Gesture Recognition On Human Pose Features Of Single Images

Abstract

Enabling robots to read intentions of humans i.e. by detecting a waving person in a restaurant or by pointing to something in a supermarket gives a huge set of possibilities for human robot interactions. Gesture recognition is a challenging problem because of the complexity and the wide variety of human gestures. In this paper, we propose a method for multi-person gesture classification for five gestures and one neutral body posture based on human pose features extracted on 2D images. Comparison between supervised machine learning methods for gesture classification are given. The results showed, that the proposed approach achieves good results on our own validation dataset and generalizes well on a public dataset.