Automatic Detection and Tracking of Suspicious Dismounts

The capability to automatically monitor areas of strategic interest at increased ranges will undoubtedly yield a key advantage on the battlefield by reducing manpower and improving knowledge of enemy target locations. The ability to effectively exploit data from multiple sensor types to find, track, and recognize targets of interest such as dismounts is key to realizing the promise AFSOC advanced sensors. Yet due to a number of well-known challenges, including the large number of target classes and aspects, long and varying viewing range, obscured targets, cluttered backgrounds, various geographic and weather conditions, sensor noise, and variations caused by translation, rotation, and scaling of the targets, effective algorithms for discriminating enemy and neutral targets are still far and few between. An integrated solution which addresses the aforementioned problems in a rigorous, methodical way is necessary to achieve the goals of AFSOC advanced sensors. Our signal processing solution works with many sensors, in many environments, and has matured to the point of natural extension to dismount classification and intent recognition. We propose to augment our tracking solution with advanced machine learning and signal processing algorithms for both appearance-based and motion-based dismount recognition. Furthermore, we propose to recognize enemy combatants through the use of algorithms that can find line-shaped objects consistent with the shape and length of weapons. BENEFIT: Toyon''s approach has successfully tracked multiple, closely spaced targets using multiple sensor types (including EO, I2, MWIR, and LWIR), mounted on multiple platforms (hand-held, building pan-tilt-zoom, and UAV). Successful completion of this work will result in a prototype that can discriminate dismounts, identify weapon-like shapes, and run in real time. Such a system will improve battlefield awareness by emitting prioritized cues only on targets of interest.