A 3D scan matching method based on multi-layered Normal Distribution Transform

Scan matching plays a significant role for 3D simultaneously localization and mapping (SLAM). Before applying the SLAM methods, two 3D data which belong to highly correlated scene has to be registered by finding the correct transformation. In this paper, we introduce a multi-layered (ML) extension of 3D Normal Distribution Transform based scan matching. In this method, point cloud is subdivided into 8ⁿ equally sized cells, where n stands for the level of layer. Unlike the NDT, the score function is described as the Mahalanobis distance. In addition, Newton and Levenberg-Marquardt methods are used to optimize the score function. The proposed method is compared with original NDT, and the optimization methods are discussed. Finally, the performance evaluation is given for experimentally obtained datasets. The approximation provides much faster and long distance measurement capabilities than ordinary NDT.