Date of Award

Summer 2013

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Electrical Engineering - (Ph.D.)

Department

Electrical and Computer Engineering

First Advisor

Edip Niver

Second Advisor

Dalton Rosario

Third Advisor

Venkataraman S. Swaminathan

Fourth Advisor

Alexander Haimovich

Fifth Advisor

Ali Abdi

Abstract

Polarimetric infrared imagery has emerged over the past few decades as a candidate technology to detect manmade objects by taking advantage of the fact that smooth materials emit strong polarized electromagnetic waves, which can be remotely sensed by a specialized camera using a rotating polarizer in front of the focal plate array in order to generate the so-called Stokes parameters: S0, S1, S2, and DoLP. Current research in this area has shown the ability of using such variations of these parameters to detect smooth manmade structures in low contrast contrast scenarios.

This dissertation proposes and evaluates novel anomaly detection methods for long-wave infrared polarimetric imagery exploitation suited for surveillance applications requiring automatic target detection capability. The targets considered are manmade structures in natural clutter backgrounds under unknown illumination and atmospheric effects. A method based on mathematical morphology is proposed with the intent to enhance the polarimetric Stokes features of manmade structures found in the scene while minimizing its effects on natural clutter. The method suggests that morphology-based algorithms are capable of enhancing the contrast between manmade objects and natural clutter backgrounds, thus, improving the probability of correct detection of manmade objects in the scene. The second method departs from common practices in the polarimetric research community (i.e., using the Stokes vector parameters as input to algorithms) by using instead the raw polarization component imagery (e.g., 0°, 45°, 90°, and 135°) and employing multivariate mathematical statistics to distinguish the two classes of objects. This dissertation unequivocally shows that algorithms based on this new direction significantly outperform the prior art (algorithms based on Stokes parameters and their variants). To support this claim, this dissertation offers an exhaustive data analysis and quantitative comparative study, among the various competing algorithms, using long-wave infrared polarimetric imagery collected outdoor, over several days, under varying weather conditions, geometry of illumination, and diurnal cycles.

Share

COinS