New Techniques for Tracking Sequences of Digitized Images.
AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OHIO SCHOOL OF ENGINEERING
Pagination or Media Count:
A model for a generalized image tracking system is presented. Characteristics of minimum norm similarity detectors are investigated. A first-order local tangent plane model for digitized imagery is used to successfully predict properties of the auto and cross distance functions for real data. A matrix signal-to-noise ratio is shown to be the natural signal-to-noise ratio for the minimum norm detection problem, and an approximation is derived and experimentally verified for an upper bound on the probability that a minimum norm detector makes a particular error. A non-linear two-dimensional filter is presented which shows a significant reduction in noise variance in low contrast regions of an image. An optimum weighted norm is derived which minimizes the probability of making a registration error, and an adaptive reference set selection algorithm is presented which maximizes the tracking signal-to-noise ratio. The adapitve reference set selection algorithm uses the histogram of gradient magnitudes and includes a new gradient estimatorclassifier with a fixed probability of error. An adaptive Kalman filter is developed to update the reference image and the filter is shown to be stable in all areas of interest. Author
- Optical Detection and Detectors