This paper deals with the problem of building a 3-D model of the environnment for a mobile robot by combining several 3-D images obtained from stereo vision. A new method for 3-D image matching is described. It is based on a coarse to fine approach allowing a good initial estimation of the robot motion. We show the importance of feature grouping and how feature groups can be used to guide the search. The experimental results demonstrate a significant reduction in the complexity of the matching problem and low sensitivity to noise.