On matching sets of 3D segments

The problem of matching 3D objects has been studied extensively in computer vision. Here we consider the case of polyhedral objects extracted from range images and represented in terms of 3D lines. Given two sets of line segments A and B the problem is to find a rigid body transformation T that minimizes a given dissimilarity measure between T(A) and B. We discuss two practical approximate solutions to the segment matching problem in 3D that can be used to recognize planar-faced objects from range data. The first method is based on indexing and the second is a minimization of the Hausdorff distance. We show the advantage of an integrated use of the two strategies improving the overall performance without degradation in the quality of the results.