Skip to Main Content
As distributed surveillance networks are deployed over larger areas and in increasingly busy environments, limiting the computation, bandwidth, and human attention burdens imposed is becoming critical. We describe a system addressing this problem which uses layered, in-network processing on each camera to filter out uninteresting events locally, avoiding disambiguation and tracking of irrelevant environmental distractors. Coupled with this is a factor-graph-based resource allocation algorithm which steers pan-tilt cameras to follow interesting targets while maintaining a "peripheral awareness" of emerging new targets. We describe this distributed attention mechanism and our implementation of this high-level architecture in a video sensor network.