In this paper, we propose a novel approach to automatically assess the performance of video sensors in a a multi-camera system for video surveillance. Performance is measured through the quality of the segmentation obtained from each sensor. The key idea of this work is that no reference segmentation is assumed to be present, but the redundancy offered by the multiple cameras is exploited to construct the reference segmentation at each time instant. In this way, data coming from the different cameras may be weighted accordingly and fused. Encouraging experimental results have been obtained on real video sequences
