I. Introduction
Human body pose recognition is an important task in computer vision which recognizes human body parts and locates them according to data from wearable sensors or different types of cameras. While traditional marker-based pose recognition [1] and motion capture system successfully capture human motion, the need of marker and the high hardware cost restrict its applications. Another solution employs multi-camera system to construct volumetric human model and derive human pose which requires professional surrounding camera array and heavy computation to reconstruct human model from multiview images [2]. This is inconvenient for many applications. Therefore, a robust marker-less pose estimating and tracking system from single view camera has considerable applications including human-computer interaction, game, smart surveillance, character animation, robotics, visual reality, and video analysis.