HMMs-based human action recognition for an intelligent household surveillance robot

Abstract

The aging of population has become a social problem and fall is a major health risk in the elderly. To this end, this paper presents a novel approach for fall detection applied to an intelligent household surveillance robot. Silhouette based features are extracted, including aspect ratio of minimal bounding box of the human silhouette, approximated elliptical eccentricity, normalized central moments and Hu moments. Fall and other human motions, such as walk, bend, run and crouch, are modeled using Hidden Markov Models (HMM) with Gaussian Mixture Models (GMM). The experimental results are evaluated by sensitivity, specificity and accuracy and the average of them reaches 88.71%, 97.56% and 96.26% respectively.