Human body modeling with partial self occlusion from monocular camera

Automated human body modeling from monocular video sequences is a challenging task because humans may possess very sophisticated postures with various types of motions. The self-occlusion problem often makes body parts invisible in a period of time. In this paper, we propose an automated contour-based 2D human modeling system which is able to deal with partial self-occlusion problem. The 2D human silhouette is decomposed into several essential parts with a step-by-step approach starting with head extraction, torso extraction, and followed by probabilistic limb configuration estimation. The modeling mechanism only uses basic human kinematic constraints and no predefined motion models are required. In this paper, without loss of generality, we demonstrate two types of human motion in the experiments and show that our approach can successfully extract 2D human body parts with partial occlusions and unpredictable torso orientation under a monocular view. The successful development of this framework can eventually be applied to all kinds of human centric event detection and behavior understanding.