Research On In-door Robot Navigation Technology Based On Stereo Vision

With the development of intelligent mobile robots,they occupy an irreplaceable position in the development of national economy.Nowadays,the application fields of mobile robots are more and more extensive in society.As a key technology,mobile robot navigation has become the focus of research scholars.Due to the high flexibility of mobile robots,they work in unknown environments in many cases,therefore it is necessary to select visual sensors as the real-time acquisition unit of mobile robots.This paper focuses on mobile robots,using stereo camera to collect environmental information to achieve robot positioning and navigation.The details are as follows:Firstly,a set of mobile robot platform based on visual navigation is built for the hardware environment of the laboratory.Since the monocular camera could not obtain the true size of the scene,a stereo camera is selected as the visual sensor.The stereo camera is calibrated and the camera distortion is corrected by calibration parameters.On this basis,the stereo visual SLAM(Simulation Localization and Mapping)technology based on feature point method is adopted,and ORB(Oriented FAST and Rotated BRIEF)is selected as the feature points of keyframe matching to estimate the position and pose of the robot.The outliers are eliminated by Random Sampling Consistent algorithm.Secondly,aiming at the problem that sparse point cloud map can not be directly used for navigation,a method for constructing a semi-dense point cloud map is proposed.This method splicing each keyframe by solving the transformation pose of mobile robot and improve the density of the point cloud in the map.The experiment proves that the method of this paper increases the geometrical information of the map while increasing the density of the point cloud,which makes the point cloud map have strong environmental description ability and can be used for navigation of mobile robots.Then,aiming at the problem of high complexity and difficult data processing for 3D point cloud map,a grid map construction method is proposed.By projecting a 3D semidense point cloud map into a 2D plane,the method reduces the complexity of the map without losing environmental information,so that the semi-dense point cloud map can be directly used for the mobile robort path planning algorithm studied in this paper.Finally,for the path planning problem of mobile robots,the search path process of the A* algorithm is studied in detail,and the A* algorithm is used as the path planning algorithm of the navigation scheme to verify the feasibility of the mobile robot navigation system studied in this paper.