Research On Indoor 3D Map Building Method Based On Vision SLAM For Mobile Robot

Simultaneous localization and mapping(SLAM)technology is the key for robots to achieve autonomous navigation,and SLAM technology mainly includes two solutions: SLAM based on vision camera and SLAM based on lidar.The vision camera is small and easy to acquire rich environmental information so that it is more suitable for the realization of the SLAM solution of the mobile robot platform.This subject has carried out the following theoretical and experimental studies on 3D map building method based on visual SLAM:1.Designed and built a 3D map building platform based on visual SLAM for mobile robot.On the basis of analyzing the actual environment of map building,this topic builds a hardware system based on binocular stereo camera and a software system based on ROS,uses an embedded development board to process the underlying data,and completes an indoor 3D map building platform for mobile robot.2.The method of acquiring 3D point cloud data based on binocular vision is studied.This subject first analyzes the optical model and distortion model of the vision camera and calibrates the camera,and realizes the stereo correction of the left and right camera image pairs.The stereo matching of the image pairs is completed based on the epipolar line search.A single-frame depth map is obtained as well as three-dimensional point cloud map by binocular distance measurement3.The method of pose estimation and optimization between two adjacent frames are studied.Aiming at the shortcomings of the descriptor method and the direct method in the tracking of key points,the method of combining key points and gray-scale tracking is studied to realize the matching and tracking of key points.The 3D-2D matching and 3D-3D matching pose estimation and optimization methods are studied,and the two matches are combined to realize the pose restoration between two adjacent frames of images.4.Researched indoor global 3D map building method.Aiming at the shortcomings of obtaining poses between two adjacent frames of images,the method of matching poses between current image and key point atlas to obtain poses is studied,and point cloud map registration is realized based on pose restoration.Key image frames are selected to construct a streamlined point cloud map.In order to verify the feasibility of the key point matching tracking,point cloud map registration,and global map construction methods studied in SLAM construction,this topic carried out corresponding experimental research on the mobile robot map construction platform,and the results show that it can guarantee more efficient tracking of about 1700 key points,global optimization of posture to achieve better point cloud map registration,the constructed environmental point cloud map error is about 1.72%,which verifies the feasibility of the method Nature,the constructed map can accurately and truly reflect the environmental information.