Sensor Data Fusion for Target Classification and Identification

Award Information
Agency:
Department of Defense
Branch
Army
Amount:
$120,000.00
Award Year:
2001
Program:
SBIR
Phase:
Phase I
Contract:
DAAH01-01-C-R110
Agency Tracking Number:
A002-1399
Solicitation Year:
n/a
Solicitation Topic Code:
n/a
Solicitation Number:
n/a
Small Business Information
SCIENTIFIC SYSTEMS CO., INC.
500 West Cummings Park, Woburn, MA, 01801
Hubzone Owned:
N
Socially and Economically Disadvantaged:
N
Woman Owned:
N
Duns:
859244204
Principal Investigator:
Adel El-Fallah
Research Engineer
(781) 933-5355
adel@ssci.com
Business Contact:
Raman Mehra
President
(781) 933-5355
rkm@ssci.com
Research Institution:
n/a
Abstract
The effectiveness of Non-Cooperative Target Recognition(NCTR) against air targets is limited by a lack of robustness in thefollowing respects: insufficient-fidelity signature simulation;statistically uncharacterizable signature variations; excessivelylargepose-model libraries; excessively large target-type libraries;misclassification of ``novel'' targets; insufficient target-identityresolvability using single sensors; and difficulties in fusing diversesources/sensors. Many of these difficulties arisefrom the fact thatconventional OPTIMAL techniques (e.g. Bayesian filtering andestimation) expect PERFECT models and, consequently, can behavevery non-optimally if the mismatch between model and reality is too great.Consequently, optimal techniques cannotbe applied as a ``cookbook''panacea---they must be augmented by ROBUSTNESS STRATEGIES thatcompensate for model-mismatch and other problems. Scientific SystemsCompany, Inc. (SSCI) and its subcontractor Lockheed Martin of Eagan MN(LM-E) propose theapplication of new ROBUST DATA FUSION ANDROBUST-BAYES FILTERING techniques to NCTR problems. In our approach we (1)use generalized likelihood functions to model the uncertainties as well asthe certainties in data; (2) fuse very disparate kinds of datausinggeneralized joint likelihood functions; (3) deal with ``novel'' targets byintroducing and modeling an ``unknown target'' type; (4) reduce the size ofthe pose-search library via fusion of kinematic data with target ID data toestimate pose; (5) ensureaccurate, stable NCTR by using a true JOINTstate-estimator; and (6) ensure computational efficiency and robustness byusing an approximate filtering scheme with efficiency O(n) or O(nlog n)and theoretically guaranteed divergence properties.Targetidentification is one of the key technologies for globalsurveillance, precision strike, air superiority and defense which arethree of the seven science and technology thrust areas identified bythe Director of Defense Research and Engineering.Commercialapplications of advanced tracking and identification systems exist inseveral areas such as: radar, biometric identification, industrialinspection, medical screening and diagnosis, failure detection andidentification, and remote sensing.

* information listed above is at the time of submission.

Agency Micro-sites

US Flag An Official Website of the United States Government