A general framework for tracking objects in a multi-camera environment

We present a framework for real-time tracking of objects. Our system consists of multiple cameras and a control unit that communicate through a network. Each camera has a general-purpose processor and a reconfigurable hardware unit embedded in it. Therefore, some computation can be performed at the point of data collection. We argue that collocating the computation with the data at vision sensors can improve performance, communication overhead and network scalability. We exploit the parameterization and tuning of the vision algorithms and present a sample tracking application implemented on our framework. We further argue that the proposed architecture can be used to implement many other real-time vision applications through hardware reconfiguration.