抄録
In this paper, we propose the technique for posture estimation from monocular image sequences. This technique estimates a posture based on detecting body parts and tracking body parts. Body parts (arms, legs, head, torso and so on) are detected without markers using motion features and a human body structure model. Against the failure of body parts detection, body parts estimated by tracking the body parts detected at the previous frame. A posture is estimated by constructing detected or tracked body parts. This technique does not require a manual operation to determine initial posture at the first frame. In the experiment, we estimated human arm posture using this technique.
