Localizing a Network of Non-Overlapping Cameras

We describe a method for recovering the extrinsic calibration parameters of non-overlapping cameras in a multi-camera system. Non-overlapping camera networks can cover a much wider area than overlapping configurations, but calibrating them is more challenging. To compensate for the lack of overlap, our approach assumes that the calibration target follows a smooth trajectory over time, as described by a stochastic dynamics model. Our algorithm then tracks the target when it comes in the field of view of each camera, and searches for calibration parameters and target trajectories that are consistent with the observed tracks. We demonstrate the idea with a network of indoor wireless cameras.