Automatic sign categorization using visual data

Abstract

This paper presents a method of visual tracking in recordings of isolated signs and the usage of the tracked features for automatic sign categorization. The tracking method is based on skin color segmentation and is suitable for recordings of a sign language dictionary. The result of the tracking is the location and outer contour of head and both hands. These features are used to categorize the signs into several categories: movement of hands, contact of body parts, symmetry of trajectory, location of the sign.