Three-dimensional Map Construction And Application Based On Visual SLAM

Simultaneous localization and mapping(SLAM)is important on robot autonomous navigation.This paper has done some work on building an accurate OctoMap based on RGB-D SLAM for 3D navigation of mobile robots and getting more accurate map points with ORB-SLAM.At present,RGB-D SLAM mostly uses two-dimensional map for mobile robots.But in some scenarios,two-dimensional map is not enough for robot navigation.Therefore,paper uses RGB-D SLAM to estimate robot pose and build OctoMap of the surrounding environment.However,outliers inevitably appear in the original OctoMap,which affects robot navigation.This paper proposes an outliers removal algorithm based on K-Nearest Neighbor and Gaussian distribution to remove outliers and build a more accurate OctoMap,which lays the foundation for robot navigation.In this paper,we use the TurtleBot robot to build the OctoMap of the laboratory and corridor.Compared with the OctoMap map constructed by the original algorithm,outliers removal algorithm based on K-Nearest Neighbor and Gaussian distribution can effectively remove outliers and obtain more accurate OctoMap.Aiming at the inaccuracy of ORB-SLAM in obtaining map points by direct linear triangulation method,this paper propose an iterative triangulation method to extract map points,which lays the foundation for building more accurate maps for ORB-SLAM and applying it to the three-dimensional navigation of robots.TUM and KITTI data sets are used in the experiment.Based on ORB-SLAM,the map points obtained by the proposed iterative triangulation algorithm have smaller re-projection error than those obtained by the direct linear triangulation algorithm.In addition,the real trajectory of monocular camera and the real pos of static marker points are recorded by OptiTrack in the laboratory scene.The three-dimensional coordinates of map points estimated by the proposed algorithm are compared with those of the original algorithm.The result proves that the map points by proposed algorithm is are more precise than original algorithm.