MIPOSE

Abstract

Giving computers the ability to learn from demonstrations is important for users to perform complex tasks. In this paper, we present an intelligent self-learning interface for dynamic human pose recognition. We capture 20 samples for an unknown pose to train a stable generative adversarial networks (GAN) system which aims to conduct data enhancement, then we adopt a threshold isolation method to distinguish relatively similar poses. A few minutes of learning time is sufficient to train a GAN system to successfully generate qualified pose samples. Our platform provides a feasible scheme for micro-intelligent interface, which can benefit to human-robot interaction greatly.