Point Cloud Registration Using Intensity Features

• Published in: Sensors and materials 2020-07, Vol.32 (7), p.2355
• Main Authors: Lin, Chien-Chou, Mao, Wei-Lung, Hu, Ting-Lun
• Format: Article
• Language: English
• Subjects: Alignment; Computer simulation; Coordinates; Field of view; Iterative algorithms; Lidar; Registration; Three dimensional models
• ISSN: 0914-4935
• DOI: 10.18494/SAM.2020.2808

In this paper, a registration method for extending point clouds is proposed. The proposed method merges several point clouds to increase the vertical field of view (FOV). However, the most popular alignment algorithm, iterative closest point (ICP), fails to extend point clouds that are captured with varying heights when most points are similar. The main issue is the tyranny of the majority, in which ground points and wall points dominate the registration result of ICP. Instead of using all points of point clouds, the proposed method only uses the intensity features to find the transformation matrix between two point clouds and then transforms the target point cloud to the coordinate system of the source point cloud. Upon merging the two point clouds, the vertical FOV can be extended. In a simulation, the proposed algorithm scans the source and the target with fixed position and varying height using a light detection and ranging (LiDAR) (Velodyne VLP-16 mounted on a tripod). The simulation result shows that the average error of alignment of the proposed system is less than 16 cm in a 6 × 6 m2 meeting room, and the average error of alignment of the proposed system using a premeasured height for compensation is less than 12 cm.