Enhanced Active Contour Models for Tracking Shape-Changing Hand Gestures in Taiwanese Sign Language

Abstract

This paper presents a vision-based system for tracking and recognizing the hand gestures that change from one shape to another for conveying a single sign gesture. The widely-used active contour model (ACM) often fails to track the objects with long knobs, such as hands with fingers because of the difficulty progressing into boundary concavities. The enhanced ACM is proposed to overcome the difficulty by combining the gradient vector flow (GVF) and the inflation pressure force as the external forces. The tracked hand shape contours were then characterized by affine-invariant Fourier descriptors for recognition. Transition probabilities between shape changes were included to improve the recognition. The gestures of Taiwanese sign language (TSL) were chosen for test due mainly to the finite set of well-defined hand shapes and trajectories. Experiments showed that the enhanced ACM was reliable in tracking changes in hand shapes and that the system was feasible for recognizing the shape-changing hand gestures.