The purpose of this study is to estimate three-dimensional motions of the sign language with a single camera. In the system, first, a three-dimensional human model with three-dimensional polyhedrons with nine degrees of freedom was prepared, and two-dimensional image sequences with a CCD camera were applied. Moreover, optical flows in three frames before current time were used for the motion estimation, so that the system can be robust to self-occlusions by performer's own face and body. In the experiment, standstill, straight, circular, arc motions and their combinations were adopted for the estimation. As the results, the system realized the gesture estimation and tracking with high accuracy. And human voluntary movement model was constructed based on optimal control. An objective function was define as follows: square of the rate of change of acceleration multiplied by an exponential function of time integrated over the entire movement. Variational calculus was used to obtain the equation expressing the 3-D wrist trajectory. As the result of parameter estimation, even a small difference in a same sign language motion was clear in the parameters.
展开▼
