Multiresolution, Multi-Scale Target Identification and Tracking using the Anisotropic Diffusion Pyramid

摘要

A three year ARO Young Investigator project has resulted in a novel target identification and tracking system based on multiresolution, multi-scale image processing methods. Through this research effort, two nonlinear image processing methods have been developed and utilized in target tracking: the anisotropic diffusion pyramid and the morphological pyramid. Coarse-to-fine target searches are implemented within the image pyramids, providing a lOOX improvement in computational expense over standard correlation-based approaches. Several improvements in tracking robustness have been achieved, including a lOX improvement in target localization over standard linear processing methods. A noise-resilient morphological diffusion method has been designed, and diffusion techniques that converge rapidly to locally monotonic signals have been developed, yielding a significant advance in nonlinear diffusion-based image processing. Further improvements to the diffusion technique have included the incorporation of multigrid methods and the generalization to multispectral imagery for multisensor tracking applications. These advances in theory have been validated with a new tracking simulator, in which several visible- wavelength and infrared image sequences have been used for testing. During the last three years, the ARO-funded project has supported 10 graduate students at Oklahoma State University and has led to over 20 refereed publications.