Thanks guys
tomorrow morning i plan to purchase ESP32-CAM on amazon
Hi! Could you please share what LED strip you used and if you do not mind, short description or link to a tutorial on how to set it up? Is the LED strip connected to the ESP32cam directly and powered from it? Or is it completely separate and turned on via an automation? I would not want to over do it but at the same time would like to improve the light conditions and the digit recognition on a digital meter.
you can try our Meter Reading Engine which can effectively recognize all kinds of meter, such as water, electricity, gas, heat, pressure, flow, temperature, in various complex conditions.https://www.youtube.com/watch?v=ep8DTBt0XC4&t=22s
That video skips over the final slides. Is the software free? Do you have it integrated with Home Assistant?
I do not have a 3D printer for the case. So how can I fix this onto my water meter? Model Zenner.
If you don’t have a printer, I would look for a suitable plastic printer at the hardware store. If necessary, maybe just get a lid and cut it to size or shape it with a hot air dryer.
I wanted to adjust the camera focus, but daaaamn that’s indeed the most difficult part! I just destroyed my first one, LOL. Luckily I bought some extra. Anyway, you should NOT use regular pliers and apply force, you’ll just destroy the whole thing.
My lesson so far has been:
- Use magnification lenses so you can see up close
- Locate exactly where the glue is, don’t cut anywhere there’s no glue, it will only make it more difficult to rotate.
I’m now printing the ‘tools’ to adjust the lens and will try to remove the glue with some alcohol first, instead of cutting first.
EDIT
Ok success on 2nd cam 
What I did this time is a applied a few times alcohol to the glue using a cotton swab, to try to dissolve most of it. Then I did NOT cut it this time at all, I held the square part between my fingers and actually DID what I told everybody not to do above: use pliers again LOL. But this time just more carefully not to destroy (squeeze) the cam. I read somehwere that you should hear the remaining glue crack and it did! After that I can now freely adjust the cam, yeeeah
I also realized I used the knife method wrong, LOL, I was cutting in a horizontal plane, that’s not correct, you should cut almost vertically and use a VERY sharp knife. But again, personally I would not use that method at all.
EDIT2 LOL also success on the 3rd cam lens (I bought a few to mess around with) and actually getting good at this now It’s actually not hard at all. I also 3d printed the ‘tools’ (as per the documentation) but they’re pretty much useless. If you just apply alcohol a few times to the glue, then take the square part between your fingers, use magnifiying lenses if possible (I just love these for projects like this: Amazon.com: COYLAPY Magnifying Glass with Light Hands Free - 1X to 14X Magnifying Glasses for Close Work, Adjustable Jewelers Loupe Headset with 1x, 1.5X, 2X, 2.5X, 3.5X Lens for Hobby, Jewelry, Crafts : Arts, Crafts & Sewing) and use pliers to turn the cam counter clockwise, using moderate strength, it should actually move pretty easily. Just be careful not to squeeze TOO hard, it shouldn’t be even necessary if you use alcohol.
Of course make sure not to spil alcohol on the lens itself.
EDIT3
Well the next problem is that the wifi antenne that’s onboard of the cam module is pretty damn weak. I also did buy an external antenne but to switch it to the external antenna, you need to do some soldering that’s actually quite hard since the resistor you need to solder is extremely small…
So currently I have a pretty bad wifi connection and adjusting the cam focus is a lot of trial and error to get it right which is especially frustrating over a bad wifi link…
All in all my first impression with this project isn’t too good, it’s all quite frustrating.
EDIT4 actually soldering the bridge for the external antenna was super easy. I followed the soldering gun method here: https://www.youtube.com/watch?v=aBTZuvg5sM8 but used normal soldering temp, not sure why everybody is complaining there, was literally less than 1 minute work.
My WIFI is finally good now. But the next problem arises and that’s my water meter is receiving sunlight and this messes up everyhthing. Should probably use a cone like structure like they’re proposing but my issue is that the water company needs to take measurements too, not sure if they’ll like that thing on the meter…
EDIT5 I was about to give up when I realized I had to move the cam a lot closer! It’s now at probably about 6-7 cm above the meter and I’m getting a good image finally… Cam distance is super important here, need to get it as close as possible… Also it seems good practice to get the rotation right from the start, so that the system doesn’t need to digitally rotate the image, which distorts it slightly. To get an idea, look at the first few posts here: Images misaligned · Issue #533 · jomjol/AI-on-the-edge-device · GitHub
EDIT6 Finally getting my first good reading now, yiiihaa, finally getting somewhere now
EDIT7 after trying different models everything SEEMS to be perfect now, but need to try it a bit more …
EDIT8 Couldn’t get it ‘perfect’ in that it sometimes misread a number, it would mistake a 3 for a 5 every now and then for example. I think to get this really good you need to do a lot of optimization, like getting really the optimal distance between sensor and cam, optimal focus etc. It’s a LOT of tweaking and even then, I’m not sure how ‘perfect’ this could end up…
I think 3d printing the casing, suited to your specific situation, is really a must.
So an update, I 3d printed a casing, got the camera about 4-5 cm above the meter, as suggested as ‘best practice’ by the developer and everything seems to finally run pretty well. It’s not 100% perfect, in my case it sometimes mistakes the last 3 for a 5, but it’s easy detectable and thus not a big issue. Other than that currently it seems to work quite well.
So if anyone wants to try, my recommendations:
- A 3d printed holder is really a must. Make sure the cam is 4-5 cm from the meter
- Adjusting the focus of the cam is really important. Use the alcohol method to loosen up the cam to be able to rotate it
- You’ll need to do some trial and error regarding the models to see what model gets you the best result.
All in all this is quite a bit time consuming to get it all right, but after you get the hang of it, this is actually quite a nice solution.
And another update: since the honeywell V200 meter that I had didnt really seemed to be supported very well, I’ve trained the model myself with 3 days of data. It’s now working near perfect. I’m sampling every minute and on a daily basis I now only see 1-2 misreadings and these get automatically filtered. So I’m really happy now with this solution!