Generalized Principal Motion Analysis: Classification of Sit-to-Stand Motions

摘要

We propose a generalized principal motion analysis (GPMA) method for analyzing temporally evolving motions of redundant systems, such as human motions. GPMA finds base functions, which are called principal motions, that maximally separate distinctive types of motions and that weaken the effects of repeated errors within each type of motion. As an example of human motions, we measured 15 types (3 participants × 5 conditions) of sit-to-stand motions by a camera-based motion capture system. Each type of motion was repeated 10 times. We then compared GPMA and PMA in terms of their ability to classify the type of motions. GPMA correctly classified all types of motions, whereas PMA correctly classified only 81 % of them, which shows that GPMA has a better ability to classify motions.