Distributed Attention for Large Video Sensor Networks