Research Of Moving Object Detection And Tracking Based On Video In Intelligent Transportation And Its Application

Intelligent transportation system(ITS)refers to a transportation service system which is formed by applying a new generation of information technology including mobile internet,computer vision technology and image processing technology to the field of transportation,and intelligent transportation system has drawn more and more attention.Intelligent transportation system can effectively alleviate traffic congestion,reduce the incidence of accidents,while improving the utilization of traffic resources.Vehicle detection and tracking has always been an important research topic in the field of intelligent transportation.Through the detection and tracking of vehicles,we can get the relevant information.This information can not only provide samples for subsequent vehicle identification and motivation judgment,but also serve as a basis for providing vehicle drivers with navigation,rescue and other services.Therefore,real-time and stable vehicle tracking becomes very meaningful.Detection and tracking are two essential parts of the vehicle tracking process,the vehicle detection module is used to obtain the initial information of the target vehicle,vehicle tracking module gets these information to complete the follow-up tracking.At present,vehicle detection and tracking algorithms are faced with many problems such as target occlusion,target size change and complex scenes.Based on the research of the existing algorithms,this thesis puts forward the corresponding improvements,designs and implements the vehicle detection and tracking system.The main work of this thesis is as follows:(1)The principle,advantages and disadvantages of typical moving vehicle detection algorithms such as background difference method,interframe difference method and optical flow method are analyzed.A background differential vehicle detection algorithm based on automatic background updating is proposed.By keeping the background continuously updated,the problem that traditional background differential models are prone to drifting is solved.(2)The vehicle tracking algorithms such as Mean Shift and particle filter are studied,and the applicability of each algorithm is summarized.The detection-based TLD(Tracking-Learning-Detection)tracking algorithm is analyzed.In view of the problem that the detection module of the TLD algorithm scans too large when samples are generated,a Kalman filter is introduced to estimate the target detection area.The tracking module of TLD algorithm uniformly generates feature points in the tracking target,it is difficult to ensure that each feature point can be sustained and stable tracking,easily lead to tracking drift.In response to this problem,Harris corner detection is introduced to extract the key points.(3)The vehicle detection and tracking system is designed.The vehicle detection module adopts a background difference algorithm based on background automatic updating.The vehicle tracking module uses an improved TLD algorithm combining Kalman filter with Harris corner detection.(4)In MATLAB2015b and Visual Studio 2013 platform,the system is programmed by calling OpenCV visual function library.Compared with the original TLD algorithm,the improved TLD algorithm shows better real-time and tracking accuracy.