Machine learning approach for gesture recognition based on automatic feature selection

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. © 2012 Springer-Verlag Berlin Heidelberg.