Vehicle Detection And Tracking Based On Monocular Vision

Problem of traffic safety has become a problem can not be ignored with high popularityof vehicle in the people’s life. How to reduce traffic accidents caused by vehicles’ collisionhas become one of the important research issues in recent years. Vehicle pre-collisionwarning system is an effective solution to this problem.The commercial vehicle anti-collision warning systems are mainly based on monocularvision. Vehicle detection and tracking are two core algorithms for such systems. Based on theanalysis and comparison of the related algorithms in car detection, we propose the newvehicle detection and tracking algorithms that can be applied to the highway scene. Maincontributions are as follows:1) For the vehicle detection in daytime highway scene, we propose a car detectionalgorithm based on vehicle’s shadow and vertical edge symmetry features. Firstly, it uses lanedetection to locate the lane area and get the shadow line under the vehicle by the threshold ofshadow. Secondly, the ROI (region of interesting) is generated by the shadow line. Finally,the vertical edge symmetry rules are used to verify whether the ROI is a vehicle object.2) For the vehicle detection in nighttime highway scene, we propose a car detectionalgorithm based on vehicle tail light’s features. Firstly, it locates the area of vehicle tail lightsprobably exist. Secondly, it uses search algorithm to get the candidate tail light regions in theimage based on color, luminance and spatial distribution features. Finally, it recognizes thevehicles through using the geometric features of vehicle’s candidate tail lights to match theminto tail light pairs.3) For the problem of stability of vehicle tracking and vibration disturbance, we proposecar tracking algorithm based on neighborhood with pre-tracking queue and stable trackingqueue to stably tracking vehicle. The algorithm that based on neighborhood will largelyreduce the area that needs to be tracked which can improve computing speed. Thepre-tracking queue is used to sift out the stable tracking target which can enhance the stabilityof tracking algorithm. The stable tracking queue that based on neighborhood is used toenhance the fault tolerances of losing target and changes of the target features which canreduce the interferences of camera shaking and background light. This measure canstrengthen immunity and accuracy of the tracking algorithm.We implements the vehicle detection and tracking algorithm in the Windows systemwith Visual Studio2008. The experimental results of highway videos show that the proposed algorithm has better stability, accuracy and real-time performance.