Detection of hand-raising gestures based on body silhouette analysis

This paper introduces a method for hand-raising gestures detection based on human body silhouette analysis in indoor environments. Past approaches have detected the gestures for isolated persons or seated persons. Our method can deal with moving persons in crowd. First, background subtraction based on integration of intensity histograms with codebook of color feature is employed to segment human bodies. And then, to deal with movements, nonrigidity and partially occlusions of human bodies, the silhouette analysis, including candidate regions (CR) search, shape feature extraction and classification, is applied to search for raised hands. Shape features in each CR instead of the entire silhouette are extracted through R-transform. At last, a hierarchical hand-raising gestures detector, consisting of two classifiers which are learnt using SVM, is used to determine whether each CR contains raised hands. Experiments show that this method can detect hand-raising gestures well, even in crowded scenes.