Hand-gesture Description Based on Skeleton Representation and Hu Moments

摘要

The techniques of Human-Computer Interaction (HCI) are rapidly developed for such application as virtual computer-aided design, human-robot interaction, virtual games, remote control of devices, smart house control, among others. In this research, all stages for visual-based static hand-gestures representation were improved. A hand tracker is built based on spatio-temporal data in a static scene and uses the results of pre-processing by filters, skin classifier, morphological processing, and visual tracking. The proposed robust hand-gesture description includes two main estimations: the topological (approximated skeleton representation) and geometric (Hu moments) descriptions of hand-gesture. The Hu moments provide the accurate gestures recognition with similar topological view. The recognition accuracy using the improved (approximated) skeleton and Hu moments calculation achieved 92–96 %.