Automatic Registration of Multiple Range Images by the Local Log-Polar Range Images

We propose a method for automatic registration of multiple range images by matching invariant feature vectors generated from the local log-polar range images. Point pairs are corresponded by finding the nearest neighbor of invariant feature vectors. The correspondence is validated, and the pairwise transformations between the input range images are determined by using the RANSAC algorithm. The registration of all input range images are determined by constructing the view tree of the input range images. The registration result of the proposed method is used as the initial value of a fine registration methods for object shape modeling.