Automatic Target Recognition Using Perceptual Organization

Period of Performance: 01/11/2000 - 11/30/2000


Phase 1 SBIR

Recipient Firm

Scientific Systems Company, Inc.
500 West Cummings Park Array
Woburn, MA 01801
Principal Investigator
Principal Investigator

Research Topics


Perceptual Organization refers to the ability of a machine vision system to organize detected features, or primitives, in images based on, for instance, Gestaltic criteria. Since some of the original work, which showed that even simple organizations, such as parallel lines and rectangles, can drastically prune the recognition search tree, there have been a number of contributions that demonstrate the importance of perceptual organization for various vision tasks such as object recognition, stereo, motion, image databases, building detection and change detection. This work proposes to develop an ATR architecture based on Perceptual Organization. This architecture, which is built on a Perceptual Inference Network, provides a framework for effectively dealing with ambiguities and the uncertainties and is able to prevent the subsequent secondary errors and artifacts from proliferating along the processing chain. This is because, ATR based on Perceptual Organization is not a feed forward process. Rather, if the subsequent processing affects earlier probability estimates, the process is able to return to these earlier steps in order to update these estimates. Integration of bottom-up processing with top-down feedback loops is crucial to obtain high performance ATR. Likewise, it is important to explicitly represent ambiguities and uncertainties arising from local processing until global inforrmation intervenes to resolve them. The Phase I will demonstrate the use of this new ATR algorithm to recognize an object such as a tank from both visible and infrared wavelengths. The Phase II will continue this development and demonstrate the ability to recognize whether any of several objects are present in a scene where the object may be partially occluded and present in various poses, the lighting on the object may be of varying intensity, and the entire image may be distorted by clutter. This demonstration will be tied to an Army testbed or platform. BENEFITS: The development of the above core technologies in ATR will serve as a foundation for Phase III commercialization. Commercial application of this technology exist in several areas such as: medical screening and diagnosis, remote sensing, road and bridge inspection, and buried waste detection.