Real-time human action recognition with Extreme Learning Machine

In this paper, we propose an effective and high accurate method for RGB-D human action recognition in real-time. We simultaneously classify data extracted from both RGB video and depth data. Two Extreme Learning Machines (ELM) are used to improve accuracy and a decision-fusion defines the class of action. This method can analyze each frame in 25ms by the accuracy higher than 95%. The results show that our method can effectively perform in real-time.