Motion Prediction Based on Eigen-Gestures

Abstract Motion prediction is important for the realization of “proactive” gesture-based man-machine interaction systems, which can react in various ways before the end of user’s action. In this paper, two motion prediction methods are proposed. The first method is based on a naive extrapolation using a reference gesture. The second method is based on eigen-gestures, which represent typical spatial and temporal variations from the reference gesture. Experimental results show that the second method outperforms the first method because the eigen-gestures are useful to compensate various changes in input gestures.