Research On Vision Based Environmental Perception Technology For Intelligent Vehicles In Complex Urban Environment

The surge in vehicle ownership has increased the pressure on transport system.Energy depletion,air pollution and traffic safety caused by vehicles have brought unprecedented pressure to human society.By virtue of intelligent sensing,intelligent decision and decision control,intelligent vehicles(IVs)can improve traffic system utilization rate effectively.IVs has become an important technical means for the developed countries to achieve intelligent transportation system(ITS).As one of the key technologies of IVs,the environmental perception algorithm based on computer vision has been widely researched in the structured environment such as highway through detecting,identifying and reconstructing and provide environmental information for IVs.In the complex urban environment,multiple targets block with each other and their movement trajectories are complex,which obstruct application of existing environmental perception algorithm.In this paper,based on machine vision environment perception as research content,muilti-target detection and tracking algorithm is designed,in order to achieve multi-target such as lanes,vehicles and traffic signs detection and tracking in urban environments.This paper main contents are as follows:(1)According to the camera characteristics,pinhole model is established and calibrated by faugeras calibration method with constrains.After calibration,internal and external parameter matrixs can be gotten which are used to estimate camera position in vehicle coordinate.Camera position and the some camera processing parameters are used to establish inverse perspective mapping(IPM)model and realize inverse perspective transformation of images.(2)According to the camera characteristics,vanishing point(VP),lane geometry and position constrainces are established which as priori knowledge constrains Hough transform,lane clustering and fitting results.In this way,lanes in images can be detected.(3)HOG feature and color histogram feature can describe the characteristics of vehicles and traffic signs respectively.In order to improve the computational efficiency and the accuracy of net classifier in vehicles and different kinds of traffic signs.Sliding window optimized by camera characteristics are used to traversal scan images` ROI area.(4)In order to solve the multi-targer long-term tracking problem in complex environment,a multi-target tracking algorithm based on historical trajectory information is designed.The historical detect results are used to predict or fuse new detect results by three steps which are short trajectory generation,long trajectory generation and historical trajectory reliability verification.And then,according to new detection results,standard deviation classifier,nearest neighbor classifier and the historical trajectory information are updated gradually.Repeat above steps and achieve multi-objective long-term tracking in complex environments.