Tracking articulated hand underlying graphical model with depth cue

Visual tracking of articulated objects in real 3D space is challenging with applications in advanced human-computer interfaces and gesture semantic understanding. In this paper, graphical model, constructing articulated human hand, and NBP algorithm embedded with CAMSHIFT, inferencing hand configuration in 3D space, are applied for visual hand tracking. We also introduce image depth cue captured by two calibrated cameras together with color and edge as observation model. Depth cue allows our method to track human hand in comparatively accurate distance from cameras, especially in cluttered scenes with objects of similar color or edge. All those image cues are converted necessarily to probability distribution applied for the graphical model frame. The proposed method promotes the tracking efficiency and robustness proved by experiments and theory analysis.