Design And Implementation Of Vehicle Violation Detection System Based On Vehicle Video

With the improvement of the national economy,residents’ private cars have gradually increased,and road traffic has entered a new stage of development and crisis.Among which the detection and supervision of vehicle violations is particularly important.This paper designs and implements a vehicle violation detection system based on the vehicle video,which mainly includes vehicle detection,separation of static and dynamic vehicles and speed estimation.Lane detection and classification.Over speeding,illegal parking,illegal lane change of three kinds of vehicle violations of the judgment.Construction of interactive interface of violation detection system and development of auxiliary function module.Practical works are as follows:(1)The design of vehicle detection and speed estimation module.Mask-RCNN network was used to segment vehicles,and the detection frame and Mask information of vehicles were obtained.Based on the improvement of conventional ORB feature extraction and matching method,image pyramid and gray centroid method were introduced for feature extraction.And RANSAC algorithm was combined to complete matching.Then estimate the vehicle camera pose,and the minimum reprojection idea is used to optimize the pose of the camera.In the separation process of moving and static vehicles,the scene flow information is calculated based on the camera pose parameters,and the dynamic and static vehicle feature points are judged based on this.At the same time,the weighted average method is used to optimize the decision threshold.And then the proportion of dynamic feature points is calculated according to the target mask information to realize the separation of moving and stationary vehicles.In the part of speed estimation,by solving the position change relationship of vehicle feature points in adjacent frames,combining camera pose and vehicle pose to estimate the speed.Finally introducing factor graph for optimization processing,which can get more accurate speed.(2)Lane detection and classification module design.Based on the traditional UFLD lane detection model,the hollow pyramid pool was adopted,the frequency attention mechanism was introduced,the loss function was improved by adding L-Dice function.The improved model could extract more abundant lane feature information while maintaining the original high detection rate,and at the same time,the focus on the lane edge was strengthened.For lane line classification,a classification algorithm based on the RGB values of lane line detection points is adopted to determine lane color and continuity and complete lane line classification.(3)Design of vehicle violation detection and decision module.Based on the above module,the vehicle detection frame,the results of vehicle motion and static determination,the speed of moving vehicles,the position and category of lane line are obtained.Through the relationship between speed and speed limit for speeding violation judgment.Judging illegal parking through the relationship between vehicle speed and lane position and category.Through the relationship between the center position at the bottom of the vehicle detection frame and the lane line,and combined with the identification and switch of the turn signal at the rear of the vehicle,the illegal lane change violation is determined.(4)Development and implementation of auxiliary function modules of the system,which including user registration and login,vehicle video upload,violation picture capture,violation record and history review based on license plate and violation behavior.At the same time,combined with the video data of mobile violation detection vehicles,a test data set based on real scenes was made.And based on this,the vehicle violation detection system designed and developed in this paper was tested for function and performance,and the relevant results met the system design requirements.