I am new to the forum. You guessed it, my wife got a phto radar ticket here in Arizona, on Interstate 10 between Phoenix and Tucson, near Eloy. It was a classic speed trap where the posted 70 MPH speed is reduced to 55 MPH at a more or less permanent consctruction zone. Se was clocked at 68 as she was coasting to 55.
I am the curious sort so decided to look into the van mounted systems by R'flex.
Luckily the plate photo is online for download from DPS and it reveals some interesting clues about some potential countermeasures. The back of the car was in shade so the flash was the dominant lighting.
The plate image is very low resolution -- only about 88 pixels wide by 35 tall. Reproduced plate characters are only 2 to 3 pixels wide, proably about 3,000 pixels total. So much for megapixel performance. I figure when the photo was snapped the car was about 150-200 ft from the camera lens, off axis by 20 to 30 degrees while going 68 MPH. This is a **very ** good system -- amazing they can do that and get such a clear image. Note that the plate frame, as well as the car is white, but actually appears dark on the image despite the full blast strobe flash. Major signal processing dinking with contrast is being done here.
The whole photo is digitally compressed with only about 7 levels from full white to full black with the range of noise of about three levels. This leaves little more than 4 intensity levels of actual data so, on the surface, it should not be too hard to fool the contrast computation. Luckily R'flex has filed patents that explain a little of the technology. The trick is that the background of the plate is highly retro-reflective so when the flash hits, the foreground numbers are backlit in high relief. After the image(s) are acquired the software scans the whole rear end image for the shape of the plate based on the rectangular edges marked by the reflective background. When the plate area is "found," the software then scales the intensities found in plate area to 4 bits - to 15 levels from full white to full black (or so claimed in the patent - I can only detect 7) then the same scaling is applied to the balance of the photo -- this is why the main photo is so overexposed in some areas and underexposed in others. The patent is not too specific on the scaling algorithm but might be something as simple as the average pixel level across the plate image or detecting weighted maximum/minimum values and assigns values in between. Also, according to the patent, they may actually do two complete simultaneous images at different exposures and then fill in one from the other to get the best contrast. Looking at the photo, the plate background is so bright the software obviously spends most of its effort in attenuating the intensity to bring out the dark plate characters -- resulting in the rest of the white car, in direct view of the flash, going dark. In short, the software's job is to dig the blacks out of the very bright background and "don't care" too much about other gradations.
What is interesting is the plate printing with the words "ARIZONA" and "THE GRAND CANYON STATE" and the left side Saguaro image, even though about the same color and darkness, have almost completely dropped into the noise. These characters are not that much smaller in area but they have the retro-reflector material on top of the printing so, despite being a dark color, to the camera they look much the same as the rest of the background: VERY interesting -- the system has the highest contrast with the retro-reflective coating on the plate "backllighting" the plate characters when the flash fires. It is the ABSENCE of the retro reflective coating on the plate characters that makes the plate characters black and highly visible, not the darkness or color of the characters themselves.
However, the mountain profile printed on the lower left of the plate is still visible at reduced intensity probably because the area is much greater in size. So clearly size and darkness underneath the retro coating also play a role, although it might have a less effect in setting the contrast.
Apparently the after market retro reflective spray-on material both diffuses the retro-reflection across the plate and adds reflective properties to the otherwise non-reflective plate characters. When the pixels are all rescaled to "normal" levels, the plate numbers are **supposed* to fall into the three levels of noise as overexposed. According to some "independent tests," it works with some cameras but not others.
But after looking at the plate photo full sized, I'm a bit skeptical that it works with R'flex. From what I have seen in photos the spray-on stuff still creates an enormous reflection when hit by the flash. Since the plate characters are non reflective behind the coating, a background pattern based on darkness alone is likely still detectable. The software will likely find the darkest gray and try to force it to black, pulling in more shades of gray. There is a real human looking at the noise distribution and my guess from looking at the photo the noise will not be randomly distributed: -- there will likely be contiguous darker pixels which follow the characters making them detectable. Still, it might be hard to distinguish between some numbers and characters which will require the operator to "guess" -- is it a 5 or a 6? Might work some of the time.
But there are clues as to how the spray system might be improved with a little more work. I plan on doing some tests::
1) Apply a thin coat of clear reflective paint with known properties such as Krylon REFLECT-A-LITE with a brush over the individual plate characters so, with a little trial and error, their reflective properties will appear as reflective as the rest of the plate background. This essentially "hides" the plate characters by making them equally reflective compared to the rest of the plate.
2) Then use a thin coat of clear diffuse paint such as Krylon Crystal Cclear over the entire plate -- it should go on a little milky, but not noticable to the eye, I think. This should cause the background to become diffuse, reducing the reflective contrast making the reflective intensity more constant across the surface.
3) Then use 3/16 inch masking tape and tape grid pattern over the plate. Then spray a final light coat of REFLECT-A-LITE over the grid to create a reflective "noise" pattern, about the size of each pixel. This hopefully breaks up any remaining reflective patterns.
4) Then construct a custom made, big-wide, flat black license plate frame/holder.
5) Then, around the plate frame/holder outside edge apply thin strips (about the width of the typical pixel: 3/16") of white, highly reflective automotive safety tape, keeping the overall size in approximately proportion to the plate itself.
So with this configuration: the detected image data will appear at the edge of the applied reflective tape instead of the edge of the plate itself: the tape is small enough in area not to affect the overall reflective intensity much, but a pixel wide enough to trick the software into taking data. The overall plate dimensions will therefore appear bigger so the real plate image will get proportionally smaller -- less data for each pixel. Then, once the software "finds" the plate defined by the edge of the tape, the software will be forced to include part of the the black plate frame/holder area as data, the equivalent of plate characters for the purpose of computing the correct contrast.
So in the end the software with its top priority to render black black, will be forced to scale the black border area to its normalized level even at the expense of overexposure and burning elsewhere. The hope is that system will intentionally scale the residual plate characters into the noise level, rendering them unreadable.
Sounds like fun.