Automated Detection and Cueing

Infrared (IR) imagery provides a rich source of information for target detection, feature-based tracking, and identification. Yet, variable operating conditions, including sensor range, viewing angle, as well as target illumination and degree of occlusion, have so far prevented the development and effective deployment of a complete solution for real-time target identification. While much attention has been devoted to discovering target features, which are invariant to orientation and affine view transformation, the resulting features have proven to be insufficiently discriminatory for large model databases. To address these and other challenges, we propose the use of scale- and affine transformation-invariant spatial and intensity features, combined with spectral histogram features, for robust, high-performance target identification. Further, we show how real-time processing capability can be achieved through efficient candidate target segmentation, followed by a hierarchical classification structure. This framework ensures that the most computationally demanding operations are only performed for a small number of most likely candidate target matches. Throughout, we show how storage and bandwidth requirements can be minimized, to enable deployment of the complete system in a variety of resource-constrained environments, leading to reduced operator workload and improved combat effectiveness.