Research And Application Of Vehicle Detection And Tracking Algorithm In Traffic Video

Vehicle detection and tracking are the basic components of intelligent transportation systems.They provide a basis for following high-level information analysis such as vehicle counting and driving behavior detection.At present,the detection of moving vehicles has the problem that the detection effect cannot be ensured simultaneously in the accuracy and realtime performance.Vehicle tracking has the problem of poor vehicle scale adaptability and antiocclusion capability.This paper is to study vehicle detection and tracking algorithms in traffic video aiming at obtaining traffic information quickly and accurately.The main research contents are as follows:(1)The traditional vehicle detection algorithm has the problem that the accuracy and realtime of the detection effect cannot be taken into consideration.This paper proposes a moving vehicle detection algorithm based on trust interval PBAS.The initial background model is built using multi-frame interval images.Area layer information is more likely to reflect background complexity.It is proposed to calculate the background complexity by using regional structure complexity and color complexity.This paper sets the confidence interval to determine whether the current background model is suitable for updating according to whether the current traffic condition and location of the pixel are within the confidence interval,and accurately and quickly detect the moving vehicle.(2)The vehicle tracking algorithm based on correlation filtering has less dependence on background information,which leads to the deviation of tracking results.And the algorithm is difficult to adapt to changes in vehicle scale.Based on the KCF tracking algorithm,a vehicle scale adaptive tracking method using background information is proposed.It takes the target vehicle and the surrounding background image as samples,and uses the minimized error model and the cyclic matrix to create a position classifier.The target position is determined based on the peak value of the position classifier output response.In this way,the accuracy of the vehicle position tracking of the classifier in a complex environment is improved.At the same time,samples of the same vehicle image on different scales are collected,and the scale classifier is established by using the minimum error model and the cyclic matrix theory.The target scale is determined based on the peak value of the scale classifier output response.In this way,the adaptability of the classifier to vehicle scale changes during tracking is improved.(3)The vehicle in the video sequence is extracted using the moving vehicle detection algorithm.At the same time,the virtual detection area is set to determine whether the vehicle determines the rule in the virtual area.Count the number of vehicles driving out of the virtual area to achieve statistics on traffic flow.The tracking algorithm is used to determine the position coordinates of the target vehicle in each frame.Fit the motion trajectory to determine whether the vehicle has driving behavior such as retrograde and lane change.