Clutter-Reducing, Scalable Image Processing Methods for Target Acquisition and Tracking

摘要

A three year ARO project has resulted in important novel non-linear image processing techniques that aim to facilitate target detection and tracking from video sequences involving severe clutter and change of target scale. One of the non-linear techniques developed is the morphological locally monotonic (LOMO) filter. Morphological LOMO filter is scalable to target sizes and eases scale sensitive target detection. Another significant development is the speckle reducing an isotropic difflision (SRAD) method for eradicating clutter and noise from radar and ultrasound imagery. A third development is a generic connected filtering technique called the inclusion filter, which has been shown to improve tracking in clutter. Target shape and size as well as other set theoretic criteria can be constrained using the inclusion filter, without distorting target edges. Yet another significant advancement in target tracking has been made through developing projection model active contour models and related Monte Carlo methods. The projection model tracking methods have been shown to be resilient to severe clutter and change of target scales. The ARO project has supported 10 graduate students and 3 undergraduate students in the process of earning engineering degrees from University of Virginia. The project has also led to over 25 peer reviewed publications.