A system for the recognition of the hand posture against uniform background on monotonic image sequence is presented. The system employs the characteristic of the hand structure that reduces the amount of the necessary information that can be acquired from features on hand images. The characteristic of the hand structure has already been successfully employed for the recognition of the hand posture. By combining this characteristic with the fingertip and the fingernail information, all finger joint angle could be calculated.
Our system reaches the recognition rate of 4 frame per second with no cumbersome device of marked glove. The proposed system allows robust recognition against the occlusion phenomena caused by finger folding.