Recognizing Hand Gestures with CALI

Abstract

Human computer interaction techniques that do not rely on devices are often perceived as more natural by users. Many of these, include hand pose recognition as an interaction technique appealing to users. In this paper we describe and evaluate two techniques for hand pose recognition, based on CALI, a general library for gesture recognition. This library was initially designed for calligraphic recognition, however recent usage shows that CALI is able to support other applications. One unexplored research area includes its application to hand pose recognition, even though there are already different approaches to the subject using techniques such as Hidden Markov Models or Model-based tracking. We developed and tested a new approach to recognize hand poses taking advantage of the features obtained from CALI. To explore this approach we implemented two techniques. The first recognizes bare-hands by their outer contours, the second uses color marks on each fingertip to track the hand and recognize its pose. Experimental results show that both approaches present recognitions rates around 93%.