Spotter: Exploiting Large-Format Imagery via Foveated Visual Search

The proposed Spotter effort develops the algorithms necessary to exploit large format (LF) imagery in an operational environment with layered sensor platforms by cueing high-resolution sensors in order to reduce downstream data volume. In order to support analysts conducting intelligence, surveillance, and reconnaissance missions, image exploitation workstations must have automatic target recognition algorithms (ATR) suitable for LF sensor data which can cue close-in sensor platforms. Spotter addresses this technical challenge by developing a suite of ATR algorithms designed for LF image sources, methods to register and fuse the outputs of these algorithms, and a cueing strategy based on efficient platform tasking. We propose a suite of algorithms inspired by our work in foveated vision systems and ongoing research in human visual search. Spotter will combine ATR results using a method modeled after the top-down theory of human gaze selection. We will quantify the sensitivity of the Spotter approach to variations in LF sensor and platform configuration. Based on our sensitivity study, we will survey available LF imagery and identify sensors for which Spotter is a feasible approach. The Spotter effort builds on a large body of our existing work in innovative techniques for foveated vision and sensor tasking.