Motion Analysis by Independent Component Analysis with Phase Difference Information Among Joints

We configured a system that interpolates and extrapolates two different human movements at an arbitrary ratio for both periodic and discrete movement. This could, for example, extrapolate possible future movement from two data points from the past and present for a person with a certain disorder and enable quantitative assessment of the disorder by interpolating or extrapolating two typical movements of nondisabled and disabled persons at an arbitrary ratio and by comparing the result to the movement of subjects. We used independent component analysis involving phase difference information between joint movements. To demonstrate the system’s effectiveness, we generated three different gaits of periodic movement and conducted experiments with and without considering phase differences between joint movement. When results were implemented in a human model in computer graphics (CG) to picture movement, the system considering phase differences reconstructed the original movement naturally even with a number of independent variables as small as two. Movement analysis not considering phase differences, however, was unnatural, especially in ankle movement, due to the lack of appropriate phase differences between the knee and hip. We synthesized and evaluated discrete movement from two winning poses – one powerful and one weak – by interpolating movement I at 50% and movement II at 50%, and extrapolation of movement I at 150%. The former generated discrete movement with an in-between impression and the latter with a powerful impression. These results demonstrate that both periodic and discrete movement can be expressed by a small number of independent variables when phase difference information between joint movements is used appropriately.