Realizing a Machine Vision System for Human Motion Recovery

Abstract A technique is shown tor recovering 3-D shape ot a detomiable object without the employment of the world coordinate system and it is applied to human motion recovery. Settled video cameras take images of the object from multiple directions to yield the measurement matrix in the time lapse and factorization is applied to the matrix to obtain 3-D locations of the feature points on the object. Some experimental results are shown. Integrating human experiences to optimize algorithms is discussed from the view point of understanding human motions to optimize human/robot motion control.