Hyperspectral-Augmented Target Tracking
Abstract
With the global war on terrorism, the nature of military warfare has changed significantly. The United States Air Force is at the forefront of research and development in the field of intelligence, surveillance, and reconnaissance that provides American forces on the ground and in the air with the capability to seek, monitor, and destroy mobile terrorist targets in hostile territory. One such capability recognizes and persistently tracks multiple moving vehicles in complex, highly ambiguous urban environments. The thesis investigates the feasibility of augmenting a multiple-target tracking system with hyperspectral imagery. The research effort evaluates hyperspectral data classification using fuzzy c-means and the self-organizing map clustering algorithms for remote identification of moving vehicles. Results demonstrate a resounding 29.33% gain in performance from the baseline kinematic-only tracking to the hyperspectral-augmented tracking. Through a novel methodology, the hyperspectral observations are integrated in the MTT paradigm. Furthermore, several novel ideas are developed and implemented-spectral gating of hyperspectral observations, a cost function for hyperspectral observation-to-track association, and a self-organizing map filtering method. It appears that relatively little work in the target tracking and hyperspectral image classification literature exists that addresses these areas. Finally, two hyperspectral sensor modes are evaluated-Pushbroom and Region-of-Interest. Both modes are based on realistic technologies, and investigating their performance is the goal of performance-driven sensing. Performance comparison of the two modes can drive future design of hyperspectral sensors.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 2008
- Accession Number
- ADA484377
Entities
People
- Neil A. Soliman
Organizations
- Air Force Institute of Technology