SBIR-STTR Award

Six multisensory (FLIR/MMW) data bases will be accessed to include all those at CNVEOD including the VISION 1 and subsequent field tests, particularly those involving the RIMS and BWI MMW radars, along with the IIDAS and SAIRS FLIRS. Data from other government agencies will include that in TABILS and at HDL, MICOM, and AATD. A Phase II plan will be produced for training and testing three ATR algorithms not heretofore considered for tank-mounted applications but proven in related contexts: 1. An adaptive Bayesian statistical classifier, which in its operational implementation would be continually retrained upon the actual ground clutter within which the tank is operating. 2. A "genetic" algorithm, which uses the same feature sets as #1, but evolves a different polynomial discriminant function. Both the "genetic" and "adaptive" classifier algorithms are under development for the LONGBOW MMW radar by Systems Dynamics, who would be subcontracted to I-MATH for this SBIR. 3. Geometric hashing algorithms, which I-MATH has recently developed using a CNVEOD-provided FLIR imagery under a Phase II Air Force SBIR. Such algorithms provide a robust mechanism for multisensor fusion as well, since each feature is indexed by its geometric position. This greatly facilitates the incorporation of MMW range (z) profiles as a third coordinate to FLIR (x,y) has points.

At the end of Phase II, NVESD will be provided with a complete suite of FLIR and MMW radar algorithms that perform both stationary target detection and recognition/identification. This provides a needed but currently unfunded growth capability of the Army Target Acquisition ATD in support of the future main battle tank. These algorithms could also provide useful enhancements to the Comanche and Longbow target acquisition systems, as well as the Scout Sensor Suite ATD. These Target Acquisition/Target Recognition (TATR) algorithms will be thoroughly documented, as well as installed in source code on a turnkey SUN SPARC 10/51 workstation. Users Manuals will be provided for all software, along with training of NVESD personnel at Ft. Belvoir. The MMW radar algorithms are based upon the techniques of genetics, adaptive filtering, and geometric hashing applied to the coherent high resolution range profiles. The algorithms will be applicable to both 35 and 95 GHz radars. The FLIR algorithms are based upon the techniques of line/point extraction (using expert rule sets) and geometric hashing. The two sensor modalities are hashed separately but with those intermediate results fused together using weighted voting, recognition by components, and hybrid logic schemes. The Phase II activities will continue the processing of Yuma and Grayling data bases that were collected by NVESD. During Phase I, techniques were developed to organize, normalize, and register the data; such were demonstrated for eight views of five targets. In Phase II, those techniques will be applied to 18 views of 13 targets, as part of the algorithms training process. Additional target data representing hull defilade armor will be obtained for the development of stationary target identification algorithms.