Design And Implementation Of Indoor Robot Navigation System Based On Binocular Vision

With the development of robot,visual positioning and navigation will be one of the key technologies to be solved in the future.Binocular vision positioning and navigation as an important part of vision positioning and navigation will be focused on.Because the outdoor scene is greatly affected by various factors,indoor positioning and navigation has been attached great importance.This paper is based on binocular vision to study the design and implementation of indoor robot navigation system,aiming at reducing costs and improving navigation positioning accuracy.Only using binocular camera to collect information has achieved good results in theoretical analysis and application.This paper studies and optimizes the indoor location method based on binocular vision.Firstly,based on ORB-SLAM2 framework,a dense key-frame map of non-real-time robot operating environment is created;secondly,binocular camera calibration is carried out and the calibration parameters are used to correct the distorted coplanar alignment of the image;thirdly,ORB features are extracted from the image and sub-pixel matching of left and right image feature points is carried out using SAD algorithm to solve the depth information;secondly,word bag dictionary or projection are used.Shadow model compares the similarity of feature points to match the feature points of inter-frame images or the road marks of key frame maps.Ran SAC and EPn P algorithm are used to estimate the camera pose using matched feature points.G2 O is used to optimize the camera pose by minimizing the re-projection error of matched feature points.The number of matching points between image feature points and key frame maps is increased by tracking the road marks of key frame maps.Secondly,the matching point pair is used to optimize the camera pose.Finally,Local back-end optimization is used to optimize position and posture again and create odometer information.When the inter-frame image matching fails,the key frame map created is used to relocate.When the relocation fails,the relocation decision is made by the mobile robot.To obtain the camera posture by relocation,the bag dictionary is used to determine the camera posture initially,and then the feature point matching is used to optimize the camera posture.The design and implementation of indoor robot navigation system based on binocular vision firstly uses calibration parameters to stereo rectify the images captured by binocular cameras to obtain the deformed and coplanar alignment images;secondly,SGBM is used to calculate parallax,create depth images,create point cloud images,and simulated radar data for coplanar alignment images.In order to detect the surrounding environment information,TF coordinate transformation tree is used to manage the transformation relationship of robot coordinate system.Then,based on Gmapping framework,odometer information,radar information and robot coordinate transformation information are used to create grid map and publish it in ROS.Finally,simulated radar data,odometer information are received.The coordinate system transformation information,rast er map information and target information of the robot are first used for global path planning using A* algorithm,then local path planning using DWA algorithm,and motion commands are issued to control the robot running through the controller to complete navigation.According to the scheme proposed in this paper,firstly,the research of binocular vision odometer is completed and compared with AMCL positioning method in reality and simulation.The positioning method in this paper is superior to AMCL positioning method in accuracy and stability,and achieves good results.Then,different binocular cameras are used to create raster maps on different frameworks such as Cartographer,and their stability and accuracy are measured.The results show that the raster map created by Gmapping framework can achieve the desired goal.Finally,the indoor robot navigation system can be tested by using various binocular cameras in different environments to plan a real-time path from the current position to the target position and accurately detect obstacles to avoid obstacles to complete the navigation function.