Electro-Optic (EO) Feature-Based Target Combat Identification (CID)

SET proposes to develop a new and powerful deterministic Combat ID system for EO sensor data collected by airborne platforms. In Phase I, we have developed algorithms for reliable feature tracking and vehicle classification using geometric projection invariant descriptors and demonstrated their potentials for greatly improved CID performance under wide variations in camera range and viewing angles using NAVAIR LB-98 sequences and DARPA VIVID sequences. In Phase II, we plan to focus on implementation aspects of vehicle identification to develop a geometric projection invariant descriptor-based vehicle identification system. In particular we will address automated feature extraction, feature matching, target model construction, computation optimization, and overall system performance enhancement. The system semi-automatically extracts a target model from a training image sequence or one or more sample views of the target of interest. The system will then search through an airborne video, extract features, match the features against features predicted from candidate target models (at different poses/angles), and identify the targets of interest using a geometric projection invariant descriptor based vehicle classifier. By the end of Phase II, we will demonstrate an integrated end-to-end vehicle identification system using a commercial grade PC workstation and operate on pre-recorded airborne videos.