Dynamic segment-based sparse feature-point matching in articulate motion

We propose an algorithm for identifying articulated motion. The motion is represented by a sequence of 3D sparse feature-point data. The algorithm emphasizes a self-initializing identification phase for each uninterrupted data sequence, typically at the beginning or on resumption of tracking. We combine a dynamic segment-based hierarchial identification with a inter-frame tracking strategy for efficiency and robustness. We have tested the algorithm successfully using human motion data obtained from a marker-based optical motion capture (MoCap) system.