Background and Target Randomization Root Mean Square (RMS) Background Matching Using a New Delta T Metric Definition.

摘要

EO/IR/Laser detection of a target amidst clutter/background is a difficult problem often treated with simplistic models. Unlike noise, clutter is more complex, neither spectrally white nor statistically Gaussian. Therefore, it is insufficient to lump clutter with noise and use standard detection curves. Using current target detection models, it is extremely difficult to perform effectiveness assessments of signature management technologies for survivability of military ground vehicles. Current models do not consider the vehicle on a component-level basis and do not account for artifacts introduced into images from aliasing and varying amounts of clutter. Algorithms must be developed that quantify the effects of random backgrounds on the imaging capability of electrooptical systems to improve false alarm rates. Current trends dictate that EO/IR/Laser imaging systems must consider developments in signature management technologies and countermeasures that are driving clutter magnitudes higher than target signature magnitudes. These trends make the problem of target detection in clutter especially critical. Battelle has produced image randomization software called BATRAN (Background and Target Randomization) which computes various types of statistical distributions to randomize background and target pixels separately. The types of statistics implemented include exponential, Gaussian, log-normal, and Rice distributions for both the background and target. To generate synthetic images to assess the detection performance of thermal imaging systems and countermeasured platform signatures, a method to characterize the background and target is required so that their signatures can be statistically matched. Current methods use an area weighted average temperature difference (AWAA7), which is regarded as inadequate in representing observer's sensitivity to the inherent detection cues of the target/background/clutter signatures.