Research Of Crossing Obstacle Detection Algorithm Based On Monocular Vision

As cars become a part of our life, traffic accidents become a world-wide problem. Obstacle Detection Systems turn into a world-wide important area as systems solving traffic safety problems. Crossing obstacle becomes a hot problem as its importance in the obstacle detection systems. By summing up a large of research methods at home and abroad, a new type of crossing obstacle detection algorithm is presented in this thesis.The algorithm includes two parts:the ego-motion estimation part and the obstacle detection part. The ego-motion estimation part is the first step of obstacle detection algorithm and its main purpose is to estimate the motion parameters of camera. When the motion parameters of camera are estimated, the algorithm can go to the obstacle detection part and realize the final aim of detecting obstacle.In the ego-motion estimation phase, to avoid the influence of complex environment and solid objects, a ground feature extraction method based on inverse perspective mapping is presented. After that, in order to reflect parameters information and reduce effect of wrong ground features, Iteratively Reweighted Least Squares algorithm is used to estimate ego-motion parameters based on ground features.In the obstacle detection phase, an obstacle detection method fitter real scene is presented. When motion parameters are (WY,TX,TZ), components of optical flow corresponding to the rotation of camera is subtracted firstly based on estimated motion parameters, and then road c-velocity space based on image region is constructed to assume and verify the existence of crossing obstacle.The algorithm has a high detection rate of crossing obstacle and is less influenced by complex background. Robustness and accuracy of algorithm are proved by testing a lot of videos in MATLAB software.