Research And Implementation Of Obstacle Detection Based On Monocular Vision

This paper takes obstacle detection in the driver assistance systems as the application direction and takes images taken from the camera as the processing objects,which studies the obstacle detection based on monocular vision.In this paper we have implemented obstacle detection for a single rear view parking camera,which using the ground points' motion information and the location relationship between the vehicle and camera to recover the motion parameters of camera.The 3-D information of features recovered from the 2-D images are clustered to divide stereo obstacles for the purpose of obstacle detection.This provides potential obstacle information for drivers,which reaches the purpose of getting a warning in advance for the drivers and improve the security in the processing of driving.The main works have discussed as follows:1.Discussed feature detection algorithms.Introduced Harris feature detection algorithm and FAST9 feature detection algorithm.Compared two algorithms' repeatability under three circumstances:image rotation change、image scaling change and image adding noises.The higher repeatability is,the better algorithm is.2.Used Harris algorithm to detect features from images,then used Lucas-Kanade optical flow tracking algorithm based on pyramid to complete features matching.3.According to the location relationship between camera and vehicle,via tracking the ground points and using the RANSAC algorithm to estimate the motion parameters of vehicle.And then recovered features'3-D information based on the principle of stereo vision.4.Achieved obstacle detection via clustering the spatial points.