Research Of Obstacle Avoidance Based On Binocular Vision Patrol Robot

The perception that intelligent robots get from external environment and the path planning based on the perception information is the precondition and the key to autonomous motion.Binocular Stereo Vision can not only perceive the image information of ambient environment,but also can infer the three-dimensional information from the corresponding relationships between two images,which provide a guarantee for completing the missions.The study of this paper is based on vision system,and mainly focuses on the camera calibration and path planning.The main work contains:(1)To deal with the problems about feature point extraction in traditional calibration algorithms,paper proposes a new calibration method by using elliptical fitting.This method combines confocal conics and generalized eigenvalue decomposition.We firstly do elliptical fitting to get the outline of object,then based on the eigenvalue's invariance,we obtain the homography after three projection transformation.Then the intrinsic parameters can be gotten by generalized eigenvalue decomposition(GEVD).Benefiting from the linear model of our method,the process of calibration takes less time and the accuracy is good.(2)Based on the traditional visibility graph method,we take the question about the map building of global path planning into account and propose an improved visibility graph method.The model takes only the starting point,destination and a few of points and lines that would have effects on path planning into account.By using heuristic search algorithm to simplify the establishment of map,we build a visibility graph which have a few line-of-sight between starting point and destination.Simulation experiments show that the method take less time in map building and has fast path searching.Besides,with the help of binocular vision,we can update the map so that we can make sure the intelligent car can achieve the obstacle avoidance.The comprehensive experiment also shows the feasibility of our method.We design the patrol robot with Arduino controller and binocular camera,based on our calibration method and path planning algorithm,it achieves binocular depth measurement,path planning and obstacle avoidance.Experiments show that our method can make system stable,and improve the running speed as well.