Recognition of free gymnastics using human body feature points from silhouette and skin region

Many motion recognition algorithms use information that is acquired under forced-conditions such as adhered sensors on body joints or the specialized place where the motion analysis equipment is previously set for extracting and tracking features of human motion. This paper presents a recognition algorithm of human motion in an outdoor environment without any sensors attached on a human body. The process initiates to extract a silhouette as a target from the difference image between the background and the object where a human is in sequence. Skin regions are segmented from this silhouette. A head is determined using the principal axis of a torso that is created by multiple morphological process with respect to the silhouette. The feature points (16 pts) on a body were adopted for tracking human motion in this paper. The points are defined on skin regions and estimated by searching to follow the boundary of the silhouette. The cubic known model is used for calibrating the 3-dimensional space. A complete set is presented for recognizing human motion. This is an analysis method to check the state of feature points on input data while the database has information of feature points for moving objects. Therefore it is suitable to apply a prescript and limited motion. A state transition diagram (STD) is used to measure the coincidence between the input image and standard motions which are known to complete a sequence in the database. Free gymnastics is used as a good example of a complete set and shown as a good result in the experiments.