Now You Can Be Big Brother Too, With A Raspberry Pi License Plate Reader

If you are wowed by some of the abilities of a Tesla but can’t quite afford one, perhaps you can enhance your current ride with a few upgrades. This was what [Robert Lucian Chiriac] did with his Land Rover, to gain some insight into automotive machine vision he fitted it with a Raspberry Pi and camera with an automatic number plate recognition system.

This bracket should find a use in a few projects.
This bracket should find a use in a few projects.

His exceptionally comprehensive write-up takes us through the entire process, from creating a rather useful set of 3D-printed brackets for a Pi and camera through deciding the combination of artificial intelligence software components required, to making the eventual decision to offload part of the processing to a cloud service through a 4G mobile phone link. In this he used Cortex, a system designed for easy deployment of machine learning models, which he is very impressed with.

The result is a camera in his car that identifies and reads the plates on the vehicles around it. Which in a way has something of the Big Brother about it, but in another way points to a future in which ever more accessible AI applications self-contained without a cloud service become possible that aren’t quite so sinister.  It’s an inevitable progression whose privacy questions may go beyond a Hackaday piece, but it’s also a fascinating area of our remit that should be available at our level.

You can see the system in action in the video below the break, as well as find the code in his GitHub repository.

60 thoughts on “Now You Can Be Big Brother Too, With A Raspberry Pi License Plate Reader

  1. Look up “ALPR”….nothing new….this dude’s home made device is more like OCR.. since there is no access to the dmv database…therefore no “privacy” issues….heck, anywhere in public – there is no privacy… even if there was, the mindless sheeple don’t seem to mind “big brother” one iota …. considering the masses willingly allow constant 24 x 7 monitoring by their corporate masters (alexa for example).. once upon a time i would get paid big $$$ to “sweep” residences and businesses for listening devices…. now, it’s people complaining that their listening device isn’t sensitive enough !! … smh… world turned upside down.. I weep for the future of humanity !

    1. I plan to test today Mobile LPR app for Android.

      “Monitor any activity with simple database searches that reveal the full scan history with location of any recognized vehicle that drove past a camera on your phone.

      Export scanned data easily and integrate it with other systems. Automatically copy recognized license plate text to the clipboard and paste it surely everywhere.

      I need full image of the scanned car to be saved for postprocessing to let me verify failed plate numbers against
      make, model, type, color of the car, since if LPR recognition fails, you get wrong LP number entered into your LPR database.

      Ok, I can process images of cars, recognized LPNs in off-line mode, manually, to verify matching, but not sure what file formats, database structure is supported.

      Let me know your opinion, since some street speed cams have LRN software built in and generate ticket on-the-fly

      1. Absolutely not. When you are in your *private* home, you behave differently than in public. Sharing opinions in a comment section, and having everything you say in a private setting being monitored is a big difference.

        People trying to say privacy doesn’t exist anymore, and “I have nothing to hide”, will affect everyone, long-term. That’s the issue.

        1. That’s the thing, though. If you’re using some off-the-shelf computer system such as an x86 or ARM computer, you’re being monitored.

          You’re complaining about privacy while also leaving your windows open and paying a monthly fee to have people come and look through your windows each day. A smartspeaker or can only listen to you. A PC or smartphone logs everything you do and sends it to untold numbers of servers for surveillance and marketing purposes.

          1. Except if you run more than one PC, and the PC that is connected gets a censored set of material that the others have full access to, for example, photos and video.

    2. The privacy in public issue depends on the country you’re in. Even in some European countries, you can get in trouble for filming on the street and using or publishing the video – go read up on “Technoviking” – he successfully sued the guy who filmed him and won in German court.

      1. Would be curious how that would work if the video in question is being fed to a machine to spit out numerical data like this instead of being viewed as imagery by human eyes. Is it still considered filming if the camera is just part of a machine that produces no likenesses as an end result?

      2. Considering that it’s a small step from this project, to another that pairs the license plates with a heatmap showing the frequency of sighting of a particular plate… there are some real privacy concerns with this type of project.

      3. The filmmaker wasn’t sued until he started printing off T-shirts, licensing the video, etc years after the video first hit the internet. He made tens of thousands of euros off another person’s image and then *shocked pikachu face* he got sued by the person whose image he was using.

        Even in the US, commercial use of a person’s image is protected. You can take someone’s picture, but you cannot use it for commercial purposes (the press, covering a newsworthy event, are excepted.)

    3. LPR systems don’t need to be linked to a DMV database directly. A local text file containing the plates of interest (stolen vehicle, stolen plate, missing person affiliated to plate, wanted person affiliated to plate, etc.) is all that’s needed. When I was last managing an LPR system, that file was updated every six hours. Operators can also manually input their own plates if needed.

      Also the system we used had two cameras per pod — the other being a close-up IR camera which the OCR was done on. The other camera was just used to take a picture of the vehicle.

      1. There is so much info avilable fo free in many Euopean countries that link the number plate to the car.
        Some simple web scraping and you can verfiy the make/model, the colour, tax, mot/fitness test, insurance validity!!

        The more advanced commerical LPR cams do a lot of this themselves. Make/model/colour mathcing without databases and of course if the occupant is male or female, wearing glasses, wearing their seatbelt.
        The coronavirus outbreak in China has highlighted the ability of cam analytics to detect if facemasks are being worn either walking in the street or in the car too.

      1. Actually a trivial observation, but somehow a revelation if you apply it to people.

        People focus so much on confidence, probably because it’s easier to assess than accuracy.

        It should turn into a meme, with a high confidence level above a person’s head, and a cartoon showing a massive wrong judgement.

    1. Don’t forget that confidence stays at >98% while the same numberplate is decoded in multiple different way.
      My best guess is that the confidence level only reflects a guess of finding a numberplate, and not of the content of the numberplate.

    2. The dark truth of a lot of these ML recognition algos. They have abysmal error rates, but it doesn’t even matter—they are used as a on-demand and inscrutable pretense to funnel people into human-operated security, legal, and carceral systems. It’s no different than the polygraph test which is bunk and pseudoscientific, basically no better than doing a taro card reading of your suspect—or using drug dogs who just respond to their handler’s cues. It’s an updated technological smokescreen to manufacture probable cause or even spurious evidence in a trial and pass it off as legitimate. They don’t care if it’s inaccurate, as long as they can make that inaccuracy lean their way with some cajoling so they can seize some person or property. I mean look at the areas which have banned public facial recognition. They know this perfectly well.

    3. The “LP” tag and confidence applied to it means it is >99% it has identified a license plate.
      There is no confidence score shown on the contents of the characters.
      Which I would think would be very useful on a per-character basis, but not in whole.

      In my experience with a hit-and-run, it is far more useful to the police knowing which characters you are sure about.
      “AB_ 1_34” will narrow the options down to a single-digit number of valid plates, when then combined with a make/model and color of car limits it down to one match.
      But “ABE 1S34” is more problematic. The “S” doesn’t fit the pattern so is easy enough to make “unknown”, but that “E” is going to case problems finding matches.

        1. Uber is for boomers. Millenials can’t afford to pay someone to drive them around.

          I drive a cheap car and contribute to a 401k unlike the entitled boomers who need everything done for them and take 30% of my income for their Social Security to fund their cruises, gambling, narcotics, and healthcare.

          If boomers had some work ethic and education, the world wouldn’t be such a mess.

          Turn off Fox & Friends and read a book. Maybe get a GED.

  2. This, but add a rating system so you can quickly flag bad drivers. Make a public database of this. Then when you detect a number, look for it in the database and issue an audio warning so you know bad drivers are around :) Could be a cool way to get some actual use out of it.

          1. They used to be super easy to spot, back when everything had steel wheels. It would be a mid-large sedan, with the shittiest cheap hub covers on it, trying to make it blend in, less like a cookie-cutter fleet vehicle.

      1. I dunno, I mean devil’s advocate but people are actually spectacularly good drivers. We always weight bad experiences heavier in our perceptions and memories, but our ability to put down millions of miles between fatalities in such a chaotic and crowded and energetic system deserves a little credit. Even the bad drivers have kind of an impressive record to be honest. Humans can somehow make a maniacal system such as this work. Like self-driving cars have a whole hell of a lot to prove before their bromides about being inevitably safer than real drivers are to be believed.

        1. Humans use automobiles to kill tens of thousands of humans every year. Somehow the fact that they do it with automobiles makes it okay. We even have a “special” crime for murder using a motor vehicle, “vehicular homicide”, with a lesser penalty. You have been desensitized to murder by propaganda from the auto industry.

    1. I’ve been working on a personalized version of exactly this, based on an esp32-cam module that displays to my phone, originally as a backup camera. But since it had all this processing power, I thought adding a license plate recognition system to it, that I can signal “hey there’s a jerk” and it marks all the license plates it can see right then, and then lets me know when one of those license plates is visible again, would be really groovy. I’m coming to the conclusion that an esp32 doesn’t have anywhere near the horsepower to do this, but it’s been fun designing.

      1. You won’t get far with an “intelligent” system that assumes a 1:1 correspondence between cars and humans.

        With all those false positives you might as well just assume that everyone is out to get you.

    2. Please put me on you list of terrible drivers, and make sure you leave plenty of room for me to pass you. Or maybe you should just stay home, that way you can be sure I won’t crash into you.

    1. “Fighting dystopia with dystopia?”

      Could actually work for that, you find the gestapo’s motor pool, you record plates coming and going, you use that list of plates to avoid them.

  3. By the way…. I would recommend a different mounting system, rearview mirror mountings aren’t as strong as you think. They often don’t hold the weight of even just the mirror vibrating on that lever arm for the whole life of the car (Had several fall off on older cars and I never load them)

    1. Can confirm that on older Toyota’s they’re quite rattly.
      Granted the stock anemic 15watt speakers had been swapped out for 120 watt ones and a new head unit and amplifier, but still no subwoofer.

      1. There’s usually a plate with backangled edges that the mirror dovetails onto, these should have a screw tapped into them to clamp it tight to the plate on the screen. So you may be able to tighten it up, if it hasn’t lost the screw or stripped.

  4. Combine this technology with a public blockchain where every licence plate observation time/place gets logged by any and all participants and we would quickly end up with a database that tells us where almost all vehicles for the most part, “live”. I’m sure others with even more extensive databases could match all that up with personal identities and people’s various attributes. There’s a car coming down the street that the computer has been identified as a fox fancier so the the corner vendor quicly grabs a large cute fox stuffed animal and holds it out hoping the driver will notice and the rest is pure commerce. Spooky?, yes, and also scary. A similar lookup could try to find political contributions so ANTIFA thugs would know which windshields to hit with hammars. (or worse)

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.