eliminate errors through real-time fusion processing combined with a Kalman filtering algorithm; and the data structured processing module is used to perform data structured processing and data labeling on the fused perception data through an improved BEV-Transformer, and supplement special scene labeling combined with artificial labor to form true value perception and road label data.

[0009] Preferably, the data set scattering module comprises a data online processing and storage module, an artificial subjective evaluation module, and an automatic objective evaluation module, wherein the data online processing and storage module is used to upload and store structured data in real time, and distribute the data to the subjective and objective evaluation modules; the artificial subjective evaluation module is used to artificially label the vehicle violation or risk moment through a vehicle-mounted handheld device, record the suspicious time stamp, combine the professional judgment of traffic police to form a subjective evaluation result; and the automatic objective evaluation module is used to input vehicle driving state data, environmental element data, and road traffic rule library data through a Rete algorithm to perform rule matching, judge in real time whether the vehicle behavior conforms to the traffic rules, and automatically generate an objective evaluation result.

[0010] Preferably, the backtracking analysis platform comprises a picture playback and time stamp matching module, a subjective and objective result fusion evaluation module, and an analysis report generation module, wherein the picture playback and time stamp matching module is used to intercept the pictures before and after the vehicle violation or risk moment through a time stamp indexing algorithm, combine the suspicious time stamp recorded by the label marking module, and perform spatio-temporal backtracking and picture synchronous playback on the event; the subjective and objective result fusion evaluation module is used to analyze the subjective and objective evaluation results through a Cohen's Kappa coefficient and a confusion matrix, and identify the violation; and the analysis report generation module is used to generate a traffic rule compliance analysis report according to the evaluation results.

[0011] The second aspect embodiment of the application provides a driving data collection and analysis method, comprising: acquiring vehicle external environment, vehicle state and driver operation state data; processing and denoising the vehicle external environment, vehicle state and driver operation state data, combining Kalman filtering algorithm to fuse data, processing and labeling the data after fusion through improved BEV-Transforme, and manually supplementing labeling for special scenes to form true value perception data and road label data; according to the perception true value data and the road label data, manually labeling through a handheld device carried by the vehicle, combining professional judgment of traffic police to generate subjective evaluation results, at the same time, using Rete algorithm, combining road traffic rule database data and vehicle design operation conditions to generate objective evaluation results; according to the subjective evaluation results and the objective evaluation results, combining timestamp index algorithm to intercept the front and rear pictures of the vehicle driving at the time of violation or risk, and analyzing and generating a report through Cohen's Kappa coefficient and a confusion matrix.

[0012] Preferably, the Kalman filtering algorithm formula is:

[0013] ;

[0014] ;

[0015] ;

[0016] ;

[0017] wherein, is a state prediction value of the vehicle at time ; is an evolution law of the vehicle state from time to time ; is the best estimated state of the vehicle at time ; is an input matrix; is a control input; is an uncertainty of the state prediction of the vehicle at time ; is an uncertainty of the state prediction of the vehicle at time ; is a transpose matrix of ; is a measurement matrix; is a transpose matrix of ; is an uncertainty in the vehicle motion model; is an adjustment of the predicted state to combine new measurement values; Noise measured by vehicle sensors; For vehicle sensors at all times The measurement results; For vehicles at any time The precise state.

[0018] Preferably, the Rete algorithm formula is:

[0019] ;

[0020] ;

[0021] in, For matching degree; For the first in the road traffic rules database Rules; Indicates the vehicle's time A set of real-time state facts; For joint matching degree; Number of rules that need to be evaluated simultaneously

[0022] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement a vehicle data acquisition and analysis method as described in the above embodiments.

[0023] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a vehicle data acquisition and analysis method as described in the above embodiments.

[0024] Therefore, this application has the following beneficial effects:

[0025] The embodiment of the application detects the external traffic participants and road feature data of the vehicle through the road and environment detection module outside the vehicle body, uses the surround-view camera and the main laser radar to improve the external environment perception range and real-time perception capability of the vehicle body; with the aid of the data fusion perception module, the Gaussian filtering algorithm is used for noise reduction, the Kalman filtering algorithm is combined to eliminate errors, the data structure processing module performs structured processing and labeling on the fusion data through the improved BEV-Transformer, and the artificial special scene labeling is combined to form the true value perception and road label data, improve the perception data precision to support the evaluation test result; the sensors are fixed through the unified standard suction cup support on the roof of the vehicle, which simplifies the equipment installation, improves the equipment integration, reduces the degree of wire clutter and the calibration difficulty; the artificial subjective evaluation module marks the risk moment through the handheld device on the vehicle, and the subjective result is formed in combination with the judgment of the traffic police, the automatic objective evaluation module generates the objective result through the Rete algorithm matching data and the traffic rule library, and the real-time test evaluation capability of the subjective and objective combination is improved. Therefore, the problems of low perception range, insufficient perception accuracy, and single evaluation function in the prior art are solved.

[0026] Additional aspects and advantages of the application will be set forth in part in the description which follows, and in part will become apparent to those skilled in the art upon examination of the following and / or can be learned by practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0027] The above and / or additional aspects and advantages of the application will become apparent and be readily understood by considering the following detailed description, including the accompanying drawings, in which:

[0028] Figure 1 A structural schematic diagram of a driving data collection and analysis system according to an embodiment of the application is shown in FIG. 1;

[0029] Figure 2 A schematic diagram of a global multi-dimensional data collection module according to an embodiment of the application is shown in FIG. 2;

[0030] Figure 3 A schematic diagram of an intelligent networked vehicle roundabout passing test scene according to an embodiment of the application is shown in FIG. 3;

[0031] Figure 4 A schematic diagram of an intelligent fusion perception processing module according to an embodiment of the application is shown in FIG. 4;

[0032] Figure 5 A schematic diagram of a vehicle road test scene according to an embodiment of the application is shown in FIG. 5;

[0033] Figure 6 A schematic diagram of a data set and scatter module according to an embodiment of the application is shown in FIG. 6;

[0034] Figure 7 a schematic diagram of a vehicle urban road test scene according to an embodiment of the present application;

[0035] Figure 8 a schematic diagram of a backtracking analysis platform according to an embodiment of the present application;

[0036] Figure 9 a schematic diagram of a vehicle highway test scene according to an embodiment of the present application;

[0037] Figure 10 a schematic diagram of a driving data collection and analysis system according to an embodiment of the present application;

[0038] Figure 11 a flowchart of a driving data collection and analysis method according to an embodiment of the present application;

[0039] Figure 12 a schematic diagram of identifying expressway driving violations according to an embodiment of the present application;

[0040] Figure 13 a schematic diagram of a driving data collection and analysis method according to an embodiment of the present application;

[0041] Figure 14 a structural schematic diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0043] A driving data collection and analysis system and method are described below with reference to the accompanying drawings. In view of the single evaluation function mentioned in the background art, the present application provides a driving data collection and analysis system. In the driving data collection and analysis system, the vehicle exterior road and environment detection module detects the vehicle exterior traffic participants and road feature data using the surround view camera and the main laser radar to improve the vehicle exterior environment perception range and real-time perception capability. With the aid of the data fusion perception module, the Gaussian filtering algorithm is used for noise reduction, and the Kalman filtering algorithm is used for real-time fusion processing of each sensor data to eliminate errors. The data structure processing module performs structured processing and labeling on the fusion data using the improved BEV-Transformer, and combines artificial supplement of special scene labeling to form true value perception and road label data, improve perception data accuracy to support judgment test results. The sensors are fixed by the unified standard suction cup bracket on the roof of the vehicle, which simplifies the equipment installation, improves the equipment integration, reduces the degree of wire clutter and the difficulty of calibration, and improves the real-time test evaluation capability of the subjective and objective combination. Thus, the problems of low perception range, insufficient perception accuracy, and single evaluation function in the prior art are solved.

[0044] Figure 1 A structural diagram of a driving data collection and analysis system provided by the embodiments of the present application is shown.

[0045] The embodiments of the present application provide a driving data collection and analysis system, which comprises a global multi-dimensional data collection module 100, an intelligent fusion perception processing module 200, a data set scattering module 300, and a backtracking analysis platform 400.

[0046] The global multi-dimensional data collection module 100 is used to collect the vehicle exterior environment, vehicle state and driver operation state data in real time during the vehicle operation. The intelligent fusion perception processing module 200 is used to perform real-time fusion processing on the data, and to perform structured processing and labeling on the fusion data to form true value perception and road label data. The data set scattering module 300 is used to gather the processed data, store and manage the data, and distribute the data to the subjective and objective evaluation modules to generate subjective and objective evaluation results by artificial label dotting and algorithm rule matching. The backtracking analysis platform 400 is used to combine the subjective and objective evaluation results to perform offline backtracking, evaluation and identification on the vehicle violation or risk behavior, and to generate an analysis report.

[0047] It can be understood that, in the embodiments of the present application, the vehicle exterior road and environment detection module detects the external traffic participants and road feature data of the vehicle by using the surround-view camera and the main laser radar, thereby improving the vehicle exterior environment perception range and real-time perception capability; with the aid of the data fusion perception module, the Gaussian filtering algorithm is used for noise reduction, the Kalman filtering algorithm is combined for real-time fusion processing of the sensor data to eliminate errors, the data structure processing module performs structured processing and labeling on the fused data by using the improved BEV-Transformer, and the artificial supplement of special scene labeling is combined to form the true value perception and road label data, thereby improving the perception data precision to support the evaluation test result; the integrated equipment is used for fixing the sensors by the roof unified standard suction cup support, thereby simplifying the equipment installation, improving the equipment integration, and reducing the degree of wire clutter and the calibration difficulty; the artificial subjective evaluation module marks the risk moment by artificial labeling by using the vehicle-mounted handheld device, and the subjective result is formed in combination with the judgment of the traffic police; the automatic objective evaluation module generates the objective result by matching the data and the traffic rule library by using the Rete algorithm, and the real-time test evaluation capability of the combination of the subjective and objective is improved. Thus, the problems of low perception range, insufficient perception precision, and single evaluation function in the prior art are solved.

[0048] In the embodiments of the present application, the full-area multi-dimensional data acquisition module 100 further includes a driver operation state acquisition module, a vehicle exterior road and environment detection module, a vehicle body data acquisition module, and an integrated equipment, as shown in Figure 2

[0049] The driver operation state acquisition module monitors the driver's steering wheel control and pedal action state data in real time by using the interior-view camera and high-precision inertial navigation; the vehicle exterior road and environment detection module detects the external traffic participants and road feature data of the vehicle by using the surround-view camera and the main laser radar; the vehicle body data acquisition module acquires the vehicle's own data by using the vehicle body CAN bus and the combined navigation hardware facility, thereby forming a vehicle dynamic data set; and the integrated equipment is used for fixing the sensors by the roof unified standard suction cup support, thereby simplifying the equipment installation and reducing the calibration difficulty.

[0050] It can be understood that, in the embodiments of the present application, the driver operation state acquisition module monitors the driver's steering wheel control and pedal action state data in real time by using the interior-view camera and high-precision inertial navigation, thereby providing a basis for evaluation; the vehicle exterior road and environment detection module realizes 360° panoramic perception by using the surround-view camera and the main laser radar, thereby detecting the external traffic participants and road feature data of the vehicle and providing comprehensive external environment data; the vehicle body data acquisition module acquires the vehicle's own data by using the vehicle body CAN bus and the combined navigation, thereby improving the evaluation accuracy; and the integrated equipment is used for fixing the sensors by the roof unified standard suction cup support, thereby simplifying the installation, reducing the calibration difficulty, and improving the equipment integration.

[0051] For example, as shown in Figure 3 ​As shown, in the roundabout passing test of a certain intelligent connected vehicle, the global multi-dimensional data acquisition module plays a key role. The driver operation state acquisition module captures the steering angle of the driver's hand on the steering wheel in real time through the interior camera (accuracy ± 1°), and records the steering rate (sampling frequency 100 Hz) through high-precision inertial navigation. When the vehicle enters the roundabout, the module accurately collects 3 sets of key data of the steering wheel rotation during the driver's continuous right turn: initial steering angle 35°, maximum steering angle 120°, and return angle 0°. At the same time, it records the process of the accelerator pedal opening degree from 20% to 5%, which provides a basis for judging the operation standard. The surround view camera (resolution 1920x1080) of the vehicle body outside road and environment detection module identifies 2 social vehicles with a speed of 25 km / h in the roundabout at a distance of 30 m from the roundabout entrance. The main laser radar (scanning radius 150 m) synchronously detects the traffic markings (identification error ≤5 cm) on the edge of the roundabout and a pedestrian (12 m away from the car, moving speed 0.8 m / s) waiting at the entrance. The vehicle body data acquisition module obtains the real-time speed (32 km / h), gear (D gear), and brake pressure (0.3 MPa) of the vehicle entering the roundabout through the vehicle body CAN bus, and generates the trajectory coordinates of the vehicle in the roundabout (positioning error ± 0.3 m) combined with the combined navigation data, forming a complete vehicle dynamic data set. The unified standard suction cup support on the roof of the integrated device stably fixes the above-mentioned sensors, which are suitable for the streamlined roof of the test vehicle. The installation time is only 40 minutes, which is 50% less than the traditional support. When the vehicle is driving at a speed of 60 km / h, the sensor vibration amplitude is controlled within ± 0.5 mm, ensuring stable data acquisition.

[0052] In the embodiments of the present application, the intelligent fusion perception processing module 200 includes: as shown Figure 4 The data fusion perception module and the data structured processing module.

[0053] Among them, the data fusion perception module is used to receive each sensor data, process noise reduction through Gaussian filtering algorithm, and eliminate errors through real-time fusion processing combined with Kalman filtering algorithm; the data structured processing module is used to perform data structured processing and data labeling on the fused perception data through improved BEV-Transformer, and supplement special scene labeling combined with artificial, to form true value perception and road label data.

[0054] It should be noted that the formula of Gaussian filtering algorithm is:

[0055] ;

[0056] ;

[0057] Among them, is the weight value of the distance from the filtering center ; is the distance of the data point from the filter center; is the constant pi; is the standard deviation; is the natural constant; is the Gaussian filter weight value at the coordinate .

[0058] The BEV-Transformer formula is improved as:

[0059] .

[0060] .

[0061] .

[0062] wherein, is the cross-modal attention result; is the sensor data converted bird's eye view feature; is the lidar feature key vector; is the lidar feature value vector; is the dimension of the key vector . is the normalization function; is the transpose operation of the matrix; is the optimized bird's eye view feature; is the original bird's eye view feature data; is the multi-layer perception; is the probability of existence of obstacles or other objects in the area . is the activation function; is the function for calculating the area and the bird's eye view feature occupancy; is the value of the optimized bird's eye view feature at the coordinate .

[0063] It can be understood that the data fusion perception module of the embodiments of the present application can effectively eliminate errors and improve data accuracy and consistency by real-time fusion of sensor data through Gaussian filtering and Kalman filtering; the data structured processing module can perform structured processing and labeling on the fused data by using the improved BEV-Transformer, and can form standardized true value perception and road label data by combining artificial supplement of special scene labeling, so as to provide high-quality and standardized data basis for subsequent subjective and objective evaluation of the data set dispersion module, violation evaluation of the backtracking analysis platform, etc., and to ensure efficient operation and result reliability of the entire driving data acquisition and analysis system.

[0064] For example, as Figure 5As shown, in a certain vehicle road test scene, the global multi-dimensional data acquisition module synchronously collects the following data through the interior camera, the surround-view camera, the main laser radar, and the vehicle body CAN bus: the driver does not hold the steering wheel for 5 seconds (interior camera), there is a stationary obstacle 150 meters in front (main laser radar), the current speed of the vehicle is 60 km / h and it is in a straight line (vehicle body CAN bus), and there is a vehicle approaching at a speed of 55 km / h on the right lane (surround-view camera). The data fusion perception module first removes three false obstacle points generated by raindrops interference through the Gaussian filtering algorithm for laser radar data; and then fuses multi-source data through the Kalman filtering algorithm to eliminate the time synchronization error of each sensor (such as the 10ms time difference between camera and laser radar data), and accurately determines the obstacle position and the relative speed of the surrounding vehicles. The data structured processing module converts the fused data into bird's eye view structured information by using the improved BEV-Transformer, automatically labels the labels such as "driver distraction", "front stationary obstacle", and "neighbor approaching", and manually supplements the label of the "slippery road in rainy day" special scene, and finally forms the true value perception data including the time stamp, target coordinate, and behavior label, and the road label data, which provides accurate and standardized analysis basis for subsequent subjective and objective evaluation.

[0065] In the embodiment of the present application, the data set scattering module 300 includes: as shown Figure 6 The data online processing and storage module, the artificial subjective evaluation module, and the automatic objective evaluation module.

[0066] The data online processing and storage module is used for real-time uploading and storing the structured data, and simultaneously distributing the data to the subjective and objective evaluation modules; the artificial subjective evaluation module is used for manually labeling the vehicle violation or risk moment through the vehicle-mounted handheld device, recording the suspicious time stamp, combining the professional judgment of the traffic police, and forming the subjective evaluation result; and the automatic objective evaluation module is used for inputting the vehicle driving state data, the environmental element data, and the road traffic rule library data through the Rete algorithm to perform rule matching, judging whether the vehicle behavior conforms to the traffic rules in real time, and automatically generating the objective evaluation result.

[0067] It is understood that the online data processing and storage module in this application embodiment ensures the timeliness and integrity of the data by uploading, storing and distributing structured data in real time, providing data support for subsequent evaluation stages. At the same time, it systematically manages the data, facilitating traceability and retrieval. The manual subjective evaluation module uses a vehicle-mounted handheld device to tag and mark moments of violation or risk, combining this with the professional judgment of traffic police to form results. This can compensate for the limitations of the algorithm in complex and special scenarios, improving the flexibility and accuracy of the evaluation. The automated objective evaluation module uses the Rete algorithm to match vehicle status, environment and traffic regulations data, automatically generating evaluation results, ensuring the objectivity and standardization of the judgment, improving evaluation efficiency, and avoiding the subjective bias of manual evaluation. The collaboration of these three modules provides a comprehensive and reliable subjective and objective basis for the subsequent evaluation and identification of the retrospective analysis platform.

[0068] For example, such as Figure 7 As shown, in a real-world urban road test, the online data processing and storage module continuously received and stored 40 minutes of structured driving data, including 12 types of structured information such as real-time vehicle speed (25-70 km / h), steering angle (-30° to +25°), distance to surrounding vehicles (5-80m), and traffic light status. It generated 30 data records per second, storing a total of 72,000 data records, which were simultaneously distributed to the subjective and objective evaluation modules. In the manual subjective evaluation module, testers, using a handheld device in the vehicle, noticed at 15 minutes and 08 seconds that the driver's gaze had deviated from the road ahead for 10 consecutive seconds (internal camera data). They immediately manually tagged the data, recording the questionable timestamp 15:08:12, and simultaneously uploaded it to the system. Subsequently, combined with the traffic police's professional judgment of "driving behavior that impedes safe driving," a subjective evaluation result was formed: "The driver is at risk of distraction and requires further evaluation." The automated objective evaluation module also calls the Rete algorithm to perform rule matching on the data at 28 minutes and 15 seconds: the vehicle speed is 65 km / h (the current road section speed limit is 60 km / h), the duration is 8 seconds, and there are no emergency avoidance environmental elements (such as ambulance avoidance). It completely matches the rule in the road traffic rule library of "exceeding the speed limit by more than 10% but less than 20%", and automatically generates an objective evaluation result of "speeding violation (minor)" with specific timestamps, speed difference and other quantitative data.

[0069] In this embodiment of the application, the backtracking analysis platform 400 includes, as follows: Figure 8 As shown, the modules include video playback and timestamp matching, subjective and objective result fusion evaluation, and analysis report generation.

[0070] The picture playback and timestamp matching module is configured to intercept pictures before and after the moment of vehicle violation or risk through a timestamp indexing algorithm, and to perform spatio-temporal backtracking and picture synchronous playback on the event in combination with the suspicious time stamps recorded by the label dotting module; the subjective and objective result fusion evaluation module is configured to analyze subjective and objective evaluation results through a Cohen's Kappa coefficient and a confusion matrix, and to make a violation determination; and the analysis report generation module is configured to generate a traffic rule compliance analysis report according to the evaluation results.

[0071] It should be noted that the formula of the timestamp indexing algorithm is as follows:

[0072] ;

[0073] ;

[0074] ;

[0075] wherein, is a timestamp corresponding to the physical location of the data in the storage medium; is a reference starting timestamp of data acquisition; is a time and position conversion coefficient; is an initial offset; is a standard timestamp after unified synchronization of multi-source data; is an original timestamp collected by the first type of device; is a time deviation of the first type of device; is an index fragmentation interval corresponding to a time range; is a starting timestamp of a target time range; is an ending timestamp of the target time range; is a data fragmentation period.

[0076] The Cohen's Kappa coefficient is an index for measuring the consistency degree of two evaluation methods when classifying and evaluating the same batch of objects, and the formula is as follows:

[0077] ;

[0078] ;

[0079] ;

[0080] wherein, is an observation consistency rate; is the number of the first type of label determined by both parties consistently; is the total number of evaluation samples; Number of categories; Expected consistency rate; For the first Total number of samples in the row; For the first Total number of samples in the column; This is Cohen's Kappa coefficient.

[0081] It is understood that the retrospective analysis platform in this application uses a video playback and timestamp matching module to capture images before and after the violation or risk moment using a timestamp indexing algorithm, and combines the questionable timestamps to play back the event, ensuring that key scenes are clearly restored; the subjective and objective result fusion evaluation module uses Cohen's Kappa coefficient and confusion matrix to comprehensively analyze subjective and objective evaluation results, improving the scientificity and accuracy of violation identification; the analysis report generation module generates a traffic regulation compliance analysis report based on the evaluation results, providing complete and standardized conclusions for the assessment and identification of vehicle violations or risky behaviors; and providing support for improving driving safety and standardizing driving behavior.

[0082] For example, such as Figure 9 As shown in the image, during a highway test, the retrospective analysis platform processed a suspected illegal lane-changing incident. The video playback and timestamp matching module used a timestamp indexing algorithm to accurately capture 30 seconds of footage before and after the incident (timestamp range 10:23:15-10:23:45), simultaneously matching the questionable timestamp of 10:23:22 from manually tagged records. This clearly replayed the entire process of the vehicle crossing two lanes without using its turn signal, while also correlating with LiDAR data to reconstruct the relative positions of surrounding vehicles (only 2.3 meters from the vehicle on the left). Subjectively, traffic police determined it to be "impeding safe driving," while objectively, the Rete algorithm was used to match the "continuous lane changes without signal" rule. The subjective and objective results fusion evaluation module used Cohen's Kappa coefficient analysis (consistency coefficient 0.89), combined with confusion matrix verification (true positive rate 92%), ultimately confirming it as an "illegal lane change." The analysis report generation module generates a traffic regulation compliance report containing a timestamp, violation type, risk level (high risk), and surrounding environmental parameters, clearly indicating "violation of Article 44 of the Implementation Regulations of the Road Traffic Safety Law." This provides specific data support for subsequent driving behavior improvement and system algorithm optimization. The entire process achieves a closed loop from event reconstruction to determination, ensuring that the results are traceable and verifiable.

[0083] The driving data collection and analysis system provided in the embodiments of the present application detects the external traffic participants and road feature data of the vehicle by using the surround-view camera and the main laser radar, improves the external environment perception range and real-time perception capability of the vehicle body, and eliminates errors by means of the data fusion perception module through the Gaussian filtering algorithm and the Kalman filtering algorithm for real-time fusion processing of the sensor data. The data structure processing module performs structured processing and labeling on the fused data by using the improved BEV-Transformer, combines artificial supplement of special scene labeling to form the true value perception and road label data, improves the perception data precision to support the judgment test result, and fixes the sensor through the unified standard suction cup support on the roof of the vehicle to simplify the equipment installation, improve the equipment integration, and reduce the degree of wire clutter and the calibration difficulty. The artificial subjective evaluation module manually labels the risk moment through the handheld device on the vehicle, and the automatic objective evaluation module generates the objective result by matching the data and the traffic rule library through the Rete algorithm, so that the real-time test evaluation capability of the subjective and objective combination is improved. Thus, the problems of low perception range, insufficient perception precision, and single evaluation function in the prior art are solved.

[0084] A driving data collection and analysis system will be described below through a specific embodiment, as shown in Figure 10 , comprising:

[0085] In a certain vehicle road test scene, the global multi-dimensional data collection module works through various devices. The driver operation state collection module monitors the abnormal steering wheel operation of the driver and captures the pedal action state data in which the driver's line of sight deviates from the road for 4.2 seconds by using the interior-view camera (frame rate 30 fps) and high-precision inertial navigation at the 18th minute and 23rd second of the test. The vehicle body exterior road and environment detection module detects the dynamic traffic participant of the truck suddenly decelerating on the right lane by using the surround-view camera (resolution 1920x1080) and the main laser radar (point cloud density 200 points / m2) at the 22nd minute and 15th second, and detects the road feature of the construction area 80 meters ahead at the 25th minute and 8th second to obtain 327 valid obstacle point data. The vehicle body data collection module obtains the vehicle data such as the speed of 35-70 km / h during the whole test, 28 times of cumulative braking, and the maximum steering angle of 35° by using the vehicle body CAN bus and the combined navigation hardware facilities, and forms a vehicle dynamic data set. The integrated device fixes various sensors through the unified standard suction cup support on the roof of the vehicle, simplifies the equipment installation process by about 40%, reduces the calibration difficulty, and controls the calibration error within ±0.5°.

[0086] The intelligent fusion perception processing module processes the collected data. After receiving the sensor data, the data fusion perception module uses the Gaussian filtering algorithm formula: The noise reduction process filters out 156 noise points generated by the laser radar due to haze, and then combines the Kalman filter algorithm formula: Real-time fusion processing eliminates the 8ms time synchronization error between the camera and the radar, and corrects the obstacle distance error from the original ±3.2 meters to ±0.8 meters; the data structure processing module processes the fused perception data through the improved BEV-Transformer, using the formula: 、 、 Data structuring and labeling module automatically labels information such as "construction area (coordinates X=235.6, Y=48.2)" and "truck speed reduction (relative speed -15km / h)", and supplements the special scene label "haze visibility about 500 meters" with manual annotation, forming a true perception and road label data set containing 128 structured records.

[0087] The data set scattering module plays a role in data aggregation, storage, management and distribution. The data online processing and storage module performs real-time uploading and storage of structured data. During the 2-hour test, 3.2MB of data was received per second, and 4.6GB of information was stored, while these data were distributed to the subjective and objective evaluation modules; the artificial subjective evaluation module uses a handheld device on the vehicle to manually label the driver's line of sight deviation at 18:23:23, which is the risk moment of the vehicle, and records the suspicious timestamp. Three traffic police officers unanimously believe that this behavior belongs to "obstructing safe driving" based on their professional judgment, forming the subjective evaluation result; the automatic objective evaluation module uses the Rete algorithm formula , inputs the vehicle driving state data (speed 62km / h, speed limit 60km / h), environmental factor data (construction area distance 55m) and road traffic rule library data at 25:12:25 to perform rule matching, and generates the objective evaluation result "speeding and not maintaining a safe distance" (matching degree 0.98).

[0088] The backtracking analysis platform carries out follow-up work based on the above results. The picture playback and timestamp matching module uses the timestamp indexing algorithm to intercept the pictures from 18:23:18 to 18:23:26 (driver's line of sight deviation period) and from 25:07:50 to 25:08:30 (construction area response period), and combines the suspicious timestamp recorded by the label marking module to realize the spatio-temporal backtracking and picture synchronous playback of the event; the subjective and objective result fusion evaluation module uses Cohen's Kappa coefficient to utilize The subjective and objective evaluation results are comprehensively analyzed by calculating 0.91 and the confusion matrix (94% accuracy of violation identification), and the "driver distraction" and "overspeed" are identified as violation behaviors; the analysis report generation module generates an intersection regulation compliance analysis report containing 12 data indicators based on the evaluation results, providing detailed basis for subsequent optimization and improvement of the vehicle.

[0089] In summary, the embodiment of the application collects the driver's operation state, the vehicle's own state and external environment data through the global multi-dimensional data collection module using the interior camera, the surround camera, the main laser radar and the vehicle body CAN bus and the like, and integrates the devices to simplify the installation and reduce the calibration difficulty; the intelligent fusion perception processing module first removes the laser radar noise points by Gaussian filtering, then eliminates the time difference by Kalman filtering, corrects the obstacle distance error, and then performs structured processing by improving BEV-Transformer, and generates the true value data in combination with manual supplement of special scene labeling; in the data set scattering module, the data online processing and storage module distributes the cumulative data in real time, and the manual subjective evaluation and the automatic objective evaluation form corresponding results respectively; the backtracking analysis platform restores the key scene through the timestamp index, generates a report containing multiple indicators through coefficient analysis and matrix verification, forms a complete closed loop from data collection to result output, and provides a basis for intelligent networked vehicle intersection regulation compliance real vehicle training and optimization.

[0090] Secondly, a driving data collection and analysis method according to an embodiment of the application is described with reference to the accompanying drawings.

[0091] As shown in Figure 11 The driving data collection and analysis method includes the following steps:

[0092] In step S101, the vehicle external environment, vehicle own state and driver operation state data are acquired.

[0093] It can be understood that, by acquiring the vehicle external environment, vehicle own state and driver operation state data, the embodiment of the application can perform real-time 360° panoramic perception on the vehicle body, provide comprehensive and multi-dimensional raw information for data fusion processing, and guarantee that the data accuracy is improved to a relative true value through fusion perception; supports real-time test evaluation combining subjective and objective evaluation, which is helpful to identify vehicle violation or risk behavior and promote effective development of intelligent networked vehicle intersection regulation compliance test.

[0094] In step S102, the vehicle external environment, vehicle own state and driver operation state data are processed and denoised, the data is fused by Kalman filtering algorithm, the fused data is structured and labeled by improving BEV-Transforme, and special scenes are manually supplemented and labeled to form true value perception data and road label data.

[0095] Among them, truth perception data refers to structured data that reflects the true state of vehicle position, speed, and distance to surrounding obstacles during vehicle operation, while road label data refers to structured data containing information related to drivable areas, speed limit signs, and road features.

[0096] It is understood that the embodiments of this application use truth-aware data to reflect the true state of vehicle position, speed, and distance to surrounding obstacles, and road label data to reflect drivable areas, speed limit signs, and road feature data. This supports the subjective and objective evaluation system in accurately identifying violations or risky behaviors, ensures that the retrospective analysis platform can effectively conduct event assessment and determination, and serves as the basis for traffic regulation compliance testing.

[0097] For example, in a traffic compliance test for a certain intelligent connected vehicle, the true-value perception data, through multi-source sensor fusion processing and labeling, presents precise information such as the vehicle's real-time position (X=320.5, Y=78.3), speed of 58 km / h, distance to the vehicle ahead of 23.6 meters, and relative speed of the pedestrian on the right of -2 km / h. The road label data clearly indicates that the current road segment is an urban arterial road (speed limit 60 km / h), the drivable area width is 3.8 meters, the distance to the stop line at the upcoming intersection is 150 meters, and the lane type is a straight lane. This data is simultaneously transmitted to both the objective and subjective evaluation systems. The objective evaluation system compares the vehicle speed with the speed limit value in the road label and, combined with the distance to the vehicle ahead, determines that the vehicle is not speeding and is maintaining a safe following distance. The subjective evaluation, through the pedestrian dynamics in the true-value perception data, confirms that the vehicle is not obstructing pedestrian traffic. Ultimately, these factors together provide an accurate basis for determining the compliance of the vehicle's driving during that period.

[0098] In this embodiment of the application, the Kalman filter algorithm formula is:

[0099] ;

[0100] ;

[0101] ;

[0102] ;

[0103] in, For vehicles at any time State prediction value; For vehicle status from time At the time The evolutionary pattern; For vehicles at any time The best estimated state; The input matrix; For control input; is the uncertainty of the state prediction of the vehicle at time is the uncertainty of the state prediction of the vehicle at time is the uncertainty of the state prediction of the vehicle at time is the uncertainty of the state prediction of the vehicle at time is the transpose matrix of is the transpose matrix of is the measurement matrix is the transpose matrix of is the transpose matrix of is the uncertainty in the vehicle motion model is the adjustment of the predicted state to incorporate new measurement values is the noise in the vehicle sensor measurement is the measurement result of the vehicle sensor at time is the measurement result of the vehicle sensor at time is the accurate state of the vehicle at time is the accurate state of the vehicle at time

[0104] It can be understood that the embodiments of the present application eliminate the time deviation caused by the response speed difference of different data acquisition devices by time synchronizing the vehicle external environment, self-state and driver operation state data, and at the same time, fuse the data, integrate complementary information, eliminate redundant interference, improve the accuracy and stability of the data, and provide a basis for generating a traffic regulation report.

[0105] For example, when the vehicle is driving on an urban road, the main laser radar of the body external road and environment detection module measures the distance of the front same-direction vehicle to be 42 meters at time t1, and the look-around camera measures the distance to be 39 meters at the same time. There is a 0.15 second time difference between the two due to the response speed difference. After Gaussian filtering processing and noise reduction, the Kalman filter first predicts that the distance of the front vehicle at time t1 should be 40.5 meters according to the driving speed (60 km / h) of the vehicle itself obtained through the CAN bus, and then performs time synchronization and calibration on the laser radar and camera data, and outputs 40.2 meters after fusion. At time t2 (0.3 seconds later), the laser radar measures 38 meters, and the camera measures 36 meters. After synchronous fusion by the Kalman filter, 37.1 meters is obtained. In this process, the Kalman filter effectively eliminates the time deviation of different sensor data, fuses multi-source information to obtain more accurate vehicle distance data, and provides a reliable basis for the intelligent fusion perception processing module to subsequently perform data structured processing and labeling by improving the BEV-Transformer, and forms true value perception and road label data.

[0106] In step S103, the subjective evaluation result is generated by manually labeling through the vehicle-mounted handheld device according to the perception true value data and the road label data, combined with the professional judgment of the traffic police, and at the same time, the objective evaluation result is generated by using the Rete algorithm to match the rules combined with the road traffic rule database data and the vehicle design running conditions.

[0107] It is understood that the embodiments of this application use perceived truth data and road label data to enable manual labeling to specifically mark vehicle violations or risky behaviors. Combined with the subjective evaluation results generated by traffic police professional judgment, the results are consistent with the actual scenario. At the same time, the Rete algorithm is used to match road traffic rule base data, vehicle design operating conditions and related data with rules, thereby improving the accuracy of the generated objective evaluation results, enhancing the comprehensiveness and accuracy of the test evaluation, and promoting the development of traffic compliance testing for intelligent connected vehicles.

[0108] For example, in a city morning rush hour test, the true-value perception data showed the vehicle's current position (X=452.1, Y=120.7), speed of 42 km / h, lateral distance of 1.2 meters from the vehicle in the adjacent lane on the left, and the traffic light at the intersection ahead being red and 35.8 meters from the stop line. Road label data indicated that the intersection was a no-left-turn zone, the lane type was a straight-ahead lane, and there was a zebra crossing 5 meters before the stop line. Testers manually labeled instances where the vehicle did not slow down before the red light using a handheld device, recording the timestamp 08:15:49 and marking it as "suspected risk of cutting in front," generating a subjective evaluation result. Simultaneously, the automated objective evaluation system used the Rete algorithm to match the vehicle's speed, distance from the stop line, and other data with the "slow down and stop before a red light" rule in the road traffic rule database and the vehicle's design braking parameters, determining a potential violation and generating an objective evaluation result. Both types of results were uploaded to a retrospective analysis platform, providing detailed evidence for the traffic regulation compliance assessment of this scenario.

[0109] In this embodiment of the application, the Rete algorithm formula is:

[0110] ;

[0111] ;

[0112] in, For matching degree; For the first in the road traffic rules database Rules; Indicates the vehicle's time A set of real-time state facts; For joint matching degree; The number of rules that need to be evaluated simultaneously.

[0113] It is understood that the embodiments of this application combine road traffic rule database data with vehicle design and operating conditions to quickly match vehicle driving status data and environmental element data with rules, determine in real time whether vehicle behavior complies with traffic regulations, generate objective evaluation results, improve the efficiency and accuracy of rule matching, provide algorithmic support for automated objective evaluation, and timely and accurately identify vehicle violations or risky behaviors from the data, complementing human subjective evaluation and ensuring the comprehensiveness and authority of the evaluation results.

[0114] For example, such as Figure 12 As shown, while driving on the expressway, the system has acquired ground truth perception data (accurate vehicle speed of 92 km / h at time t1, actual distance to the vehicle in front less than the safe value, and hazard warning lights not activated) and road label data (accident point 300 meters ahead, three-lane road attributes). Based on this data, the Rete algorithm quickly performs rule matching according to rules in the road traffic rule base such as "speeding more than 10% on expressways is a violation" and "failing to maintain a safe distance and warning when encountering an accident ahead is a violation." It identifies that the vehicle has triggered two violation rules simultaneously, and then generates an objective evaluation result and marks this moment as a risk moment. At time t2, the ground truth perception data is updated to a vehicle speed of 75 km / h, the distance to the vehicle in front is within the safe range, and the hazard warning lights are activated. After the Rete algorithm matches the rules again, it confirms that the violation status has been resolved, efficiently completing the matching of real-time data and the rule base, providing reliable support for automated objective evaluation.

[0115] In step S104, based on the subjective evaluation results and objective evaluation results, and combined with the timestamp indexing algorithm, the video footage before and after the moment of violation or risk during vehicle operation is captured and played back. The Cohen's Kappa coefficient and confusion matrix are used for analysis and a report is generated.

[0116] The confusion matrix is ​​a tool that presents the number of matching or mismatched samples of subjective and objective evaluation results in a matrix form, which is used to visually demonstrate the details of the consistency between the two.

[0117] It is understood that the embodiments of this application present the number of samples with various matching and mismatches of subjective and objective evaluation results in a matrix form, which intuitively reflects the consistency or discrepancy of the subjective and objective evaluation systems, locates the type and scale of discrepancies, improves the synergy of subjective and objective evaluation, and provides a basis for optimizing evaluation rules and report generation.

[0118] For example, during the morning rush hour, a test vehicle drives along the urban expressway, and there are 15 suspected violations within 25 minutes. After system processing, the consistency of the artificial subjective evaluation and the automatic objective evaluation results is analyzed by Cohen's Kappa coefficient, and the Kappa value is 0.68, which shows that the two have moderate consistency. Then further analysis is combined with the confusion matrix: the matrix row represents the objective evaluation result, and the column represents the subjective evaluation result. Among them, 8 times (TP) are judged as violations by both objective and subjective evaluation, involving clear violation behaviors such as making and receiving calls, exceeding speed by more than 5km / h, etc.; 5 times (TN) are judged as compliance by both objective and subjective evaluation, such as normal courtesy to pedestrians, driving in the designated lane, etc.; 1 time (FP) is judged as a violation by objective evaluation but as compliance by subjective evaluation, which is an algorithm mistake of "not keeping a safe distance" when the vehicle slowly passes through the construction site; 1 time (FN) is judged as a violation by subjective evaluation but as compliance by objective evaluation, which is a short moment of looking down at the navigation that is not recognized by the algorithm, and confirmed as a distraction behavior by artificial review. From the matrix, it can be clearly seen that the consistency of subjective and objective evaluation is higher for obvious violation behaviors (TP accounts for 53.3%), and the difference mainly concentrates in the fuzzy scene (FP+FN accounts for 13.3%). After comprehensive analysis, "driver distraction" and "excessive speed" are determined as the core violation types, and the analysis report generation module generates a traffic rule compliance report containing 12 indicators accordingly.

[0119] According to the driving data collection and analysis method provided in the embodiments of the present application, the outside road and environment detection module is used to detect the outside traffic participants and road feature data of the vehicle by using the surround-view camera and the main laser radar, so as to improve the outside environment perception range and real-time perception capability of the vehicle; the data fusion perception module is used to reduce noise by using the Gaussian filtering algorithm, and to eliminate errors by using the Kalman filtering algorithm to perform real-time fusion processing on the sensor data; the data structure processing module is used to perform structured processing and labeling on the fused data by using the improved BEV-Transformer, and to form the true value perception and road label data by combining the artificial supplement of special scene labeling, so as to improve the perception data accuracy to support the judgment test result; the sensor is fixed by using the unified standard suction cup support on the roof of the vehicle, so as to simplify the equipment installation, improve the equipment integration, and reduce the degree of wire clutter and the difficulty of calibration; the artificial subjective evaluation module is used to manually label the risk moment by using the handheld device on the vehicle, and the subjective result is formed in combination with the judgment of the traffic police; the automatic objective evaluation module is used to generate the objective result by matching the data and the traffic rule library by using the Rete algorithm, so as to improve the real-time test evaluation capability of the subjective and objective combination. Thus, the problems of low perception range, insufficient perception accuracy, and single evaluation function in the prior art are solved.

[0120] A driving data collection and analysis method will be described below through a specific embodiment, as shown in Figure 13 , which includes

[0121] On a city trunk road during the morning rush hour, a test vehicle is driving on a section with a speed limit of 60 km / h. The global multi-dimensional data collection module starts working: the vehicle's own state data shows that at time t1, the vehicle speed is 68 km / h, the brake pedal is not pressed, and the distance to the vehicle in front is only 35 meters; the external environment data shows that 150 meters ahead is an intersection with a red light, and there is a bus on the right lane that is entering a station; the driver's operation state data shows that the steering wheel angle is 0°, the accelerator pedal is pressed by 10%, and the line of sight is not directed towards the intersection direction. These data are collected in real time through devices such as interior cameras, surround view cameras, laser radars, and vehicle CAN bus, forming a raw data set.

[0122] The above raw data is first processed by the Gaussian filter algorithm formula: to remove noise caused by sensor vibration and light changes, such as narrowing the fluctuation range of vehicle speed data from ±2 km to ±0.5 km; then through the Kalman filter algorithm formula: fusion is performed to eliminate the 0.1 second response time difference between laser radar and camera, obtaining the fused data: accurate vehicle speed 67.8 km / h, actual distance to the vehicle in front 34.6 meters, intersection red light status confirmed. Subsequently, the improved BEV-Transformer performs structured processing on the fused data, labeling road features such as "intersection red light" and "bus entering station", and supplementing the label for the special scene of "driver's line of sight deviation" combined with artificial labeling, finally forming true value perception data (including vehicle dynamics and driver state) and road label data (including intersection attributes and traffic participant information).

[0123] Based on the generated true value perception and road label data, artificial evaluators use handheld devices on the vehicle to label at time t1, combined with the professional judgment of traffic police on "not reducing speed near intersection" and "following too close", generating the subjective evaluation result of "high risk"; at the same time, the Rete algorithm formula: calls the road traffic rule library, matches the two rules of "speeding more than 10% on speed limit section" and "not reducing speed at red light intersection", finds that the current vehicle speed exceeds the speed limit by 13% and does not reduce speed, automatically generates the objective evaluation result of "two violations", and both mark time t1 as a risk point.

[0124] The system combines the subjective and objective evaluation results and uses the timestamp indexing algorithm to intercept the 10-second video playback before and after time t1: it can be seen that the vehicle gradually accelerates from 65 km / h to 68 km / h, the vehicle in front has started to decelerate, the driver frequently looks at the phone, and the vehicle in front has started to decelerate. Subsequently, the Cohen's Kappa coefficient formula: The consistency of the subjective and objective results is calculated, and the score is 0.82 (high consistency), the confusion matrix shows that the two judgments of 'overspeed' and 'non-deceleration' are unbiased, finally an analysis report is generated, which clearly indicates that the vehicle is at risk due to overspeed, following too close and driver distraction at t1 time, and it is suggested to strengthen the monitoring of the vehicle speed before the intersection and the driver state.

[0125] In summary, the embodiment of the application obtains the original data of the vehicle, the environment and the driver, and forms the true value perception and road label data through noise reduction, fusion, structured processing and labeling; combined with artificial label dotting and algorithm rule matching, the subjective and objective evaluation results are generated, and finally the report is generated through backtracking analysis. The accuracy of data processing and the efficiency of rule matching are improved by using multiple algorithms, the comprehensiveness of evaluation is improved by artificial supplement and professional judgment, and a systematic and high-precision solution for driving safety analysis is provided.

[0126] Figure 14 The structure schematic diagram of the electronic device provided by the embodiment of the application is provided. The electronic device can include:

[0127] The memory 1401, the processor 1402, and the computer program stored in the memory 1401 and executable on the processor 1402.

[0128] The processor 1402 executes the program to implement the driving data acquisition and analysis method provided in the above embodiment.

[0129] Further, the electronic device further includes:

[0130] The communication interface 1403 is used for communication between the memory 1401 and the processor 1402.

[0131] The memory 1401 is used to store the computer program executable on the processor 1402.

[0132] The memory 1401 can include a high-speed RAM (Random Access Memory, Random Access Memory) memory, and can also include a non-volatile memory, such as at least one disk memory.

[0133] If the memory 1401, the processor 1402 and the communication interface 1403 are implemented independently, the communication interface 1403, the memory 1401 and the processor 1402 can be connected with each other through a bus and complete communication between each other. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 14 Only one thick line is used to represent the bus in the figure, but it does not mean that there is only one bus or only one type of bus.

[0134] Optionally, in a specific implementation, if the memory 1401, the processor 1402 and the communication interface 1403 are integrated on a chip, the memory 1401, the processor 1402 and the communication interface 1403 can complete communication between each other through an internal interface.

[0135] The processor 1402 can be a CPU (Central Processing Unit), or an ASIC (Application Specific Integrated Circuit), or an integrated circuit configured to implement one or more embodiments of the present application.

[0136] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the driving data collection and analysis method.

[0137] In the description of the specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example" or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the specification, the illustrative description of the above terms is not necessarily for the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the skilled in the art can combine and combine the different embodiments or examples described in the specification and the features of the different embodiments or examples without contradiction.

[0138] Furthermore, the terms "first", "second", etc. are used herein only to describe different steps or features and do not imply a relative importance or a defining characteristic. Thus, features defined with "first", "second" etc. can implicitly or explicitly include one or both of the features. The meaning of "a", "an" and "the" includes singular and plural referents.

[0139] Any process or method described in flow diagrams or otherwise described herein can be understood as representing code modules, segments, or portions of code that include one or more executable instructions for implementing specific logic functions (or steps) of the custom logic, and the various embodiments of the application can include additional or fewer steps or processes in alternative implementations. As such, these processes or methods can be embodied in hardware, software, firmware, or combinations thereof, and can be implemented in one or more of the above implementations with or without express support of the specified computing environment.

[0140] It should be understood that portions of the present application can be realized with hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be realized with software or firmware stored in memory and executed by a suitable instruction execution system. As such, if realized with hardware and in another embodiment, any of the following technologies or their combinations known in the art can be used: discrete logic circuitry having logic gates for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), etc.

[0141] Those skilled in the art of the present technology can understand that all or part of the steps carried out by the above-mentioned embodiment methods can be completed by programs instructing relevant hardware, and the programs can be stored in a computer readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0142] Although the above has shown and described the embodiments of the present application, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A vehicle data acquisition and analysis system, characterized in that, include: The system comprises a full-domain multi-dimensional data acquisition module, an intelligent fusion sensing and processing module, a data distribution module, and a retrospective analysis platform; among which, The full-domain multi-dimensional data acquisition module is used to collect data on the vehicle's external environment, vehicle status, and driver operation status in real time during vehicle operation. The intelligent fusion perception processing module is used to perform real-time fusion processing on the data, and to perform structured processing and labeling on the fused data to form true value perception and road label data. The data distribution module is used to aggregate the processed data, store and manage the data, and distribute it to the subjective and objective evaluation module. Subjective and objective evaluation results are generated by matching manual labeling with algorithm rules. The retrospective analysis platform is used to combine subjective and objective evaluation results to conduct offline retrospective analysis, assessment, and identification of vehicle violations or risky behaviors, and generate an analysis report. The retrospective analysis platform includes a video playback and timestamp matching module, a subjective and objective result fusion evaluation module, and an analysis report generation module. The subjective and objective result fusion evaluation module is used to comprehensively analyze the subjective and objective evaluation results using Cohen's Kappa coefficient and confusion matrix, and to identify violations.

2. The driving data acquisition and analysis system according to claim 1, characterized in that, The comprehensive multi-dimensional data acquisition module includes a driver operation status acquisition module, an external road and environment detection module, a vehicle body data acquisition module, and an integrated device. The driver operation status acquisition module monitors the driver's steering wheel operation and pedal actions in real time using an internal camera and high-precision inertial navigation. The external road and environment detection module detects external traffic participants and road feature data using a surround-view camera and a main LiDAR. The vehicle body data acquisition module acquires vehicle data via the vehicle's CAN bus and integrated navigation hardware, forming a dynamic vehicle data set. The integrated device uses a standardized suction cup bracket on the roof to fix sensors, simplifying installation and reducing calibration difficulty.

3. The driving data acquisition and analysis system according to claim 1, characterized in that, The intelligent fusion perception processing module includes a data fusion perception module and a data structuring processing module. The data fusion perception module receives data from various sensors, processes it for noise reduction using a Gaussian filtering algorithm, and performs real-time fusion processing using a Kalman filtering algorithm to eliminate errors. The data structuring processing module processes and annotates the fused perception data using an improved BEV-Transformer, and supplements it with special scene annotations by combining manual input to form true-value perception and road label data.

4. The driving data acquisition and analysis system according to claim 1, characterized in that, The data distribution module includes an online data processing and storage module, a human subjective evaluation module, and an automated objective evaluation module. The online data processing and storage module uploads and stores structured data in real time and distributes the data to the subjective and objective evaluation modules. The human subjective evaluation module uses a handheld device in the vehicle to manually tag moments of vehicle violations or risks, records questionable timestamps, and combines this with professional judgment from traffic police to form a subjective evaluation result. The automated objective evaluation module uses the Rete algorithm to input vehicle driving status data, environmental element data, and road traffic rule database data for rule matching, determines in real time whether vehicle behavior complies with traffic regulations, and automatically generates an objective evaluation result.

5. A vehicle data acquisition and analysis system according to claim 1, characterized in that, The video playback and timestamp matching module is used to capture images before and after the moment of vehicle violation or risk using a timestamp indexing algorithm, and combine them with the questionable timestamps recorded by the tagging module to perform spatiotemporal retrospective of the event and synchronize video playback; the analysis report generation module is used to generate a traffic regulation compliance analysis report based on the evaluation results.

6. A method for collecting and analyzing driving data, characterized in that, include: Acquire data on the vehicle's external environment, vehicle status, and driver operation status; The data on the vehicle's external environment, vehicle status, and driver operation status are processed and denoised. The data is then fused using a Kalman filter algorithm. The fused data is then structured and labeled using an improved BEV-Transformer. Additional labels are added manually for special scenarios to form true-value perception data and road label data. Based on the perceived true value data and road label data, manual labeling is performed using a vehicle-mounted handheld device. Combined with the professional judgment of traffic police, subjective evaluation results are generated. At the same time, the Rete algorithm is used to perform rule matching with road traffic rule database data and vehicle design operating conditions to generate objective evaluation results. Based on the subjective and objective evaluation results, and combined with the timestamp indexing algorithm, the system captures the before and after footage of the vehicle's violation or risk moments while driving. The subjective and objective evaluation results are then analyzed using Cohen's Kappa coefficient and confusion matrix, and a report is generated.

7. The method for collecting and analyzing driving data according to claim 6, characterized in that, The Kalman filter algorithm formula is as follows: ； ； ； ； in, For vehicles at any time State prediction value; For vehicle status from time At the time The evolutionary pattern; For vehicles at any time The best estimated state; The input matrix; For control input; For vehicles at any time Uncertainty in state prediction; For vehicles at any time Uncertainty in state prediction; for The transpose of the matrix; For measurement matrix; for The transpose of the matrix; Uncertainties in the vehicle motion model; To adjust the predicted state to incorporate the new measurements; Noise measured by vehicle sensors; For vehicle sensors at all times The measurement results; For vehicles at any time The precise state.

8. The method for collecting and analyzing driving data according to claim 6, characterized in that, The Rete algorithm formula for: ； ； in, For matching degree; For the first in the road traffic rules database Rules; Indicates the vehicle's time A set of real-time state facts; For joint matching degree; The number of rules that need to be evaluated simultaneously.

9. An electronic device, characterized in that, include: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the driving data acquisition and analysis method of claim 7.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed, they implement the driving data acquisition and analysis method of claim 7.

Images