Using Common Field of View Detection for Multi Camera Calibration

Uncertainty-based multi camera calibration from relative poses can basically be applied to multi camera systems in which some camera pairs do not have a common field of view. However, there are limitations to this method. In practice, the uncertainty measures are not perfect and sometimes fail to assign a high uncertainty to invalid relative pose estimates. We hence suggest using a specialized measure to remove camera pairs without a common field of view before calibrating. Furthermore, in case of an ad hoc camera network placed in two or more separate rooms, we first have to separate the cameras before we can calibrate. We use common field of view detection for this task and show that the common field of view threshold can be automatically determined if the number of rooms is known. In quantitative experiments, we show that common field of view detection improves the multi camera calibration and that separating cameras into rooms works well.