A Coarse to Fine 3D Registration Method Based on Robust Fuzzy Clustering

An important problem in computer vision is to determine how features extracted from images are connected to an existing model. In this paper, we focus on solving the registration problem, i.e. obtaining rigid transformation parameters between several 3D data sets, whether partial or exhaustive. The di culty of this problem is to obtain a method which is robust with respect to outliers and at the same time accurate. We present a general method performing robust 3D localization and tting based on a fuzzy clustering method. The fuzzy set approach is known for its practical e ciency in uncertain environments. To illustrate the advantages of this approach on the registration problem, we show results on synthetic and real 3D data. 2 List of symbols alpha. beta. IR special double R, set of real number. P mathematical sum, sigma. standard deviation of a Gaussian noise. 8 mathematical for all. p mathematical square root. (R;T ) rigid transformation: rotation R and translation T . 3