Recognizing Target Variants Using Transformational Adaptivity

One of the greatest obstacles to robust Automatic Target Recognition (ATR) is achieving a high level of performance in operating conditions outside those for which the system is designed. The bio-mimetic map-seeking circuit (MSC) has provided remarkably simple solutions to extending operating range for recognizing 3D targets allowing recognition from any viewpoint, tolerating clutter and distractors anywhere in the field of view including near the target, up to a substantial degree of occlusion. Its performance is negligibly impaired by imagery resolution down to fewer than a dozen cycles on target. Nevertheless, map-seeking has been limited, as other model-based vision approaches have been, to recognizing targets exactly or highly similar to the stored models. General Intelligence Corp proposes, as a component of a general purpose ATR system, an extension to the map-seeking approach to 3D object recognition which will allow plausible variants of stored models to be recognized. This extension of the map-seeking circuit's abilities, termed "transformational adaptivity," will make it possible to recognize articulations, plausible morphs and aggregations of known target models on the fly. As important, this solution will be able to report the parameters of the variation for further stages of decision-making by machine or human operator.