Multi-camera surveillance with visual tagging and generic camera placement

A common goal in many vision applications is to identify and track human objects with distinctive visual features or "tags". Examples range from identifying distinct soccer player by his jersey number to locating the face of an individual that produces a match in a face recognition system. In this paper, we made two contributions to this "visual tagging" problem. First, we propose a general framework for camera placement. This framework can measure the performance of any particular camera placement using simulation method. The optimal placement strategy can be obtained by iterative grid-based linear programming. Second, we focus on tracking specific colored tags used in a privacy-protecting visual surveillance network. By building a color classifier for tag detection and using epipolar geometry between multiple cameras for occlusion handling, our proposed system can identify, track and visually obfuscate individuals whose privacy in the surveillance video needs to be protected.