You are here

Innovative Electro Optic Signature Exploitation for Recognition Advancements

Description:

OBJECTIVE: This topic seeks innovative methods for deriving a sparse set of physical target features that can be used for exploitation of air to ground signature data collected from electro-optic measurement systems including EO, IR and LADAR. DESCRIPTION: Current methods for exploiting EO signature data include statistical pattern recognition techniques and model based approaches. Model-based approaches use simulated data in combination with measurement data to define features for potential exploitation. Current approaches designing algorithms based on feature extraction do not guarantee robustness because the extracted features are not linked to reliably known causal physics. Second, limitations of current approaches also include the lack of linkage to a systems theory model for disciplined trade space design of algorithm features. Innovative methods for extracting salient physics based features that are associated with robust physical mechanisms are needed. Objects of interest for this topic include civilian vehicles including passenger vehicles and sport utility vehicles and dismounts. Although some exploitable features may change with background, this topics seeks methods that identify the dominant causal physics that both characterize the object and underlay the optimal inference solution. Associating the exploitable information content with the underlying causal physics based observables that each sensor can provide is a first step towards the development of new exploitation concepts and algorithms. Analysis methods are needed to efficiently produce the sparse cueing of salient physical features critical to the exploitation potential which exists within the full signature data. In addition, methods of modeling realistic sources of signature uncertainty within a systems theory prototype are required for exploitation concepts and algorithm development. New analytical approaches are specifically sought for identifying observable physical features that can be reliably exploited. Methods for discovering salient features that contribute to exploitation should also account for and model uncertainties in the measurement process such that they are robust to realistic sensor measurement effects. Realistic uncertainties in civilian vehicles and dismounts should also be modeled. Feature robustness and persistence should be assessed using metrics tied to systems theory such that methods proposed have a theoretical basis. Example system theories that have application to discrimination could include Sparse Bayesian Learning Theory and Information Theory. 1. Sufficiently accounting for, or eliminating uncertainty in sensor feature measurements. 2. Sufficiently accounting for uncertainty sources in the object of interest. 3. Reliance on a priori information. If the proposed analysis methods rely on a database that is developed offline, sources of measured or methods of simulating data should be specifically identified in the proposal. 4. Target Feature Exploitation. Enhanced methods of deriving object physical properties from sensed observables that account for realistic sensor limitations should be specifically addressed. PHASE I: Develop and conduct proof-of-concept demonstration of innovative physical feature based discrimination approaches using simulated data that model generic sensor characteristics. Characterize physical features with exploitation potential using metrics theoretically linked to system theory. Establish a mathematical foundation for cueing to underlying physical mechanisms responsible for inference. PHASE II: Mature algorithms and methods developed on Phase I results and demonstrate technology using controlled measurement data. PHASE III: Mature algorithms and methods developed on Phase II. Develop user friendly software tool and demonstrate technology using realistic sensor data. REFERENCES: 1. Keinosuke Fukanaga, Introduction to Statistical Pattern Recognition, 2nd Edition, Academic Press, 1990. 2. Richard O. Duda, Peter E. Hart and David G. Stork, Pattern Classification, Wiley 2001. 3. T. Cover and J. Thomas, Elements of Information Theory, New York, Wiley, 1991. 4. Tipping, Michael E."Sparse Bayesian Learning and the Relevance Vector Machine,"Journal of Machine Learning Research, June 2001, 211-244. 5. M. Bell,"Information Theory and Radar: Mutual Information and the Design and Analysis of Radar Waveforms and Systems", Ph.D. Dissertation, Department of Electrical Engineering.
US Flag An Official Website of the United States Government