New Physics-Based Methodology for Optimizing Tracking and ATR Performance via Feature-Level Fusion of Multi-Sensor Data

We propose a methodology for improving perfomrance of ATR and tracking algorithms. We discover and develop high-value features and templates extracted from multi-sensor and multi-look data, and enhance their utility using model-based and learing-based feature level fusion (and hypothesis level fusion). This optimization is performed in a statistical framework for robustness, and is cast in an integrated platform that links ATR and tracking algorithms with target physics databases, data processing and feature extraction capabilties, and noise and scenario generators. Genetic algorithms are used to drive the optimal feature/template selection and fusion engine design process. Feature/template and fusion utility is assessed by measuring tracker and ATR MOPs such as track purity, and probability of correct assocation and id over a suitable distribution of operating conditions, scenarios and noise processes. The data modalities considered include radar, laser and infrared. Super resolution methods are applied to develop improved feature types from image and profile data, and to enhance the quality of their construction. The methodology can be used to improve performance MOPS across diverse scenarios, reduce computational complexity, provide natural objective functions for sensor resource management, and inform design of Advanced Technology Demonstration (ATD) experiments.