Development of Laser Radar ATR Algorithms: Phase 2. Military Objects

Abstract : Under this contract, MTI has developed a software system for the automatic recognition of military targets occurring in laser radar range data. The algorithm is designed to operate over a variety of scenarios: the targets may occur in multiple numbers and aspects, and may be partially occluded due to other vehicles, irregularities in the terrain, or background clutter. The particular algorithm delivered to CNVEO under this contract is configured to detect and classify three particular targets (M35, M60, M113), but the system is based on very general principles concerning invariant rigid body object recognition and easily extended to additional targets, clutter types, sensor effects, and degrees of freedom in the viewing angles. In fact, this work represents an enhancement and special purpose implementation of a generic object recognition system developed at MTI. Initially, this system was developed for the classification of two-dimensional shapes in highly degraded, visible light intensity data; in particular, optical character recognition served as a convenient and challenging prototype problem, and MTI has now developed a commercially viable software product capable of accurate identification of printed characters in noisy and cluttered images, such as high magnification photographs of alphanumeric identifications on silicon wafers.