Machine Learning Approach for Gesture Recognition Based on Automatic Feature Selection

Xiubo Liang, Franck Multon, Weidong Geng

Abstract: In this paper we propose a machine learning approach to design strong classifiers based on the most relevant combination of 1444 weak classifiers based on pose parameters. This classifier is embedded in a three-layers recognition system which enables us to recognize 70 different gestures performed by various users with high style variability; the recognition ratio is 97.5% with our approach.