2D upper human body pose modelling using windowing and template matching techniques

Abstract

This paper presents the upper human body pose modelling using windowing and template matching techniques. Indoor image sequences are used as input. Silhouette and colour features are extracted in the human body parts detection stage. The head and torso pose are estimated using windowing technique. Meanwhile the upper and lower arms are estimated using template matching technique. Each body part is represented by rectangular shape where its sizes depend on the predetermined approximation. The proposed techniques are tested using 30 image sequences from different users. These techniques can overcome self-occlusion, hand crossing and it is adaptive to illumination changes.