Real-time multi-hand posture recognition

Abstract

A new method based on Haar-like and topological feature is proposed for multi-hand posture recognition in this study. Initially, a statistical method is used to detect the region of the hand according Haar-like features and color segmentation technique. With this method, a group of hand posture regions can be detected in real time with high recognition accuracy. Then, the topology is applied on the detected regions in order to classify the different postures. The experimental results show that the new approach achieves satisfactory performance when applying this new method to human-robot interaction.