Automatic Target Detection and Tracking (ATD&T) Algorithms for Small Autonomous Projectiles

Given a projectile with an onboard imaging device undergoing smooth motion, we can estimate the displacement between the focal plane cen-troid and the desired aimpoint via models customarily employed in video compression to support target detection andmeasurement within small predicted areas (AOIs) of each image frame. Our Phase-I re-search has shown that (1) a prefilter/processor can be applied to raw imagery to select candidate target AOIs, (2) candidate targets can be classified to reject clutter,and (3) motion models can accurately predict target positions in subsequent frames, given the underlying dynamics. This result enables the use of local image processing, sig-nificantly increasing computational efficiency. We have also investi-gatedsensor output compression (e.g., MPEG codec) and the further re-duction of computational cost via the introduction of MPEG motion vec-tors instead of a separate dynamic model. This requires no special hardware, and can be interposed in channelstransmitting compressed or uncompressed imagery.Phase-II will develop: (1) advanced motion-directed ATR software, to run on (2) low volume, low power hardware (COTS components) that support real-time computation and (3) can also process wavefront coded imagery produced by special optical devices thatare currently under consideration for the project.Techniques developed under this topic are applicable to military target detection and image or video compression as well as domestic applications such as telecommunications. The commer-cial benefits for telecommunications are considerable. Other usesinclude faster detection of moving targets in law enforcement, and security applications.