Hand detection and gesture recognition exploit motion times image in complicate scenarios

Abstract

Hand gesture recognition in complicate scenario is still a challenging problem in computer vision domain. In this paper, a novel hand gesture recognition system is presented. To detect the exact hand target from complicate scenarios, the color and motion clues are used to obtain potential hand regions. And then a method named Motion Times Image (MTI) is proposed to identify the optimal hand location. The R-transform descriptor is used to describe the hand shape features and an offline trained Support Vector Machine with Radial Basis Function kernels (RBF-SVM) is exploited to perform the hand gesture recognition task. Extensive experiments with different users under dynamic and complicate scenarios are conducted to show its high recognition accuracy and strong robustness. © 2010 Springer-Verlag.