The problem of registering two partially overlapping range images taken from different views is dealt with in this paper. The authors describe the RANSAC-based DARCES method, which can solve this problem without any initial estimation. When the shape of the scene data set is fully contained in the shape of the model set, the data-aligned rigidity-constrained exhaustive search (DARCES) algorithm checks all possible data alignments of two given 3D sets in an efficient way, while requiring no preprocessing and no initial estimates of the 3D rigid motion parameters. To solve the partially overlapping 3D registration problem, the random sample consensus (RANSAC) scheme [1] is integrated into the DARCES procedure. For the noiseless case, the basic algorithm guarantees that the solution it finds is the true one, and the time complexity is relatively low. Computation time is reduced by using the constraints provided by the rigidity of the objects. The method can be used when there are no local features in the 3D data sets.
The paper is clearly written, and experimental results for two different scenes are presented.