I wouldn't be surprised if the ATR technique was comicly simple, like CS101 simple. BTW, that's not necessarily a critiscism of the implementation.
My guess would be the software, new with ATR, takes the difference, P-frame, of the PAX image and a reference image based on body type. This can be done with any one of dozens of redily available compression algorithms. Then it counts the number of macroblocks within a certain grid size, or even contiguous grids. Differences above a threshold are marked on the screen.
There's nothing remotely revolutionary about this, like is said, it's CS101. If they, L3 or one of their subs, did come up with a better techinque, awesome.
I would be suspicious if anyone tried to claim any sort pattern recognition, along the 'fuzzy/neural' line of research. Kinect looks like the first viable commercial application of this so it'll be decades before the government gets there.
In my observations, if the subject of the scan presents
anything to justify the alarm, they are cleared. I have never, ever seen anyone rescanned.
Originally Posted by
RadioGirl
I'm absolutely sure that it's reflection/scatter, not a through transmission. The attenuation by the human body at these frequencies is so great that a transmission system would only generate a silhouette where the beam missed the body completely, and blackout where the beam hit the body.
I believe they've said as much.