Accurate Link Identification and Posture Measurement from Motion Capture Data
To model human hand motion, it is important to measure and reconstruct human hand postures accurately. Product engineering requires reconstructions to be accurate to within a few millimeters at the finger tip.
An optical motion capture system, in which many reflective markers are put on the skin surface, is used to model hands of various size with adequate degrees of freedom. Due to a hand's high degrees of freedom (about 30), only a limited number of markers can be used to avoid occlusion and subsequent failure of marker identification. In addition, skin movement over a human hand is relatively large for each joint's range of motion. Also, the previous technique of link identification developed for the entire body cannot be used effectively.
Therefore, we propose a link identification method from calibrating motions and associating them with a catalogued set of calibrated motions and related anatomical knowledge.
Link Model of a Hand
When we deal with a hand represented as a link model shown in Figure 1(b), one of the characteristic points is how to deal with the palm. The palm is divided into two parts so that the palmer arch may be studied without directly considering the CM joints.
Calculation of Link Structure from Calibrated Motion
We use marker set as shown in Figure 2 to avoid the failure of marker identification by occlusion. When solving the high-dimensional parameter estimation problem, some simplifications, based on the anatomical characteristics of the link structure, are made to compensate for the decreased marker information (Refer to "Link Structure from Medical Images". In addition, the adverse effects of the large range of skin movement and the relatively small range of joint motion can be avoided by regulating the calibrated motions.
Using the proposed method, the posture used to grasp cylinders with various radii was reconstructed with an average error of 1.6[mm] at the finger tips (Figure 3).