3-D Object Creation for Recognition

The problem of registering point sets with rigid motions is fundamental in computer vision. It arises commonly in applications where measured sets of 3-D data must be aligned in a common coordinate system, as in reverse engineering where multiple 3-D scans are combined to construct a complete surface model and in medical imaging where the fusion of information over a range of modalities provides comprehensive diagnostic information. In this work our target application is the task of building 3-D geometric models from multiple range images, which becomes useful in the generation of complete model representations in the database of a model-based recognition system and may also be used in the pose estimation of a single view of the same object from an arbitrary direction.