Human Action Recognition Algorithm: Risk Notification Service on the Android Environments

Abstract

This paper proposes a human action recognition algorithm which can be efficiently applied to a real-time intelligent surveillance system. The proposed method classifies human actions into walking, sitting, standing up and unusual action like faint and falling down. Also, in the case of detecting unusual actions, it offers an alarm to smartphones to monitor the object of interest. This method models the background, obtains the difference image between input image and the modeled background image, extracts the silhouette of human object from input image and recognizes human actions. In order to recognize human actions, the proposed method uses direction vector of movement and link codes of movement dependent histogram (LC-NMDH). Firstly, NMDH is computed by dividing the motion information histogram into ten parts and saving the median value of each part. LC-NMDH is defined as the values which records a chain code of NMDH of each part. Human actions are classified into walking, sitting, standing up and unusual action. The proposed method was examined on seven people through sequences captured by a web camera. The proposed algorithm efficiently classified human actions, detected unusual actions and provided an alarm service. The test result showed more than 99 percent recognition rate for each action by the proposed method.