Research on Mobile Robot Navigation Based on Stereo Vision

摘要

A novel method for vision navigation is presented.Considering the complexity of real-time and the real road,the characteristics of the normalized RGB value are used to divide the road image into road region and dubious obstacle region.The region of water and crack are eliminated by reprojection and the correlation of the pixel gray.With the correlation matching and constant-local-disparity,real matching points are found which can be used to produce the maximal encasing box for the obstacle on the ground.The relative velocity of the obstacle is calculated based on the mean and variance of the hue.The least change principle for one obstacle and the optimal principle for multi-obstacle help the robot to avoid the obstacle. Simulation results demonstrate the effectiveness of the obstacle avoidance in this method.