Objective The terrestrial laser scanner can flexibly and accurately obtain three-dimensional (3D) point cloud data of ground objects. Thus, it is widely used in heritage protection and restoration, building 3D model reconstruction, and deformation monitoring. In practical applications, owing to its limited field of view, the scanner should conduct multistation and multiview scanning to obtain complete substation cloud information of the object to be measured. Therefore, the registration of a multisite cloud is necessary to obtain the complete point cloud information of the object to be measured, which is also the premise of scene reconstruction, segmentation, and classification. The traditional point cloud registration methods, such as the registration method using target or landmark points, rotating platform method, automatic registration method based on point cloud features, and auxiliary registration method using indoor Global Positiaoning System (iGPS) world coordinate system, usually face problems of weak robustness and low efficiency. In this study, a registration method of the terrestrial laser scanning point cloud based on a laser tracker is proposed to overcome the above defects.
Methods In the proposed method, the position sensor of the terrestrial laser scanner is turned off, so that the scanning coordinate becomes an independent coordinate system relative to the scanner body. Four spherical prism bases of the tracker are fixed on the body of the 3D laser scanner. The coordinates of the four base points in the scanning coordinate system are calculated using a “calibration” process. Because the scanning coordinate is an independent coordinate system that remains unchanged relative to the scanner body, the base point coordinates remain unchanged in any scanner station coordinate system. After the scanner moves, the laser tracker is used to measure the coordinates of the four base points in the tracker coordinates. Combined with the “calibration” results, the relative rotation and translation matrix of the scanning coordinate system of any station can be solved. Additionally, a mathematical model of the transformation relationship between the scanning coordinate system before or after the moving station can be established to achieve accurate registration of the multiview point cloud.
Results and Discussions In the indoor environment, a small industrial device and a gypsum statue are scanned from four stations. The proposed registration method is used for coarse point cloud registration. The coarse registration results (Figs. 7 and 9) showed that the proposed coarse registration method can accurately register two adjacent point clouds. After iteration closest point (ICP) fine registration of small industrial devices, the coarse registration effect on the upper edge of the device (yellow box in Fig. 6) is better than that of the coarse registration, which can be seen by naked eyes; however, it is not noticable. No significant difference between the coarse and fine registration results of gypsum statues is observed. Thus, the coarse registration effect of the proposed method worked perfectly, which can provide a sufficient initial pose for fine registration. Tables 1 and 2 show that the maximum coarse registration error of statue is 0.71 mm, with an average of 0.59 mm; the maximum coarse registration error of the small industrial devices is 0.68 mm, with an average of 0.64 mm, and the coarse registration result with high accuracy is obtained. Thus, the ICP algorithm only needs a few iterations to obtain accurate registration results. After ICP fine registration, the error is less than 0.2 mm.
Conclusions Aiming at the registration problem of multistation terrestrial laser scanning point cloud, a method that uses a laser tracker to assist the registration is proposed. By placing a spherical prism base on the scanner body, the position transformation matrix of the scanner before and after moving the station is calculated. A mathematical model of the transformation relationship between the scanning coordinate system before and after the station moving is established. Using this model, the data of multistation ground laser scanning point clouds are unified to the same coordinate system, and the coarse registration of multistation scanning point clouds is completed, providing a good (coarse registration error is not greater than 0.71 mm) and robust initial value for ICP algorithm. The proposed method is easy to operate and has high accuracy. Without opening the scanner position sensor, only one “calibration” is needed in the entire measurement process.
