Camera network topology estimation by lighting variation

Abstract

The goal of this paper is to find connections between cameras in a large surveillance network. As a proxy for camera pairs whose fields of view overlap spatially, we find pairs that are affected by a common light source. We propose multiple measures of lighting variation and show that we can reliably detect nearby cameras even without direct overlap. The relationships discovered by our process can be used for problems such as automated tracking and re-identification across large camera networks. We demonstrate our method on a campus network of 20 and 26 cameras and evaluate its accuracy and performance.