Extraction of skeleton features using human silhouette and skin

Information that is acquired position by adhered sensors on a body has been commonly used for the three-dimensional motion recognition system. However, if human use adhered sensors, it must be the constrained for their acting and restricted for moving area by those sensors. In this paper, we use optical capture systems like CCDs or movie cameras for overcoming those demerits. Exercising human images, which are taken by optical capture systems in the outdoor, are used in this paper. Silhouette is detected by background subtraction with several morphological processes. Then we extract skeleton features using skin region and silhouette on a human body. In a real-time image sequence, a silhouette of human body is extracted by background subtraction between background and the reference image that has a human posture with upright and unfolded both arms to the each side. In the normal case of human posture, which is upright and looking forward posture, we can easily see three remarkable skin regions. One of them is a face and others are hands or arms. The principal axis is calculated from the torso region, which is extracted by several erosion and dilation process. A skin region that is laid on the principal axis and upper from centroid is estimated to the face. Skeleton feature points, which are essential condition to track the human gesture, are obtained from a skin region, a body silhouette, and kinematical conditions of body structure.