Research On Key Issues In Vehicle Safety Driving Assistant Techniques Based On Machine Vision

Driver assistance system is an important part of the intelligent transportationsystem. Its purpose is to reduce the traffic incidence, as well as to improve the roadtraffic safety. In view of the environmental detection within driver assistance system,the problems of the vision based lane detection, stereo matching, and pedestriandetection are addressed in this paper. The research topics are summarized as follows:(1) Lane detection based on the structural road is studied. Firstly, the OTSU andCanny edge detection operator are combined to segment image, and then, the lane lineis detected with Hough transform algorithm; On this basis, in order to reduce theimpact of light and changes in the lane line, the paper propose a sub-regional adaptivealgorithm to detect lane line, moreover, the results is optimized with the principle ofregional growth, after that, the lane line is fitted with the method of least squares.(2) Binocular vision matching is studied. The binocular matching system is dividedinto three parts: calibration, rectification and matching. The left and right CCD iscalibrated respectively by Zhang·s calibration method based on a flat-screencalibration template, and then, the translation and rotation parameters of the twocameras are calculated. After that, the binocular images are rectified according to theresults of calibration, so that the matching points in the same pixel row; Finally,accurate disparity maps are computed using Bayesian approach to stereo matching.(3) Pedestrian detection in the static environment is studied. The HOG feature andsupport vector machine are combined to detect pedestrians. The overall process ofpedestrian detection is divided into two steps: training and testing, which are analyzedin this chapter. Finally, the HOG feature and the principles of classification based onsupport vector machine are addressed in this paper.