Research Of Vehicle And Road Recognition Based On Machine Vision

In recent years, vehicle number of worldwide continues to increase,especially inChina,simultaneously,more and more vehicle driving safety problem on road isgrowing.Therefore, The demand of driver assistance system for safe driving areincreasingly strong. The vehicle and road detection system based on machine visiondescribed by this paper is a part of vehicle driving assistance system.By this system, firstly,process and analyze the image shooted by single-camera.Then,compute the locationinformation of road lane and vehicle.With the above information,this system can improvethe safety of vehicle driving on road.Firstly,we introduce two key techniques in this paper,lane detection and vehicledetection.And explain the concept of machine vision and introduce the image pre-processwhich is necessary before the two key techniques.Secondly, we make a research on random hough transform alogorithm used in lanedetection.Considering the accuracy and timeliness factors,we use the area constraint andangle constraint to enhance the accuracy and timeliness.Thirdly, we make a research on vehicle detection method. According to the demand ofaccuracy and timeliness of vehicle detection,we make a solution for the above demand.Thissolution include two part of modules which are hypothesis generation and hypothesisverification.As enhancements for accuracy and timeliness of two modules, we design twoframes for the two mudules.The hypothesis generation frame is made up of the AdaBoostmethod and Haar-like feature.The hypothesis verification frame is made up of SVM methodand HOG feature.According to above two frames,we combine the timeliness advantage ofAdaBoost frame,which include rapid selection of adaboost and rapid computing ofHaar-like feature, with the accuracy advantage of SVM frame,which include the SVMclassify ability of high dimensions feature and the rich gradient information of HOGfeature.According to the system implementation and test, the vehicle and road recognition system in the paper get a high perfomance on the detection rate and timeliness.