Method to Automatically Register Scattered Point Clouds Based on Principal Pose Estimation

Three dimensional (3-D) modeling is important in applications ranging from manufacturing to entertainment. Multiview registration is one of the crucial steps in 3-D model construction. The automatic establishment of correspondences between overlapping views, without any known initial information, is the main challenge in point clouds registration. An automatic registration algorithm is proposed to solve the registration problem of rigid, unordered, scattered point clouds. This approach is especially suitable for registering datasets that are lacking in features or texture. In general, the existing techniques exhibit significant limitations in the registration of these types of point cloud data. The presented method automatically determines the best coarse registration results by exploiting the statistical technique principal component analysis and outputs translation matrices as the initial estimation for fine registration. Then, the translation matrices obtained from coarse registration algorithms are used to update the original point cloud and the optimal translation matrices are solved using an iterative algorithm. Experimental results show that the proposed algorithm is time efficient and accurate, even if the point clouds are partially overlapped and containing large missing regions.