I'm looking at a theft that occurred in my town of just over 7000 people the other day. The perps pulled right up in front of an HD camera, but because the camera is probably 15 feet up, I'm having the worst trouble getting the license plate numbers.
Unfortunately, our local LEOs aren't much help; I've got make, model, and range of years on this vehicle, but they apparently can't be bothered looking up what I have been able to extract from the license plate. I'm fairly certain I've got the first 3 characters, but the next 3 are proving elusive.
I'm starting to think I'm dealing with the luckiest idiots on the planet; the motion sensor where they broke in failed to detect them, and they never went into the other part of the building, where there was more valuable equipment, but active motion sensors.
I'd be leaning toward this being an inside job, but with only 4 employees, I'd call it unlikely. Only the one camera captures any view of the license plate, and I'm pulling my hair out, using Amped Five and the various tools that are supposed to deblur license plates. But it's almost like the IR on the camera overcompensates once they turn off their headlights, even. The plate becomes a total whitewash.
Anyone got suggestions/advice on this? I picked the wrong day to give up barbiturates. (Obligatory Airplane! reference)
Naturally, I don't want to just post my progress so far, but if anyone's interested in seeing how far I've gotten, I could message you an image or two.
Sometimes the data is simply not there. (
However (it costs nothing) try the tool I suggested here
https://www.forensicfocus.com/Forums/viewtopic/p=6586815/#6586815
From the little I have seen of these it needs a lot of patience and a lot of attempts to (hopefully) get something readable.
jaclaz
Any chance the vehicle stopped when the plate came in view for frame averaging.
Yes, the vehicle came to a stop right in front of one of the security cameras. I think I've managed to pull the first 3 letters, but the next in the sequence is 3 numbers, and what I thought started with a 1 apparently does not.
I'm currently fishing for this one; found a web site that looks up plate numbers and gives make, model, and year of vehicles. Since our local PD doesn't want to be bothered looking up possible numbers, I'm running through this site to see what hits close to what I can make out.
I have exported the entire video's frames into .bmp images, but I'm not terribly experienced with image manipulation. Of all the things I've done, this one has me learning a lot the hard way.
I don't mind a challenge, but this is a huge puzzle with pieces missing, and for a newbie to this area, it's a killer.
Only the one camera captures any view of the license plate, and I'm pulling my hair out, using Amped Five and the various tools that are supposed to deblur license plates.
Don't understand that. Not do I understand what you mean by 'deblur' – motion deblur or some other kind? Basically, you need to have some idea of digital image processing, and some ideas of the theory involved.
idea 1 can you get single frames off the video that don't show any motion blur?
If so, you may be able to superimpose them, and average out, beating down any random noise you may have. (Later Already suggested by another poster, I see)
Idea 2 What kind of frame image would you get? bilevel? greyscale? some kind of colour? Is the plate histogram 'nice' with well separated black & white, or has it been all squeezed together? I'd look for some kind of equalization operation, if possible focussed on the plate.
But it's almost like the IR on the camera overcompensates once they turn off their headlights, even. The plate becomes a total whitewash.
That may mean that all information gets pushed out of the image. But it could mean that it just gets moved to a higher pixel range, closer to flat white, and is not clearly visible. 'Stretching' those pixels (say, with values 230-255) over the entire dynamic range (0-255) may help. Again, this is equalization.
Idea 3 Again, if you have 'something' there, but it cannot be quite 'seen', you may try edge detection on the plate to see if you can tease out any structure. (This depends a lot on how big the plate is in the final image.) Perhaps it's named 'Laplace transform' or 'Sobel filter', or some such. (This is one area where some knowledge of image processing theory can be useful, if just for those names … )
(You may also find it easier if you look at the relevant part of the image in false colours, rather than in plain colour or greyscale.)
The difficult part is to convince anyone that the data you may find actually was present in the image, and not something you added on your own, by all these strange operations.
It looks like we may have our thieves. For some reason, the shop that was robbed just decided to check and see if the vehicle involved was one they had worked on. It appears that it was.
That means they already had the plate number, the VIN number, and the customer's name, address, etc. They just had to look for it.