DEVELOPMENT OF A VOXEL BASED LOCAL PLANE FITTING FOR MULTI-SCALE REGISTRATION OF SEQUENTIAL MLS POINT CLOUDS
Keywords: SLAM, Point Cloud Registration, Indoor Mapping, Sequential Scan Matching, Voxel-based Registration, MLS
Abstract. The Mobile Laser Scanner (MLS) system is one of the most accurate and fastest data acquisition systems for indoor and outdoor environments mapping. Today, to use this system in an indoor environment where it is impossible to capture GNSS data, Simultaneous Localization and Mapping (SLAM) is used. Most SLAM research has used probabilistic approaches to determine the sensor position and create a map, which leads to drift error in the final result due to their uncertainty. In addition, most SLAM methods give less importance to geometry and mapping concepts. This research aims to solve the SLAM problem by considering the adjustment concepts in mapping and geometrical principles of the environment and proposing an algorithm for reducing drift. For this purpose, a model-based registration is suggested. Correspondence points fall in the same voxel by voxelization, and the registration process is done using a plane model. In this research, two pyramid and simple registration methods are proposed. The results show that the simple registration algorithm is more efficient than the pyramid when the distance between sequential scans is not large otherwise, the pyramid registration is used. In the evaluation, by using simulated data in both pyramid and simple methods, 96.9% and 97.6% accuracy were obtained, respectively. The final test compares the proposed method with a SLAM method and ICP algorithm, which are described further.