Research On Autonomous Navigation System Of Mobile Robot Based On Visual SLAM

With the development of computer technology,electronic information technology,and machine vision technology,mobile robots have been widely used in warehousing,restaurants,and personal homes.Autonomous navigation implementation is an essential capability for mobile robots and the basis for performing various other tasks.As the two key technologies of autonomous navigation systems,SLAM and path planning have always been the focus of research in the field of robotics.The SLAM technology that integrates vision was widely studied,because it can accurately sense the surrounding environment,and it was inexpensive.In this paper,the overall framework of autonomous navigation system of mobile robot was established,and it was divided into top-level human-computer interaction module,middle-level navigation algorithm implementation module and low-level motion control module.The experiment platform of mobile robot was built and the hardware components were analyzed,and the software framework of the mobile robot system was established based on the ROS.The Kinectl camera was calibrated with the MATLAB camera calibration toolbox,which laid a foundation for later research.For feature extraction,the SURF algorithm and ORB algorithm have been analyzed,and the ORB algorithm was selected as the feature extraction algorithm according to the experimental results.The FLANN(Fast Approximate Nearest Neighbor)algorithm in the feature matching method was selected to match image feature,and the RANSAC algorithm with erroneous matching was proposed for quadratic feature matching.The EPnP algorithm was used for inter-frame motion estimation,which improved the robustness of camera pose estimation.By analying the loop detection method based on the module of bag of word,the cumulative error of the front-end visual odometer was corrected,and a more accurate camera pose was obtained.By using the g2o to optimize pose,the final optimized pose of the robot was obtained.The overall implementation process of the visual SLAM proposed in this paper was summarized.and the data provided by TUM was used to generate 3D point cloud map and octree map,which verified the feasibility of the visual SLAM system.In order to improve the defects of extra turning points and few search directions in the traditional A*algorithm,the method of A*global path planning using 16-direction search was proposed.Aiming at the defects of local minimum and target unreachability of traditional artificial potential field algorithm,an improved local path planning method for artificial potential field was proposed,and the simulation of related algorithms was carried out in MATLAB.The experiment of visual SLAM was carried out in the corridor environment through the experimental platform that has been establish,which verified the effect of the visual SLAM system proposed in this paper.By using the constructed environment map and the proposed path planning algorithm,the autonomous navigation experiment of the mobile robot was carried out,which verified the effect of the autonomous navigation system.