Artificial neural networks for 3-D rigid motion analysis

A novel approach to 3-D rigid motion analysis using artificial neural network techniques is proposed. This approach differs from previous works in the ways it tackles the problems of matching and motion parameter extraction among two or multiple frames. A two-dimensional Hopfield network is configured to enforce matching between two frames, and a three-layered learning network is constructed to extract motion parameters with fast implementation speed and flexibility in processing information from multiple frames. Experiments on both synthetic and real data sets are conducted to corroborate the proposed techniques.