Key-layered normal distributions transform for point cloud registration

A new scan matching algorithm is proposed using the concept of key layers. In the conventional multi-layered normal distributions transform (MLNDT), the number of layers and iterations per layer are fixed and mismatches in point clouds occur due to the limited number of optimising iterations per layer. Moreover, the accuracy of registration is low and the number of layers is heuristically determined in MLNDT. The proposed key-layered normal distributions transform (KLNDT) works well with both enhanced success rate and accuracy. It is also possible for KLNDT to register in higher layers than the traditional MLNDT.