Edge Enhanced Depth Motion Map for Dynamic Hand Gesture Recognition

Chenyang Zhang, Yingli Tian

Abstract: In this paper, we propose a novel approach to recognize dynamic hand gestures from depth video by integrating Edge Enhanced Depth Motion Map together with Histogram of Gradient descriptor. The novelty of this paper has two aspects: first, we propose a novel feature representation, Edge Enhanced Depth Motion Map (E2DMM), balancing the information weighing between shape and motion, which is more suitable for hand gesture recognition, second, we further employ a dynamic temporal pyramid to segment the depth video sequence to address temporal structure information of dynamic hand gestures. Histogram of Gradient is applied on E 2 DMM to generate vectored representation. Comparison study has been conducted with the state-of-the-art approaches and demonstrates that our approach can achieve better and more stable performance while keeping a relative simple model with lower complexity as well as higher generality.