Improving Feature-Level Data Assocation for Multi-Mode Sensors
Small Business Information
6 New England Executive Park, Burlington, MA, 01803
AbstractWe propose a methodological advance for performance optimization of feature-level data asociation in tracking and ATR algorithms, thereby improving MOPs such as probability of correct associations (Pca), probability of correct identification (Pid), andtrack purity. Our approach augments GMTI kinematic reports with high-value features extracted from multi-mode and/or multi-sensor data sets. High-efficiency selection methods are used to discover the feature subset maximizing the expection of Pca over anensemble of operating conditions and measurements noise; a process that ensures performance robustness. The selection process addresses issues of feature independence at the outset. The joint use of the optimal features represents a stong form oflow-level fusion, thereby maximizing information content of each datum. The rigorous physics-based formulation enables well-posed feature estimation. Although our sensor focus is on radar modes (GMTI, HRR, SAR, ISAR), the generality of the methodsupports extension to additional data types (e.g., IR, LADAR) and multiple time-frame optimization. Super-resolution techniques are viewed as a key method for improving feature quality. The proposed work provides a new paradigm for jointly optimizingtracking, recognition, and association performance, and derives an efficient and robust numerical implementation. Improving the reliability of data association will benefit applications that require confident object identification and/or tracking.Examples include surveillance of ground stationary and moving targets, missile and decoy discrimination, civilian or military vehicle tracking in cluttered environments via video monitoring, commercial distributed traffic monitoring for Intelligent HighwaySystems, law enforcement applications such as border surveillance, facial recognition, and fingerprint recognition. Medical applications include aiding diagnosis from imagery from dissimilar sensors such as X-ray, untrasound, magnetic resonance, PET scansand CAT scans.
* information listed above is at the time of submission.