Cheap ANPR

[Scuzi]

I’m posting this in GD because I’m not sure where else it fits.

Background:
Asking for a friend. He has a private lane which is used frequently by cars travelling at up to 20, maybe 25mph. He would like to keep an automated log of the vehicles using the lane, and also have the ability to get push alerts on his phone if certain specified VRMs access it. Images are not required as he has full CCTV coverage already. He just needs the text log accessible over a network share and the push alerts.

I know there are COTS ANPR systems and cameras available but they are expensive and indiscreet. I was wondering if there was the possibility of a Raspberry Pi based solution. I have found some R Pi project logs but they have been for low and slow traffic areas, e.g. personal driveways so I don’t know if they’d be up to the task.

What say you OcUK?

 

[Scuzi]

Speed humps if it's speed that's the issue?

Nah that’s not the problem. He has people regularly accessing his property without permission and he wants to know about it ASAP even when he isn’t there.

 

[Scuzi]

Something like this? https://magpi.raspberrypi.org/articles/anpr-car-spy-raspberry-pi

I saw that but, admittedly not being up to date with a lot of this stuff, I wondered if it wouldn’t be potent enough to deal with, at peak, say 30 vehicles per hour?

 

[Scuzi]

I seem to recall there are some open source number plate recognition systems.

Edit how about this?

https://github.com/openalpr/openalpr

That looks like it could be the software answer. In terms of hardware, what are the general thoughts on what sort of beef is needed to run this?

 

[Scuzi]

From what the article says, I don't see why not? For the cost of a rPi it's not an expensive roll of the dice either.

Good point. For the sake of, what £50(?), plus a camera (I’m guessing a generic IP camera?) it might be worth a shot just to experiment.

 

[Scuzi]

We have a private road nearby, the owner has installed a gate across it and opens it at peak times and (presumably) pays someone to collect a toll.

That would be ideal but a bit overkill for the situation. In an ideal world, there would be an automated gate which operates on a VRM whitelist, triggered by ANPR. But that’s iteration 10+. The idea right now is to do this for under £200!

 

[Scuzi]

Camera (24GBP) https://thepihut.com/collections/raspberry-pi-camera/products/raspberry-pi-camera-module
Pi Zero (12.80GBP) https://thepihut.com/products/raspberry-pi-zero-wh-with-pre-soldered-header

That’s the sort of budget we are working to!
I imagine it will need weather proofing and whatnot, but that aside, considering the layout of the are and the feasible places the kit can be placed, it might need some telephoto capability. Come to think about it, IR night vision would also be required. But I can see from the offerings on that site that either is a possibility. Something to ponder.

Thanks!

 

[Scuzi]

If he's keen to keep costs that low (sub £200) then why not just a gate with a (combination) padlock or lock of some sort for the people who are supposed to be travelling down the lane? I know it's extra faff but...

Also just a basic automated gate and issuing the people with remotes for the gate would presumably be less faff than using ANPR... I don't see why ANPR would be needed if a gate is involved and CCTV is already present as the gate negates many visitors... all you'd have left are perhaps the odd tailgate incident before it closes etc..

All good points but where’s the fun in that?

A locked gate isn’t feasible as the faff involved with stopping and getting out to open and close is not tenable. The users of the lane change frequently and administering the dishing out of keys, codes, or fobs would be a pain.

He just wants to know who is using his lane and when. That’s a higher priority than access control right now.

 

[Scuzi]

Nvidia jetson or older TX1 would be better than a Pi. IIRC it can analyse 6 HD streams at once.

I have just found this project:
https://medium.com/@alannewcomer/license-plate-recognition-with-a-jetson-nano-e94c6ff683bc

Definitely seems more up to the task. I’ll explore further.

All of the projects I have seen so far have been for use cases involving static cars. I wonder how a clean image of a moving car at a distance of 10m could be sourced for analysis. I think the only way to find out is to prototype it. Unless anyone else has ideas?

 

[Scuzi]

Update for anyone interested or considering something similar.

This project was put on ice for a while but a trespass and access rights dispute reignited it. When I took a look again, there is a very user friendly and competent solution which was borne from the OpenALPR project, Rekor Scout, although it does require subscription of 8 USD per month.

We have installed a relatively cheap spare HD IP camera covering the lane in question. It is Lorex brand, manufactured by Dahua I think. Going for around £70 used on eBay. Nothing fancy but gives a 4k resolution picture and is IP66 rated. It runs via some external grade CAT5e to a PoE switch in the house. Rekor Scout is currently running on his W10 daily computer and is doing an impressive job of picking up pretty much every plate going by, even on the faster traffic, and in sub-optimal lighting conditions. Accuracy for plates is >95% so far. Even more impressive is that if it doesn't catch a plate, it can detect the profile of the vehicle and give a make and model. This is working with >80% accuracy.

Night performance needs some work but I think this can be improved with a separate IR source closer to the vehicles so the reliance on the camera's IR source is reduced.

The next step is to host the software on a much lower power unit. It does seem to be quite CPU intensive at times, due in most part I think to the sensitive motion detection driving lots of unnecessary image analysis. I'm going to experiment with different settings to see if it can be optimised. Alternatively, a camera with a longer lens would give a narrower field of view which could be better targeted; this will be considered if software optimisation isn't effective.

Overall project cost so far:
Camera - £73
Cable - ~£10
Software subscription - $96 per year.
6 pints - £28.80
TOTAL (1 year) - £195.80

Potential costs:
Camera with a longer focal length - £100-150.
IR light source - £25.
Power hookup for IR light - £20.
Separate machine to host it - ~£200?

 

[Scuzi]

Obviously there's power at the lane because you're running the ANPR equipment so why couldn't you have just installed a powered barrier with fob/tag entry?

Similar to the ones you see at coastal caravan parks

No mains power at the camera mounting point. It runs on PoE.

If you can find a powered barrier with ANPR controlled whitelist for under £200 I’m all ears!

Does this guy own a chain of chicken restuarants?

It’s nothing sexy like that, just stubborn people being stubborn!