A method for the automatic registration of terrestrial laser scanner point clouds

As part of a system being developed to manage large terrain reconstructions from laser scanner point clouds, a methodology based on the use of spin images to automatically pair-wise register point clouds has been implemented. Individual point clouds are meshed via Delauny triangulation, the mesh is filtered to exclude undesirable points from further processing, importance values are determined to identify salient surface features, spin images are extracted from each point cloud and then conjugate points established by image correlation, similarity measures and geometric filtering. The registration is implemented as a simple 3D conformal transformation. Initial testing on three datasets was done with the developed system and the results compared to those produced by the Cyclone point cloud software. The results indicate that the developed methodology is reliable but could be improved with a more sophisticated orientation strategy such as that provided by an iterative closest point algorithm.Department of Land Surveying and Geo-Informatic