Polarization-Based Image Understanding Research for Automatic Target Recognition

摘要

Major developments in a new promising approach to Automatic Target Detection and Recognition have taken place this year building off of previous research developments in Polarization Vision. Preliminary results have already demonstrated tangible enhanced capabilities for RSTA in the UGV program for Detection of heavily occluded military vehicles, woodland camouflage nets, and, augmented features for Target Recognition. Basic research in Polarization Vision has continued with the development of a material classification method having the capability of detecting metallic materials under polarized illumination of clear skylight, complementary in nature to existing methods. New photometric invariants have been discovered for object recognition under multiple light illumination based upon the relationship between the local covariance matrix of photometric values and Gaussian curvature. For 3-D images in Medical Imaging of complex tubular anatomy (e.g., lung bronchial and vascular trees) accurate automated methods have been developed for extraction of central axis geometry which is of large importance to medical researchers.