Automatic visual recognition of armed robbery

We propose a method by which to analyze silhouettes and recognize a classic holdup position of armed robbery. In such a situation, one actor levels his or her arm while another actor raises his or her arm(s) into the air. The core of this algorithm is skeleton analysis. We attempt recognition by first segmenting the skeleton of the silhouette into separate pieces of the body, then identifying the positions of the arms. We show that our algorithm correctly utilizes skeletons to identify parts of the human body and recognize these holdup positions.