3D LiDAR-Based Point Cloud Map Registration Using Spatial Location of Visual Features

In this paper, a novel approach for the registration method of 3D point cloud maps is presented. During the operation of unmanned system, the construction of 3D maps for the environment is an important prerequisite for many navigation tasks. Generally 3D maps are widely utilized to recognize the location of the unmanned agent. Traditional map aligning techniques that use ICP (Iterative Closest Point) method relate points having the closest distance in different cloud maps in order to combine and extend the 3D maps. The method can be easily adopted in registration among sparse point cloud maps or consecutive scanning problems. However, this approach takes long computation time when aligning large scale maps including a lot of points. Therefore, an improved 3D point cloud map registration method is proposed to register precisely and effectively two maps using low-cost cameras. By combining odometry information derived from the SLAM (Simultaneous Localization and Mapping) procedure and the 3D position of the selected image feature points, location of the coexisting places in both maps are extracted. Then, the estimation of rigid body transform between the origins of each map is achieved. The effectiveness of the presented method is quantitatively validated by experiment on challenging instances of the merging problem and comparison with an existing registration method.