The Research On Robot Obstacle Detection Methods Based On Stereo Vision

Stereo vision is one of the most important research direction of machine vision andbinocular stereo vision is a way to simulate eyes directly how to deal with scene. Based ontriangulation principle, binocular stereo vision technology can reconstruct the 3-d informationof an object by processing its images. It is used in a wide range, and has become the researchfocus in the field of stereo vision. In the field of robots study, fast detection of obstacles is oneof the key technologies to realize the robots autonomic navigation. Therefore, adoptingbinocular stereo vision to further study robot obstacle detection methods is significant. Wehave researched the key technologies of binocular stereo vision. Detecting and locatingobstacles existing in the area of robot forward is the foundation for autonomic navigation andobstacle avoidance.Based on the research of robot obstacle detection methods, this paper mainly includes thefollowing content: binocular camera calibration, stereo images matching, 3-d reconstructionof obstacles and so on.Accurate camera calibration is the base to realize obstacles detection with stereo visiontechnology. This part introduces the camera model, camera calibration principle and somecommon camera calibration methods in detail. In this paper, the improved calibration methodof Zhang is used for camera calibration, this method with good reliability, high accuracy andsimple process, can mainly meet the demands of binocular vision system. In order to satisfythe real-time characteristic of robot obstacle detection, this paper selects a stereo matchingmethod based on SIFT operator, and combines with the polar line constraint conditions. Thiscan reduce the search space from 2-d to1-d, then uses the 128-d SIFT descriptor for the match.During the experimental process, the binocular vision system has been placed parallelly, in the3-d reconstruction process of the stereo matching points, on the basis of the projectiontransformation matrix, we adopt a method based on projection linear equation to calculate its3-d coordinates, then realize obstacle detection and location.Proved by the Matlab simulation experiment, the robot obstacle detection method basedon binocular stereo vision in this paper has good real-time characteristic and higher accuracy.Generally speaking, this method proposed in this paper can be applied to other binocularvision system well. It is beneficial for the follow-up study.