Human recognition on combining kinematic and stationary features

Both the human motion characteristics and body part measurement are important cues for human recognition at a distance. The former can be viewed as kinematic measurement while the latter is stationary measurement. In this paper, we propose a kinematic-based approach to extract both kinematic and stationary features for human recognition. The proposed approach first estimates 3D human walking parameters by fitting the 3D kinematic model to the 2D silhouette extracted from a monocular image sequence. Kinematic and stationary features axe then extracted from the kinematic and stationary parameters, respectively, and used for human recognition separately. Next, we discuss different strategies for combining kinematic and stationary features to make a decision. Experimental results show a comparison of these combination strategies and demonstrate the improvement in performance for human recognition.