Deductive Approach to Processing High-Level Video Activity Queries in UAV Networks

Abstract

We envision a network of airborne video sensors (AVSs) being used for detection of unusual activities in emergency and disaster control situations, monitoring of unplanned events, etc. In this preliminary work, we address the problem of declarative specification and optimized evaluation of high-level video activities. In particular, we propose use of a declarative deductive framework, wherein high-level activities are defined in terms of the low-level events using deductive rules. Here, the low-level events are the primitive events generated at individual AVS nodes using computer vision techniques. In our envisioned system, the high-level activities are essentially translated into a distributed collection of AVS operations run on individual AVS nodes. We propose certain optimization opportunities that prune the set of targets being tracked or events being detected by an AVS node. In effect, our envisioned system can form a basis of macro-programming of AVS applications and automatic translation into energy-efficient distributed code.