Robust range image registration using a common plane

This paper presents a method for the registration of range image pairs that were captured from different locations. The proposed approach assumes that a common 3D plane is least visible in both scans. Based on this assumption the number of unknown transform parameters reduces from six to three. An orthogonal rectification process is then used to convert the 3D data into 2D depth images and correspondences that are found by a simple 2D image matching process are used to determine the remaining unknowns. We propose a robust plane detection algorithm to detect 3D planes in the scans and describe the algorithm for the computation of the ortho-rectified range images. A correlation based image matching process is used to establish correspondences between the rectified depth images. The experiments are carried out on typical facade data from a city scanning project