|
The output signals from FL1R sensors suffer from many degradation mechanisms including high clutter noise, detection noise and strong dependence on weather and time of aquisition. The degradation mechanisms are more pronounced for ground terrain scenes, which are the cases we are considering.
An automatic target detection system, must overcome these degradations. Degradations due to the target's own thermal signature distortions (scale, rotations, aspect angles, partially hidden parts and screening by smoke or dust) have to be tackled too. Thus, "automatic FLIR target detection" involves a massive computational burden. Here we introduce a system based on a highly parallel compact hybrid thermal imagery signal processor in which the main computational route is executed in an analog optical processor module. The input video signal, after electronical on-line preprocessing is introduced via a spatial light modulator to a coherent optical processor. The image is optically prefiltered using an appropriate synthetic filter, and then Fourier transformed- using a lens, in a 4f configuration. A transformation replicator is utilized in order to obtain several spatially separated replicas of the transform. Each one of the Fourier transforms is spatially multiplied by a different optically implemented algorithm and then optically inverse transformed. The output signal is collected by an array detector and is transfered to a digital post processor. Some of the optically implemented algorithms are used for background rejection. The other algorithms are used for the detection of regions of activity and targets by their thermal signatures. The detection algorithms were designed in order to achieve high detection rates with target scale invariance. Preliminary examinations of the overall processor have yielded high positive detection rates, and medium false alarm rates.
|
