3-D object recognition with symmetric models: symmetry extraction and encoding

Object recognition systems which employ solid models and range data have been a topic of interest for several years. Model databases have the potential to become large in some environments. This paper proposes a pair of techniques for incorporating knowledge of the symmetries of object models into the recognition process. The effects of symmetric models on the speed of an object recognition system is examined in the context of an implemented system employing invariant feature indexing as a correspondence-building mechanism. Groups of model surfaces are enumerated and examined to yield a list of segment label permutations which summarize the model´s symmetry. This symmetry extraction process is followed by a symmetry encoding procedure which replaces groups of features which are indistinguishable because of symmetry with a single prototype feature group. Experiments with a large model database demonstrate the utility of these symmetry extraction and encoding techniques.<>