Map-building Research On Spatial Sparse Feature Based On Stereo Vision

In recent years,visual SLAM has been received great attention by domestic research institutions.There has been a lot of exciting achievement by study on visual SLAM.Map-building,as one important part of visual SLAM,is a scientific research direction of innovation.A reasonable description of the map is significant to localization calibration and path planning of mobile robot.The stereo vision system can directly restore the depth information of the object by the fusion of the two images.Because its advantages of high precision,high speed,simple structure and low price,the binocular vision system has been widely used in the field of mobile robot map building.This paper mainly study the construction of environment map by using stereo vision technology,and proposed a sparse feature map construction method which has excellent performance.With the screening process,the feature points of the map have the advantages of high accuracy,and can provide a solid data support for the real-time mobile positioning and path planning of the intelligent robot.Firstly,this paper introduces the current situation of the research on the field of the robot environment map building and the related technologies,and expounds some important and difficult points in the process of map building.Secondly,this paper studies the application of ORB feature point extraction and matching algorithm in binocular stereo vision image processing,and focuses on the research and improvement of the RANSAC algorithm which can remove the false matching points.Through experiments,the feasibility and real-time performance of the proposed scheme are verified.Finally,we design a scheme,which can construct spatial sparse feature environment map platform.The scheme first extracted and matched the feature points of the left and right frame images captured by the camera.Then the scheme carried out the key steps such as the depth perception of the space,the combination of the tracking,the map building,the optimization and so on.The experimental results show that the proposed method is effective and feasible by using the well-known KITTI data set.