People correspondence in multiple-camera setup

To track people successfully in multiple cameras, one needs to establish correspondence between objects captured in each camera. In this paper, a framework for tracking human motions from sequences of color images in a fixed uncalibrated multi-camera setup is presented. We propose a novel scheme based on artificial positioning data (APD) to find the relationship between individuals in each camera scene by determining the valid positions of people. Thus, the system is independent of the fields of view (FOVs) and yet able to identify each individual consistently throughout the scenes. The proposed approach is fast and computationally efficient.