Location invariant features for relative hand position classification

Abstract

Hand posture recognition is applied in various research areas such as an automated sign language translation system, and manual-based human-computer interface. One of the problems found in magnetic tracker-based hand posture recognition is caused by change of user body location while using the system. In this paper, by using orientation measures obtained from the 6-DOF location sensing device, a hand posture feature which is invariant to change in a user's absolute body location is derived. This invariance property is achieved by exploiting the constraints imposed by human arm kinematics, as well as by the feasible and typical range of hand postures allowed by most sign languages. Results from the experiment with real measurements are included.