A Parts-based Method for Detecting People in Various Poses

Abstract

Many scene analysis methods have been proposed in recent years that approach the challenging task of detecting the many people in various poses when they are partially occluded. In this paper, we present a parts-based method that can efficiently detect humans in various poses and states of occlusion. The proposed method segments depth images into small regions; these regions are then classified as human or nonhuman. The method then merges the regions classified as human into one region per person and outputs bounding rectangles for those regions. Experimental results show that this method is more effective than a previous method in detecting occluded people in various poses.