Fast Gesture Recognition Algorithm Based on Superpixel Distribution and EMD Metric

Abstract

This paper presents an efficient gesture recognition algorithm based on superpixel distribution and EMD (Earth Mover’s instance) metric with the help of the Kinect depth camera. In the first step we make full use of the depth and skeleton information from Kinect to accurately locate and segment hands from cluttered backgrounds. In the second step we adopt superpixel distribution to describe gestures and the SLIC algorithm works out superpixels. At last EMD metric is applied to measure the distance between two superpixel distributions. And the FC-EMD algorithm is proposed to calculate EMD distance immediately. Experimental results show that the proposed FSP-EMD can speed up and accurately detect hands, extract gesture features and calculate EMD distance. The running time is smaller when compared to the F-EMD and SP-EMD algorithms.