CN118072260B

CN118072260B - Vehicle offset analysis method and system based on image

Info

Publication number: CN118072260B
Application number: CN202410465787.2A
Authority: CN
Inventors: 刘月清; 刘哲良
Original assignee: Shandong Shuangbai Electronic Co ltd
Current assignee: Shandong Shuangbai Electronic Co ltd
Priority date: 2024-04-18
Filing date: 2024-04-18
Publication date: 2024-08-16
Anticipated expiration: 2044-04-18
Also published as: CN118072260A

Abstract

The invention relates to the technical field of image recognition, in particular to a vehicle offset analysis method and system based on images. The method comprises the following steps: acquiring environment data and road images; carrying out fine environmental modeling based on the environmental data and the road image to generate a fine three-dimensional road model; constructing a signal lamp distribution network based on the refined three-dimensional road model to generate the signal lamp distribution network; the method comprises the steps of carrying out contour segmentation on a passing vehicle through a signal lamp distribution network to obtain a vehicle running image; performing vehicle behavior simulation analysis on the vehicle running image to generate a vehicle behavior simulation analysis image; and carrying out vehicle deviation analysis on the vehicle behavior simulation analysis image to generate a vehicle deviation analysis data set. According to the invention, through fine environment modeling, signal lamp distribution network construction and angle adjustment camera and vehicle offset risk assessment, the efficiency and the accuracy of vehicle offset analysis are improved.

Description

Vehicle offset analysis method and system based on image

Technical Field

The invention relates to the technical field of image recognition, in particular to a vehicle offset analysis method and system based on images.

Background

The rise of computer vision lays a foundation for vehicle offset analysis, and the development of deep learning technology thoroughly changes the pattern of image recognition. In particular Convolutional Neural Networks (CNNs), have been proposed and developed that enable computers to learn and extract higher-level features from images, which allows vehicle offset analysis to be implemented in dependence on more complex, accurate models. As urban traffic becomes increasingly complex, real-time requirements for vehicle offset analysis are becoming increasingly high. While conventional computer vision methods often have difficulty meeting the real-time processing requirements, deep learning-based methods can achieve real-time through optimization algorithms and hardware acceleration, and these technological advances allow vehicle offset analysis to be completed in seconds. Besides image data, vehicle offset analysis can be combined with other sensor data such as radar, laser radar and the like to perform multi-mode information fusion, and the fusion can improve the accuracy and the robustness of offset analysis, so that the system can better cope with various complex situations. However, in the prior art, vehicle offset analysis is still required to be performed on the basis of images photographed at different angles of a camera, so that the operability of the offset analysis is complex, and meanwhile, the prior art often lacks of fine processing on environment data and road images, so that the accuracy of the vehicle offset analysis is not high under complex road conditions.

Disclosure of Invention

Based on the foregoing, it is necessary to provide a vehicle offset analysis method and system based on images to solve at least one of the above-mentioned problems.

To achieve the above object, an image-based vehicle offset analysis method includes the steps of:

an image-based vehicle offset analysis method, comprising the steps of:

Step S1: acquiring environment data and road images; carrying out fine environmental modeling based on the environmental data and the road image to generate a fine three-dimensional road model;

step S2: constructing a signal lamp distribution network based on the refined three-dimensional road model to generate the signal lamp distribution network; the method comprises the steps of carrying out contour segmentation on a passing vehicle through a signal lamp distribution network to obtain a vehicle running image; performing vehicle behavior simulation analysis on the vehicle running image to generate a vehicle behavior simulation analysis image; performing vehicle deviation analysis on the vehicle behavior simulation analysis image to generate a vehicle deviation analysis data set;

step S3: according to the vehicle deviation analysis data set, carrying out angle adjustment on the signal lamp camera on the refined three-dimensional road model to generate a signal lamp angle adjustment camera; the camera is adjusted based on the angle of the signal lamp to perform vehicle offset animation production, so that vehicle offset animation is obtained;

Step S4: performing vehicle deviation risk assessment on the vehicle deviation animation so as to generate vehicle deviation risk assessment data; and uploading the vehicle offset risk assessment data to a cloud platform for offset risk monitoring, and performing response alarm processing when the vehicle offset risk prediction data is greater than a preset standard risk threshold.

According to the invention, through the detailed modeling of the environment data and the road image, the road condition including the road curvature, the gradient, the traffic sign and the like can be more accurately understood, and the vehicle behavior simulation and the offset analysis can be more accurately facilitated. The signal lamp distribution network is established, so that the behavior of the vehicle in different signal lamp states can be simulated, and the vehicle deviation condition can be predicted better. By analyzing the vehicle running image, real-time vehicle position, speed and other information can be acquired, and vehicle behavior simulation and offset analysis can be facilitated. The simulation analysis is carried out based on the vehicle running image, so that the behavior of the vehicle under different conditions can be predicted, and data support is provided for subsequent offset analysis. By carrying out offset analysis on the vehicle behavior simulation analysis image, potential vehicle offset conditions can be identified, and a basis is provided for subsequent prediction and monitoring. And adjusting the angle of the signal lamp camera according to the offset analysis data, thereby being beneficial to improving the accuracy and coverage range of monitoring. Through animation, the vehicle deviation condition can be intuitively displayed, and the model is convenient to understand and train. Through model training and prediction, the offset risk can be early warned in advance, and preventive measures can be taken and the occurrence probability of traffic accidents can be reduced. The data is uploaded to the cloud platform for monitoring, so that real-time monitoring and response to vehicle offset risks can be realized, and road safety is further improved. Therefore, the vehicle offset analysis efficiency and accuracy are improved through the refined environment modeling, the signal lamp distribution network construction, the angle adjustment camera and the vehicle offset risk assessment.

In the present specification, there is provided an image-based vehicle offset analysis system for performing the above-described image-based vehicle offset analysis method, the image-based vehicle offset analysis system comprising:

The three-dimensional model construction module is used for acquiring environment data and road images; carrying out fine environmental modeling based on the environmental data and the road image to generate a fine three-dimensional road model;

The vehicle offset analysis module is used for constructing a signal lamp distribution network based on the refined three-dimensional road model to generate a signal lamp distribution network; the method comprises the steps of carrying out contour segmentation on a passing vehicle through a signal lamp distribution network to obtain a vehicle running image; performing vehicle behavior simulation analysis on the vehicle running image to generate a vehicle behavior simulation analysis image; performing vehicle deviation analysis on the vehicle behavior simulation analysis image to generate a vehicle deviation analysis data set;

The camera angle adjustment module is used for adjusting the angle of the signal lamp camera on the refined three-dimensional road model according to the vehicle deviation analysis data set to generate a signal lamp angle adjustment camera; the camera is adjusted based on the angle of the signal lamp to perform vehicle offset animation production, so that vehicle offset animation is obtained;

The offset monitoring module is used for carrying out vehicle offset risk assessment on the vehicle offset animation so as to generate vehicle offset risk assessment data; and uploading the vehicle offset risk assessment data to a cloud platform for offset risk monitoring, and performing response alarm processing when the vehicle offset risk prediction data is greater than a preset standard risk threshold.

The vehicle deviation prediction system has the beneficial effects that through the refined environment modeling and the vehicle behavior simulation analysis, the system can more accurately predict the vehicle deviation behavior, so that an alarm can be sent out in time and corresponding measures can be taken, the occurrence of traffic accidents can be reduced, and the traffic safety can be improved. The system can identify traffic bottlenecks and high-risk areas through the signal lamp distribution network and the vehicle deviation analysis data set, and provide data support for traffic management departments, so that traffic flow is optimized, and traffic jams are reduced. By predicting the vehicle deviation risk, the system can give an alarm in time to help drivers and traffic management departments to take measures for avoiding accidents, thereby reducing the occurrence rate of traffic accidents and related costs, including casualties, vehicle damage, road maintenance and the like. By monitoring and analyzing the vehicle offset path trajectory data, the system can identify road conditions and traffic flow, provide real-time traffic information for drivers, and help them to select more efficient travel routes, thereby improving traffic efficiency. Therefore, the vehicle offset analysis efficiency and accuracy are improved through the refined environment modeling, the signal lamp distribution network construction, the angle adjustment camera and the vehicle offset risk assessment.

Drawings

FIG. 1 is a flow chart illustrating steps of an image-based vehicle offset analysis method;

FIG. 2 is a flowchart illustrating the detailed implementation of step S2 in FIG. 1;

FIG. 3 is a flowchart illustrating the detailed implementation of step S22 in FIG. 2;

FIG. 4 is a flowchart illustrating the detailed implementation of step S24 in FIG. 2;

The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.

Detailed Description

The following is a clear and complete description of the technical method of the present patent in conjunction with the accompanying drawings, and it is evident that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, are intended to fall within the scope of the present invention.

Furthermore, the drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus a repetitive description thereof will be omitted. Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically separate entities. The functional entities may be implemented in software or in one or more hardware modules or integrated circuits or in different networks and/or processor methods and/or microcontroller methods.

It will be understood that, although the terms "first," "second," etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another element. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element, without departing from the scope of example embodiments. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.

To achieve the above object, referring to fig. 1 to 4, an image-based vehicle offset analysis method includes the steps of:

In the embodiment of the present invention, as described with reference to fig. 1, a schematic flow chart of steps of an image-based vehicle offset analysis method according to the present invention is provided, and in this example, the image-based vehicle offset analysis method includes the following steps:

In the embodiment of the invention, the environment data, such as the laser radar, the camera, the GPS and the like, are acquired by using various sensors and devices so as to acquire the related information of the road and the surrounding environment. The road image is specifically shot through a camera or unmanned aerial vehicle and other equipment, and a real-time image of the road is obtained. And preprocessing the acquired environmental data and road images, including denoising, correcting, aligning and the like, so as to improve the accuracy and reliability of subsequent modeling. Based on the preprocessed environment data and road images, environment modeling is performed by using computer vision and image processing technology. The road image can be converted into a three-dimensional model by adopting a three-dimensional reconstruction technology, and meanwhile, the detailed information such as the geometric shape, the marking, the traffic sign, the guideboard and the like of the road is considered. And the data acquired by different sensors are fused by adopting a data fusion technology, so that the comprehensiveness and accuracy of environmental modeling are improved. The actual condition of the road is considered, including the road curvature, gradient, road width, road surface condition, etc., to more truly reflect the road environment. And converting the data obtained by the environmental modeling into a refined three-dimensional road model. Including the road stereo structure, road elevation, traffic facilities, etc. for subsequent steps.

in the embodiment of the invention, three-dimensional topographic data of a road is acquired by using laser radar (LiDAR) scanning or combining with photogrammetry technology. The method utilizes computer aided design software such as AutoCAD or SolidWorks to create a refined three-dimensional road model, and specifically utilizes the technologies of Geographic Information System (GIS), computer Aided Design (CAD), photogrammetry and the like. Based on a refined three-dimensional road model, a signal lamp in an image is detected and identified by utilizing a computer vision technology, the detection is realized by a target detection algorithm, such as YOLO, SSD and the like, or a characteristic-based method, such as a Haar cascade detector and the like, the detected signal lamp is subjected to attribute classification, such as a traffic light, a pedestrian light and the like, the detected signal lamp is subjected to position estimation, the specific position of the detected signal lamp in a road is determined, a signal lamp distribution network is constructed according to the detected signal lamp position and the attribute, the intersection and a road section are connected in a topological way, and the signal lamp is identified on the intersection, so that the network can identify the position and the type of the signal lamp on the road, and the deep learning and image processing technology is utilized. The constructed signal lamp distribution network is utilized to carry out contour segmentation on vehicles on roads, and the contour segmentation is realized by a semantic segmentation or instance segmentation technology, and deep learning and computer vision technology are used. Based on the vehicle running image obtained by segmentation, the behavior of the vehicle on the road is simulated and analyzed by using a physical simulation and behavior modeling technology, and the method relates to a rule-based model and combines physical, machine learning and simulation technologies. The vehicle behavior simulation analysis image is analyzed to detect whether the vehicle deviates from a predetermined travel track, specifically by image processing, such as target detection and track tracking algorithms.

according to the embodiment of the invention, the angle of the signal lamp camera in the refined three-dimensional road model is adjusted by analyzing the offset condition and the position information in the data set according to the vehicle offset. The angle of the camera is simulated or actually adjusted, so that the vehicle offset behavior can be effectively captured, and the camera can cover an offset area on a road. Based on analysis results, the angle of the signal lamp camera is adjusted by using mechanical control and automation technology, which involves mechanical design by Computer Aided Design (CAD) software and actual angle adjustment by using microcontrollers such as Arduino or Raspberry Pi. According to the adjusted camera angle, computer graphics and animation technology are used to produce an animation showing the vehicle offset condition, three-dimensional modeling and animation software such as Blender or Maya is required, and real-time rendering technology such as Unity or Unreal Engine is involved to enhance visual effects.

In the embodiment of the invention, the key frame images are extracted from the vehicle deviation animation, the characteristics related to the vehicle deviation, such as the vehicle position, the vehicle speed, the vehicle direction and the like, are extracted according to the requirements, a set of evaluation algorithm is designed, and the vehicle deviation risk is evaluated according to the extracted characteristics, so that the vehicle deviation risk is realized based on rules, statistical analysis or other image analysis technologies. For example, the evaluation may be based on the magnitude, frequency, and environmental conditions of the vehicle offset. And applying a risk assessment algorithm to each key frame image to generate corresponding vehicle offset risk assessment data, wherein the corresponding vehicle offset risk assessment data comprises information such as frame number, risk level, offset angle, offset amount and the like. And uploading the generated vehicle offset risk assessment data to a cloud platform for storage and monitoring. And ensuring the safety and reliability of data transmission, facilitating subsequent monitoring and analysis, setting a risk threshold on the cloud platform, facilitating timely detection of abnormal conditions, and determining the threshold according to historical data or expert advice. And triggering an alarm processing mechanism when the vehicle deviation risk assessment data is larger than a preset standard risk threshold value. Responding to the alert may include sending an alert notification, triggering an automated control system to correct the offset, or notifying the driver to take necessary action to reduce risk.

Preferably, step S1 comprises the steps of:

step S11: acquiring environment data and road images;

Step S12: converting the three-dimensional point cloud data of the environmental data to generate environmental point cloud data;

step S13: carrying out image enhancement on the road image to generate a road enhanced image; carrying out data space alignment on the environment point cloud data and the road enhanced image to generate road environment space alignment data;

step S14: and carrying out fine environment modeling based on the road environment space alignment data to generate a fine three-dimensional road model.

According to the invention, the environment data and the road image are fused, and the more comprehensive environment information is provided by combining the point cloud data and the image data. Through refined environment modeling, a more real and accurate three-dimensional road model is generated, and the perception and decision making capability of an automatic driving system are improved. The data space alignment is carried out on the environment point cloud data and the standard road image, so that errors caused by data inconsistency can be reduced, and the consistency and reliability of the data are improved. The road image is enhanced, so that the image quality and the information extraction efficiency can be improved, and better input data can be provided for the subsequent steps. By comprehensively utilizing the point cloud data and the image data, the perception capability of an automatic driving system to the environment can be improved, so that complex traffic environments and road conditions can be better dealt with.

In the embodiment of the invention, the environmental data including the position, distance, shape and other information of surrounding objects is acquired by using the sensor (such as LiDAR, camera and the like). Meanwhile, a camera or other visual sensor is used for acquiring road images, and visual information of the road is captured. Environmental data acquired from LiDAR or like sensors is converted into three-dimensional point cloud data to represent the spatial location and shape of surrounding objects. Enhancing the road image to improve the image quality and accuracy, wherein the image enhancement: contrast, brightness, etc. of the image are enhanced to improve image quality and the degree of visualization of the features. And carrying out space alignment on the environment point cloud data and the preprocessed road image, ensuring that the environment point cloud data and the preprocessed road image are in the same coordinate system, carrying out refined environment modeling based on the road environment space alignment data, including building a three-dimensional road model, identifying traffic signs, lane lines and the like on the road, and analyzing and identifying the environment data by specifically utilizing a deep learning technology to extract more environment characteristics so as to build a more accurate model.

Preferably, step S14 includes the steps of:

Step S141: data filtering is carried out on the road environment space alignment data, and road environment space filtering data are generated; extracting road feature points from the road environment space filtering data to generate road feature point data, wherein the road feature point data comprises road edge data and road center line data;

Step S142: carrying out road plane fitting according to the road edge data and the road center line data to generate an initial three-dimensional road image; projecting the initial road image into the initial road model for texture mapping to generate a three-dimensional road texture image; carrying out road model scene enhancement on the initial three-dimensional road image to generate a three-dimensional road scene enhancement image;

The road model scene enhancement step specifically comprises the following steps:

Adding road marks and marks to the initial three-dimensional road image to generate a three-dimensional road mark image;

Building and tree elements are added to the three-dimensional road sign marking image, and a three-dimensional road element enhanced image is generated;

Performing illumination simulation on the three-dimensional road element enhanced image to generate a three-dimensional road illumination enhanced image;

performing pavement material optimization on the three-dimensional road illumination enhanced image to generate a three-dimensional road scene enhanced image;

Carrying out image time sequence combination on the three-dimensional road mark line image, the three-dimensional road element enhanced image, the three-dimensional road illumination enhanced image and the three-dimensional road scene enhanced image to generate a three-dimensional road scene enhanced image;

Step S143: carrying out pavement detail analysis on the three-dimensional road scene enhanced image to generate pavement concave-convex data and pavement crack data; and carrying out environment dynamic element modeling on the three-dimensional road scene enhanced image according to the road surface concave-convex data and the road surface crack data to generate a refined three-dimensional road model.

According to the invention, through filtering and feature point extraction on the road environment space alignment data, noise and unnecessary information in the data can be reduced, and meanwhile, key feature points such as edges and center lines of a road are extracted, so that accurate road geometric information is provided for subsequent steps. The road edge data and the central line data are utilized for plane fitting, an initial three-dimensional road image can be generated, then the initial three-dimensional road image is projected into a model for texture mapping, the sense of reality and detail of a road are increased, and the road surface has a stereoscopic sense and a fitting sense. By adding road signs, building and tree elements, illumination simulation, pavement material optimization and other operations, the road scene can be more complete and vivid, and the sense of reality and visual effect of the whole scene are enhanced. And different enhanced images are combined in time sequence, so that more continuous and dynamic road scene images can be generated, and the simulated road environment is more vivid and credible. By carrying out pavement detail analysis on the road scene enhanced image, detail data such as concave-convex and crack of the pavement can be obtained, and then dynamic element modeling is carried out according to the data, so that the sense of reality and detail display of the model are further improved.

In the embodiment of the invention, a common filtering algorithm, such as mean filtering, gaussian filtering or median filtering, is used, and an appropriate filter is selected according to specific situations. The extraction of the road feature points can use computer vision technology, such as an edge detection algorithm (such as Canny and Sobel operators), a straight line detection algorithm (such as Hough transformation) and the like, so as to extract the feature points of the road edge, the central line and the like. The road plane fitting specifically uses a fitting algorithm such as a least squares method or RANSAC to estimate a road plane model from the edge data and the center line data. The texture mapping specifically adopts texture mapping technology, the fitted road image is projected onto the surface of an initial road model, the road model scene enhancement involves the steps of adding road marks and marked lines, adding building and tree elements, illumination simulation, optimizing road surface materials and the like, and the texture mapping is specifically realized by using the technology in computer graphics, such as mapping, geometric model addition, illumination model, material mapping and the like. The pavement detail analysis specifically uses a computer vision technology and an image processing method, such as feature extraction, morphological operation, edge detection and the like, so as to obtain detail data such as concave-convex and crack of the pavement. The dynamic element modeling specifically uses three-dimensional modeling technology, such as object modeling, particle system, animation, and the like, and generates a dynamic road scene through the acquired pavement detail data.

Preferably, step S2 comprises the steps of:

step S21: confirming the positions of the signal lamps based on the refined three-dimensional road model, and generating signal lamp position distribution images; constructing a signal lamp distribution network according to the signal lamp position image to generate a signal lamp distribution network;

step S22: the method comprises the steps that a traffic light distribution network is used for carrying out past vehicle contour segmentation on a road enhanced image to obtain a vehicle running image; performing vehicle behavior simulation analysis on the vehicle running image according to the refined three-dimensional road model to generate a vehicle behavior simulation image, wherein vehicle behavior simulation analysis data comprise a vehicle normal simulation image and a vehicle abnormal simulation image;

Step S23: carrying out vehicle journey analysis on the vehicle anomaly simulation analysis data to generate vehicle journey analysis data; according to the vehicle path analysis data, carrying out high-risk accident road section coding on the vehicle behavior simulation image, and generating a high-risk accident road section coding image;

step S24: and carrying out vehicle offset analysis on the vehicle normal simulation image and the vehicle abnormal simulation image according to the high-risk accident road section coding image, and generating a vehicle offset data set.

The traffic flow can be effectively managed and controlled by confirming the positions of the signal lamps and constructing the distribution network, and the traffic safety is improved. By analyzing the vehicle running image, data such as traffic flow, vehicle running track and the like can be obtained, so that traffic flow analysis and road use condition evaluation are performed. Vehicle behavior simulation based on a three-dimensional road model can simulate the running condition of a vehicle on a road, including normal running and abnormal behaviors (such as overspeed, lane change and the like), and is helpful for evaluating the rationality of road design and the influence of vehicle behaviors. By carrying out path analysis and high-risk accident path section coding on the abnormal simulation analysis data, the path section with potential safety hazard can be identified, and measures are taken in advance to prevent traffic accidents. By carrying out vehicle offset analysis on the normal simulation image and the abnormal simulation image, the off-track condition in the running process of the vehicle can be identified, and references are provided for improving road design and traffic management.

As an example of the present invention, referring to fig. 2, the step S2 in this example includes:

In the embodiment of the invention, the three-dimensional model data of the road is obtained by using technologies such as laser radar (LiDAR), photogrammetry, structured light and the like, and the data is processed and reconstructed by using computer vision and a three-dimensional reconstruction algorithm so as to obtain a high-precision road model. Based on a road model and an image processing technology, the signal lamp position on the road is identified and confirmed, and the signal lamp detection and position confirmation are carried out by the technologies of target detection, image segmentation, feature extraction and the like. And marking the position information of the signal lamps on the road model according to the confirmed signal lamp positions, and generating corresponding images, wherein computer graphics and rendering technology and image processing algorithm are required to generate clear signal lamp position distribution images. Based on signal lamp position images, a signal lamp distribution network is constructed, and an image processing technology, an image analysis algorithm and a network analysis method are involved to identify signal lamps on a road and construct connection relations among the signal lamps.

In the embodiment of the invention, the road enhancement image is processed by using a deep learning technology such as a Convolutional Neural Network (CNN) or a semantic segmentation model so as to identify and segment the vehicle outline in the image, thereby helping to accurately extract the vehicle information on the road. The image output by the signal lamp distribution network is processed by utilizing an image segmentation algorithm, such as semantic segmentation or instance segmentation, so as to segment the outline of each vehicle in the image, and specific common technologies include Mask R-CNN, U-Net and the like. Advanced laser radar (LiDAR), photogrammetry and other technologies are used to obtain refined three-dimensional model data of the road, including road geometry, road surface conditions and other information. And combining the road model and the vehicle contour segmentation result, and performing simulation analysis on the behavior of the vehicle by utilizing computer graphics and physical simulation technology, wherein the simulation analysis relates to a vehicle kinematic model, a collision detection algorithm and the like. According to the vehicle behavior simulation analysis result, generating simulation images under the conditions of normal running and abnormal running of the vehicle, wherein the normal simulation images are specifically generated based on conventional traffic behaviors, and the abnormal simulation images need to consider factors such as emergencies, traffic accidents and the like.

In the embodiment of the invention, the abnormal simulation analysis data of the vehicle is processed by using data mining and statistical analysis technology to analyze the information such as the distance, the speed, the acceleration and the like of the vehicle, and methods such as time sequence analysis, cluster analysis and the like are involved to identify abnormal behavior patterns or high-risk road sections. The vehicle path analysis data and the road network data are combined to code the potential high-risk accident road sections, and the potential high-risk accident road sections are evaluated specifically based on factors such as historical accident data, road conditions, traffic flow and the like, and the codes usually represent the positions or characteristics of the high-risk accident road sections in the form of numbers or symbols. The vehicle behavior simulation image is associated with the high-risk accident road section codes, and the vehicle behaviors related to the high-risk road sections in the image are coded, which relate to image processing and computer vision technology to identify the road section positions and the vehicle behaviors in the image. According to the coding result, a high-risk accident road section coding image is generated and used for visually displaying the high-risk road section and the vehicle behavior related to the high-risk road section, so that traffic managers and planners can be helped to more intuitively know traffic safety conditions, and corresponding preventive measures are adopted.

In the embodiment of the invention, the image processing technology and the road safety analysis algorithm are used for analyzing traffic accident data and marking high-risk accident road sections, and the technology of a Geographic Information System (GIS) is related. And extracting a normal simulation image and an abnormal simulation image of the vehicle, wherein the images are generated according to the previous vehicle behavior simulation analysis, and the image data comprise the driving conditions of the vehicle on different road sections. The normal simulation image and the abnormal simulation image of the vehicle are analyzed to identify whether the vehicle deviates from an expected running track, and the image processing technology and the target detection algorithm are involved. The analysis results are arranged into a data set form, including the offset position of the vehicle, the offset degree, the road section information and the like, and can be used as a data basis for subsequent research and analysis.

Preferably, step S22 includes the steps of:

Step S221: the method comprises the steps that a traffic light distribution network is used for carrying out past vehicle contour segmentation on a road enhanced image to obtain a vehicle running image; drawing a passing vehicle running path of the vehicle running image to generate a vehicle running path image;

step S222: according to the refined three-dimensional road model, vehicle running track simulation is carried out on the vehicle path data of the vehicle path and station, and a vehicle track simulation image set is generated;

Step S223: performing time stamp analysis on the vehicle track simulation image to generate a vehicle running time stamp; according to the vehicle running time stamp, vehicle swing frequency analysis is carried out on the vehicle track simulation image set, and vehicle swing frequency data are generated; performing Fourier transform on the vehicle swing frequency data to generate a vehicle swing spectrogram;

Step S224: according to a vehicle yaw lateral speed discrimination formula, performing vehicle stability calculation on a vehicle rocking spectrogram to generate a vehicle stability value; comparing the vehicle stability value with a preset standard stability value, and when the vehicle stability value is greater than or equal to the preset standard stability value, performing corresponding image screening on the vehicle track simulation image set to generate a normal simulation image; when the vehicle stability value is smaller than a preset standard stability value, the vehicle track simulation image set performs corresponding image screening, and an abnormal simulation image is generated;

Step S225: and carrying out image integration on the normal simulation analysis data and the abnormal simulation analysis data so as to obtain a vehicle behavior simulation image.

According to the invention, the vehicle contour segmentation and the driving path drawing are carried out on the road enhancement image, so that the driving track of the vehicle on the road can be clearly displayed, and the vehicle movement condition can be understood. Based on the refined three-dimensional road model and the path data of the vehicle approach station, the running track of the vehicle is simulated and generated, and a real data basis is provided for subsequent behavior analysis. By performing timestamp analysis and wobble frequency analysis on the vehicle track simulation image and applying Fourier transformation, a vehicle wobble spectrogram can be obtained, so that the motion characteristics of the vehicle can be more comprehensively known. And carrying out stability calculation on the swing spectrogram according to a vehicle yaw lateral speed discrimination formula, and comparing a stability value with a preset standard, so that a normal simulation image and an abnormal simulation image can be effectively distinguished, and potential safety risks can be identified. And integrating the normal simulation analysis data and the abnormal simulation analysis data to obtain a complete vehicle behavior simulation image, thereby providing a basis for further research and analysis.

As an example of the present invention, referring to fig. 3, the step S22 in this example includes:

In the embodiment of the invention, the image of the road scene is acquired by using a camera or other sensors, and the image is influenced by illumination, weather and other factors, so that preprocessing is required to enhance the quality and definition of the image. The vehicle is segmented from the road background using an image segmentation algorithm, such as semantic segmentation or instance segmentation. For the vehicle profile obtained by the segmentation, a trajectory tracking algorithm, such as a kalman filter, a particle filter, and the like, is used to estimate and predict the motion trajectory of the vehicle. The driving path of the vehicle is drawn on the vehicle running image according to the estimated track, and the estimated track is projected on the image, and the moving direction of the vehicle is represented by a line segment or an arrow. And superposing the drawn vehicle driving path on the original road enhanced image to generate a vehicle driving path image, once the vehicle contour segmentation image is obtained, identifying the position and the direction of the vehicle by using an image processing algorithm in a computer vision technology, and drawing the driving path of the vehicle by tracking the position of the vehicle. Common methods include kalman filtering, optical flow methods, or feature point based tracking methods. The vehicle travel path is superimposed on the original road image, and a vehicle travel path image is generated, specifically by drawing the vehicle travel path on the original image, or combining it with the road enhancement image.

In the embodiment of the invention, the data are obtained from laser radar scanning, aerial photogrammetry, satellite images or other mapping technologies by obtaining three-dimensional road model data with high precision, wherein the road model contains detailed information such as geometric shape, curvature, gradient, intersection, sign and the like of a road. And acquiring the data of the station path to be passed by the vehicle, wherein the data comprise information such as a starting point, an ending point, a passing point, a running speed, an acceleration and the like, and the data are from GPS track data, traffic simulation or other actual vehicle running data. The simulation of the running track of the vehicle is carried out by utilizing the obtained three-dimensional road model and the vehicle path data, and the simulation running is carried out on the three-dimensional road model according to the path indicated by the vehicle path data, and the factors such as the speed, the acceleration, the steering angle and the like of the vehicle are considered. In the process of vehicle running track simulation, the position information of the vehicle at each time step is recorded, and a vehicle track simulation image set is generated by intercepting images of a three-dimensional road model at corresponding positions or by rendering the motion of the vehicle on the model into images in real time.

Step S223: performing time stamp analysis on the vehicle driving path image to generate a vehicle driving time stamp; according to the vehicle running time stamp, vehicle swing frequency analysis is carried out on the vehicle track simulation image set, and vehicle swing frequency data are generated; performing Fourier transform on the vehicle swing frequency data to generate a vehicle swing spectrogram;

In the embodiment of the invention, the time stamp information in each vehicle driving path image is extracted, specifically, the time stamp information is realized through metadata of an image file or other related information, and the time stamp is converted into a uniform time format, and is sequenced and processed for subsequent frequency analysis. Frequency analysis is performed on the vehicle track simulation image set by using a signal processing technology to identify the swing frequency of the vehicle, and particularly, frequency spectrum analysis is performed on time series data by adopting a Discrete Fourier Transform (DFT) or other frequency domain analysis methods to acquire frequency components. Fourier transform is applied to the vehicle rocking frequency data, which is converted to the frequency domain. Fourier transforms transform time domain data into frequency domain data in order to better analyze the contributions of different frequency components in the vehicle motion.

According to the embodiment of the invention, the vehicle swing frequency spectrogram is calculated according to the vehicle yaw lateral speed judging formula to obtain the vehicle stability value, and the method involves integrating or weighting and summing the amplitudes or the energies of certain frequency ranges in the swing frequency spectrogram to calculate the vehicle stability degree. A standard steady state value representing a threshold value of a normal vehicle motion state is preset, and the standard value is specifically set based on historical data, industry standards or expert experience. And when the running stability value of the vehicle is greater than or equal to the preset standard running stability value, the vehicle is considered to normally run. At this time, the corresponding vehicle track simulation image is selected as the normal simulation image. When the running stability value of the vehicle is smaller than the preset standard running stability value, the vehicle is considered to be abnormal in movement, and in this case, a corresponding vehicle track simulation image is selected as an abnormal simulation image.

Step S225: and carrying out image integration on the normal simulation image and the abnormal simulation image, thereby obtaining a vehicle behavior simulation image.

In the embodiment of the invention, the normal simulation image and the abnormal simulation image are subjected to superposition fusion operation, and the normal and abnormal conditions are displayed in one image by adjusting the transparency or the mixed mode of the images. And displaying the normal simulation image and the abnormal simulation image in the same image side by side, and enabling the normal simulation image and the abnormal simulation image to be visible at the same time through horizontal or vertical arrangement, so as to obtain the vehicle behavior simulation image.

Preferably, the vehicle yaw lateral velocity determination formula in step S224 is specifically as follows:

In the method, in the process of the invention, Expressed as the speed of the vehicle in the lateral direction,Represented as an upper time limit for speed discrimination,Indicated as the current moment of time,Represented as the vehicle at the current timeThe stiffness of the lateral forces to which steering is subjected,Represented as the vehicle at the current timeThe driver gives the steering wheel of the vehicle an input angle,Represented as the vehicle at the current timeAt the same time the speed in the direction of advance,Represented as the nominal speed of the vehicle,Represented as the vehicle at the current timeThe resultant of damping and stiffness experienced during longitudinal movement,Represented as the vehicle at the current timeThe rotational speed of the vehicle about the vertical axis.

The invention analyzes and integrates a vehicle yaw lateral speed discrimination formula, whereinSteering wheel input is describedInfluence on the lateral speed of the vehicle. Lateral stiffness when steering wheel angle is largeThe yaw lateral velocity of the vehicle will be more strongly affected. By combining the lateral stiffness with the steering wheel angle, and by normalizing the longitudinal speed, the effect of steering input on the lateral movement of the vehicle can be more accurately described. In the formulaDescribes the yaw rate of the vehicleImpact on vehicle yaw lateral velocity. When the yaw rate of the vehicle is large, the longitudinal stiffnessThe yaw lateral velocity of the vehicle will be more strongly affected. By combining the longitudinal stiffness and the yaw rate, and by normalizing the longitudinal rate, the effect of yaw motion on lateral motion of the vehicle can be more accurately described. When the lateral velocity judgment formula of the yaw of the vehicle is used, the velocity of the vehicle in the lateral direction can be obtained, and the velocity of the vehicle in the lateral direction can be calculated more accurately by applying the lateral velocity judgment formula of the yaw of the vehicle. The formula comprehensively considers the influence of a plurality of factors such as steering input, lateral rigidity, longitudinal rigidity and transverse speed of the vehicle on the yaw lateral speed, so that the motion stability of the vehicle is more fully described. By combining the factors such as steering wheel rotation angle, lateral rigidity, longitudinal rigidity, yaw rate and the like, the formula can more accurately simulate the actual movement condition of the vehicle, can be used for evaluating and simulating the movement stability of the vehicle, has important significance in vehicle design and control algorithm development, and is beneficial to optimizing the drivability and safety of the vehicle.

Preferably, step S24 includes the steps of:

Step S241: carrying out high-risk accident road section pixel color depth detection on the high-risk accident road section coded image according to the refined three-dimensional road model, and generating pavement tire mark pixel color depth data; carrying out pavement pothole area segmentation on the high-risk accident road section coding image through pavement tire mark pixel color depth data to generate a pavement pothole area image;

Step S242: carrying out vehicle body balance inclination analysis on the vehicle abnormal simulation image by utilizing the pavement pothole area image to generate vehicle running inclination angle data;

Step S243: carrying out vehicle steering wheel image region segmentation on the vehicle abnormal simulation image through a refined three-dimensional road model to generate a vehicle driving steering wheel image; calculating the rotation angle of the running steering wheel of the vehicle on the running steering wheel image of the vehicle, and generating rotation angle data of the running steering wheel of the vehicle;

step S244: calculating the vehicle cornering probability of the vehicle running inclination angle data and the vehicle running steering wheel rotation angle data to obtain a vehicle cornering probability value; setting a cornering threshold value for normal simulation analysis data based on the vehicle cornering probability value, and generating a vehicle running cornering threshold value; and carrying out vehicle deviation analysis on the high-risk accident pavement image through the vehicle running cornering threshold value to generate a vehicle deviation analysis data set.

According to the invention, through carrying out pixel color depth detection on the high-risk accident road section coded image and generating the pixel color depth data of the road surface tire mark, the road surface tire mark of the high-risk accident road section can be effectively identified, and the method is important for road maintenance and traffic safety, and can help to find and repair the problem area of the road surface in time. The road surface indentation area is segmented by using the road surface tire mark pixel color depth data to the high-risk accident road section coding image, so that the road surface indentation area image is generated, the positioning and analysis of the indentation problem on the road are facilitated, important information is provided for road maintenance personnel, so that maintenance measures can be taken in time, and the occurrence of traffic accidents is reduced. By utilizing the pavement pothole area image to perform vehicle body balance inclination analysis on the vehicle abnormal simulation image, vehicle running inclination angle data can be acquired, balance of the vehicle in the pavement pothole area can be evaluated, and information on vehicle behaviors and stability can be provided for drivers and vehicle research personnel. By dividing the vehicle steering wheel image area of the vehicle abnormality simulation image, generating a vehicle running steering wheel image, and calculating the rotation angle data of the vehicle running steering wheel, detailed information about the steering behavior of the vehicle can be provided, which is helpful for understanding the operation of the driver and the control characteristics of the vehicle. The vehicle cornering probability value can be obtained by calculating the vehicle running inclination angle data and the vehicle running steering wheel rotation angle data, which is beneficial to evaluating the cornering risk of the vehicle and providing the safety index of vehicle running.

As an example of the present invention, referring to fig. 4, the step S24 in this example includes:

In the embodiment of the invention, the image processing technology is used for analyzing the high-risk accident road section coded image, detecting pixels related to road surface scars, and calculating the color depth values of the pixels, wherein the color depth refers to the brightness value or the color intensity value of the pixels, and specifically, the scars on the road are identified by determining the range of the color depth. And extracting the color depth data of the detected pavement tire mark pixels to form pavement tire mark pixel color depth data, wherein the data can be used for subsequent pavement pothole area segmentation. The pavement pit area in the high-risk accident road section coding image is separated from other areas by utilizing pavement tire mark pixel color depth data through an image processing and segmentation technology, so that a pavement pit area image is generated, and the pit situation of the pavement can be analyzed more accurately.

In the embodiment of the invention, the vehicle abnormal simulation image and the extracted pavement pothole area image are subjected to pixel level alignment, so that the vehicle simulation image is ensured to be consistent with the actual pavement pothole position. A tire contact area of the vehicle is extracted from the pixel-level aligned image, and an inclination angle of the area is calculated. The inclination angle is determined in particular by calculating the change in pixel coordinates or the shift of the center point of the tire. Similar analysis was performed on all tire contact areas of the vehicle, resulting in tilt angle data for each tire. And (3) calculating the average inclination angle of the whole vehicle or the inclination degree of each tire by integrating the inclination angle data of all the tires.

In the embodiment of the invention, by constructing a refined three-dimensional road model, particularly using computer graphics technology or extracting from real road data, the model should include the geometric shape, width, side slope and other relevant characteristics of the road. Using vehicle simulation software or computer graphics techniques, an image is generated that contains a vehicle anomaly simulation that will be used for vehicle steering wheel image area segmentation and steering wheel rotation angle calculation. The image processing technology is used for carrying out steering wheel image region segmentation on the vehicle abnormal simulation image, extracting the image of the steering wheel for vehicle running, and specifically, an image segmentation algorithm such as threshold segmentation, edge detection, region growing or deep learning method is applied to separate the image of the steering wheel from the background. And calculating the rotation angle of the running steering wheel of the vehicle on the image of the running steering wheel of the vehicle, generating corresponding data of the rotation angle of the running steering wheel of the vehicle, and detecting the position and the boundary of the steering wheel by image processing or computer vision technology to acquire the geometric characteristics of the steering wheel. The steering wheel angle of rotation is calculated using a suitable algorithm (e.g., based on feature matching, template matching, or a deep learning method) by comparing the difference between the steering wheel image of the current frame and the reference steering wheel image.

According to the embodiment of the invention, the vehicle cornering probability value is calculated by combining the vehicle running inclination angle data and the vehicle running steering wheel rotation angle data through statistical analysis, the data reflects the cornering condition of the vehicle in the running process, and the possibility of occurrence of the cornering of the vehicle is estimated through probability calculation. And setting a cornering threshold value for the normal simulation analysis data based on the calculated cornering probability value of the vehicle, wherein the threshold value is determined according to the actual situation and the safety standard and is used for judging whether the cornering risk exists in the running process of the vehicle. By reasonably setting the cornering threshold, potential cornering problems can be effectively identified, and the driving safety of the vehicle is improved. And (3) carrying out vehicle deviation analysis on the road surface image of the high-risk accident by using the set vehicle running cornering threshold value, identifying the vehicle deviation condition existing on the high-risk road section by combining vehicle simulation data through image processing and analysis technology, and generating a corresponding vehicle deviation analysis data set, wherein the data can be used for further safety evaluation and establishment of preventive measures.

Preferably, step S3 comprises the steps of:

Step S31: setting a virtual camera view angle of the refined three-dimensional road model according to the vehicle deviation analysis data set, and generating virtual camera view angle data; performing illumination condition simulation on the virtual camera view angle data to generate virtual camera parameter configuration data;

step S32: performing scene fusion on the vehicle offset analysis data set and the standard road image based on the virtual camera parameter configuration data to generate a vehicle offset analysis fusion scene;

step S33: analyzing the fusion scene according to the vehicle offset to adjust the angle of the signal lamp camera, and generating a signal lamp angle adjusting camera; and (3) carrying out vehicle offset animation production on the road enhanced image based on the angle adjustment camera of the signal lamp to obtain the vehicle offset animation.

According to the invention, the road model is subjected to refined setting according to the vehicle deviation analysis data set, so that the actual road condition including the road surface shape, the bending degree and the like can be more accurately simulated, and the reality and the accuracy of scene simulation are improved. The virtual camera view angle data are generated and the illumination condition simulation is carried out, so that the simulation of the vehicle deviation condition under different illumination conditions is facilitated, the diversity and fidelity of scene simulation are improved, and more data support is provided for subsequent analysis. The virtual camera parameter configuration data, the vehicle offset analysis data set and the standard road image are subjected to scene fusion, so that the virtual scene and the real road condition can be combined, the actual driving environment can be better simulated, and the reality and the accuracy of the simulation effect are improved. And the angle adjustment of the signal lamp camera is carried out on the vehicle deviation analysis fusion scene, so that the signal lamp conditions under different angles can be simulated, more scene changes can be provided, and the simulation complexity and accuracy can be increased. And (3) carrying out vehicle offset animation production on the road enhanced image based on the signal lamp angle adjustment camera to obtain vehicle offset animation, namely recording the road enhanced image by using the adjusted signal lamp angle camera, and creating a real simulation for the offset action of the vehicle so as to further analyze and evaluate the behavior of the vehicle.

In an embodiment of the invention, a refined three-dimensional road model is created from a vehicle offset analysis dataset by using Computer Aided Design (CAD) software or a three-dimensional modeling tool. The position and the orientation of the virtual camera are determined to simulate the visual angle of the vehicle, and the camera calibration and the geometric transformation technology are involved. The method comprises the steps of simulating scenes under different illumination conditions by using a computer graphics technology, including a ray tracing or real-time rendering technology, and generating illumination information of a virtual scene by considering the position, intensity and color of a light source and the reflection characteristics of materials. The image processing and computer vision technology is used for fusing the data under the visual angle of the virtual camera with the standard road image, and the image registration, perspective transformation and mixing technology are involved, so that the reality and consistency of the fused scene are ensured. The angle of the signal lamp camera is adjusted by using a geometric transformation technology to be matched with the road environment in the fusion scene, and a camera calibration technology is required to determine adjustment parameters. And using computer graphics and animation production software to produce vehicle offset animation according to the image recorded by the adjusted signal lamp angle camera, wherein the vehicle offset animation comprises technologies of motion tracking, key frame animation, texture mapping and the like so as to simulate the motion and offset process of the vehicle on the road.

Preferably, step S4 comprises the steps of:

Step S41: dividing a key image frame set of the vehicle offset animation to obtain a vehicle offset key frame image;

Step S42: performing vehicle offset risk assessment on the vehicle offset keyframe image, thereby generating vehicle offset risk assessment data; and uploading the vehicle offset risk assessment data to a cloud platform for offset risk monitoring, and performing response alarm processing when the vehicle offset risk data is greater than a preset standard risk threshold.

The invention is helpful to reduce the processed data quantity and improve the processing efficiency by extracting the key image frames in the vehicle offset animation, and is helpful to evaluate the risk more accurately by focusing attention on the key moment when the vehicle offset is likely to happen. And performing risk assessment on the vehicle deviation key frame image so as to generate vehicle deviation risk assessment data, which is beneficial to the system to assess the risk degree of vehicle deviation according to the image content and scene information and provides basis for subsequent monitoring and processing. And uploading the evaluation data to a cloud platform for offset risk monitoring, so that the vehicle offset risk can be monitored in real time, and the abnormal situation can be found in time. When the vehicle deviation risk data is larger than a preset standard risk threshold value, the response alarm processing can help to take measures in time, potential accident occurrence is avoided, driving safety is guaranteed, sensing and coping capacity of the vehicle deviation risk are improved, accident risk is reduced, and road safety level is improved.

In the embodiment of the invention, the characteristics related to the vehicle deviation in the key frame image are extracted by an image processing technology, wherein the characteristics comprise the position, the direction, the speed, the relative position relation with a road and the like of the vehicle. The extracted features are analyzed, modes and rules related to the vehicle offset risk are found, and the vehicle offset risk is achieved through calculating indexes such as an angle, an offset speed and the like of the vehicle offset. Based on the results of the feature analysis, a set of rules or rule sets are designed for determining the degree of risk of vehicle misalignment. For example, if the angle at which the vehicle deviates from the lane exceeds a certain threshold, or the distance of the vehicle from the road edge is less than a certain limit, a higher risk of deviation is considered to exist. And carrying out risk prediction on each key frame image according to a rule or a rule set, and generating corresponding vehicle offset risk data comprising information such as frame number, offset angle, offset amount and the like. And uploading the generated vehicle offset risk data to a cloud platform for storage and monitoring. And setting a risk threshold on the cloud platform, and triggering a corresponding alarm processing mechanism when the vehicle deviation risk data exceeds a preset standard risk threshold. When it is detected that the risk of vehicle drift exceeds a threshold, the cloud platform sends an alert to notify the relevant personnel or system, performs a corresponding response and process, such as sending an alert message to the driver or relevant department, and taking necessary action to reduce the risk.

The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

The foregoing is only a specific embodiment of the invention to enable those skilled in the art to understand or practice the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An image-based vehicle offset analysis method, comprising the steps of:

Step S2: constructing a signal lamp distribution network based on the refined three-dimensional road model to generate the signal lamp distribution network; the method comprises the steps of carrying out contour segmentation on a passing vehicle through a signal lamp distribution network to obtain a vehicle running image; performing vehicle behavior simulation analysis on the vehicle running image to generate a vehicle behavior simulation analysis image; performing vehicle deviation analysis on the vehicle behavior simulation analysis image to generate a vehicle deviation analysis data set; step S2 comprises the steps of:

Step S21: confirming the positions of the signal lamps based on the refined three-dimensional road model, and generating signal lamp position distribution images; constructing a signal lamp distribution network according to the signal lamp position distribution image to generate a signal lamp distribution network;

Step S22: the method comprises the steps that a traffic light distribution network is used for carrying out past vehicle contour segmentation on a road enhanced image to obtain a vehicle running image; performing vehicle behavior simulation analysis on the vehicle running image according to the refined three-dimensional road model to generate a vehicle behavior simulation analysis image, wherein the vehicle behavior simulation analysis image comprises a vehicle normal simulation image and a vehicle abnormal simulation image; step S22 includes the steps of:

Step S222: performing vehicle running track simulation on the vehicle running path image according to the refined three-dimensional road model to generate a vehicle track simulation image set;

step S224: according to a vehicle yaw lateral speed discrimination formula, performing vehicle stability calculation on a vehicle rocking spectrogram to generate a vehicle stability value; comparing the vehicle stability value with a preset standard stability value, and when the vehicle stability value is greater than or equal to the preset standard stability value, performing corresponding image screening on the vehicle track simulation image set to generate a normal simulation image; when the vehicle stability value is smaller than a preset standard stability value, performing corresponding image screening on the vehicle track simulation image set, and generating an abnormal simulation image;

step S225: performing image integration on the normal simulation image and the abnormal simulation image so as to obtain a vehicle behavior simulation analysis image;

Step S24: according to the high-risk accident road section coding image, carrying out vehicle offset analysis on the vehicle normal simulation image and the vehicle abnormal simulation image to generate a vehicle offset data set; step S24 includes the steps of:

step S244: calculating the vehicle cornering probability of the vehicle running inclination angle data and the vehicle running steering wheel rotation angle data to obtain a vehicle cornering probability value; setting a cornering threshold value for normal simulation analysis data based on the vehicle cornering probability value, and generating a vehicle running cornering threshold value; carrying out vehicle deviation analysis on the high-risk accident pavement image through a vehicle driving cornering threshold value to generate a vehicle deviation analysis data set;

Step S3: according to the vehicle deviation analysis data set, carrying out angle adjustment on the signal lamp camera on the refined three-dimensional road model to generate a signal lamp angle adjustment camera; the camera is adjusted based on the angle of the signal lamp to perform vehicle offset animation production, so that vehicle offset animation is obtained; step S3 comprises the steps of:

Step S33: analyzing the fusion scene according to the vehicle offset to adjust the angle of the signal lamp camera, and generating a signal lamp angle adjusting camera; the camera is adjusted based on the angle of the signal lamp to perform vehicle offset animation production on the road enhanced image, so as to obtain vehicle offset animation;

2. The image-based vehicle offset analysis method according to claim 1, wherein the step S1 includes the steps of:

step S11: acquiring environment data and road images;

3. The image-based vehicle shift analysis method according to claim 2, characterized in that step S14 includes the steps of:

step S142: carrying out road plane fitting according to the road edge data and the road center line data to generate an initial three-dimensional road image; projecting the initial three-dimensional road image into an initial road model for texture mapping to generate a three-dimensional road texture image; carrying out road model scene enhancement on the initial three-dimensional road image to generate a three-dimensional road scene enhancement image;

4. The image-based vehicle deviation analysis method according to claim 1, wherein the vehicle yaw lateral velocity discrimination formula in step S224 is as follows:

；

5. The image-based vehicle offset analysis method according to claim 1, wherein step S4 includes the steps of:

6. An image-based vehicle offset analysis system for performing the image-based vehicle offset analysis method of claim 1, the image-based vehicle offset analysis system comprising: