Research On Road Scene Segmentation And Vehicle Detection Algorithm Based On Stereo Vision

With the rapid development of computer science and robotics,advanced driver assistance systems(ADAS)has become a research hotspot of intelligent vehicles,and has been used in military,civil and other related fields.To ensure the safety of intelligent vehicles on the road,we must identify the free space,which is the world region where navigation without collision is guaranteed.Therefore,the free space and the obstacle detection algorithms,especially the algorithms based on stereo vision has been highly concerned by the domestic and foreign research institutions.In this paper,we mainly studied road scene segmentation and vehicle detection based on vehicle stereo vision.The main work is as follows:(1)Reviewed the relevant literature,understood camera calibration,image rectification and other preparatory work on the stereo vision system,studied geometric model of stereo vision and construction of u-v disparity(2)Built a mobile robot to simulate vehicle stereo vision system.Got outdoor images by stereo camera on mobile robot.Using Zhang's calibration method to complete the calibration of the camera.SGBM was used to stereo matching,and the dense disparity map was obtained.(3)To solve the problem threshold selection problem in road line detection using standard Hough transform,we proposed using gray weighted Hough transform to model the road.We handle the program directly on gray images,instead of binary images.(4)Proposed a method improves detection accuracy of free space boundary in some specific situation,using dynamic programming(DP)in u-disparity map.We took three columns into consideration at the same time and changed the smooth term,which will significantly increase the smoothness of slope boundary.To satisfy the high speed requirement,the algorithm was optimized and speeded up.(5)Detected vehicles through adaptive threshold.The adaptive threshold will calculate the accumulation of points required for an obstacle based on its distance(disparity)and will be related to the characteristics of the stereovision system.Experiments executed on KITTI data set and Campus data set show that the proposed methods can detected free space and vehicles efficiently.