Trajectory modeling in gesture recognition using CyberGloves® and magnetic trackers

The recognition of human gestures is important for several human-computer interaction applications. In this paper, we develop a gesture recognition system that uses the condensation-based trajectory matching/recognition algorithm. The gesture data are collected using a pair of CyberGloves® measuring hand-joint angles and three magnetic trackers that determine 3-D hand positions. The multi-dimensional gesture data are subsequently recognized by matching against trajectory models using probability measures. In our experiments, we evaluate the efficiency of our proposed gesture recognition system using three different gesture sets, viz. directional movements, static hand-shapes and American sign language (ASL) gestures. Experimental results show high recognition rate and signer-independence but less robustness to co-articulation effects.