Add support for 7 segment displays

[johnnychicago]

Hello --

My city has given me a new water meter that has a 7 segment LCD display (and LORA communication that I cannot listen to, but that is another topic).

I was trying to get this going, and despite a reasonably ok camera picture, I do not get any recognition. ?debug says:

array(2) {
  ["TesseractOcrException"]=>
  string(185) "Error! The command did not produce any output.

Generated command:
"tesseract" - "/tmp/ocr8XrpfU" -c "tessedit_char_whitelist=0123456789"

Returned message:
Estimating resolution as 240"
  ["readDigits() : !is_numeric()"]=>
  string(51) "Could not interpret "". Using last known value 1212"
}

It occured to me that the OCR maybe is not trained at all to read 7 segment numbers? Could that be - or is my image just too bad and I should fix that?

[nohn]

Hi @johnnychicago I have no idea ;) If you could provide the image you are using + your config (the latter is optional), I can investigate, if that is possible.

However: If you your meter already has a 7 segment display, there is a 99.9% chance, you can grab that information digitally and precisely instead of relying on image recognition.

[johnnychicago]

Hey, thanks for looking at this, @nohn !

Here'a photo of the meter and a screenshot of the ?configure page, also the config.php. I am perfectly happy for you to tell me the photo is crap, and that's my problem. I just used a camera and light I had flying around to prototype, and got a bit sad that nothing returned.

As for the meter, it seems like city went to lengths to make this non-accessible. It has no connectors whatsoever, only transmits via Lora, and the transmission is encrypted (I agree with that, unencrypted over-the-air transmission seems not appropriate). So I have to read the meter optically or not at all.

Previous meter was mechanical, but had a connector that would close once per liter consumed, that was nice.

config.php.txt

[nohn]

It seems like the OCR tool I am using (Tesseract) is not able to properly do OCR on 7 segment display style text. It even fails for optimized test data like this:

Others seem to have found a way using OpenCV preprocessing: https://github.com/DevashishPrasad/LCD-OCR, but that's a route I won't go for now. I'll nevertheless keep this ticket open.

[johnnychicago]

Hey @nohn thanks for analysing this.
I will not be able to integrate OpenCV into the container in any way.
I found the seven segments integration in home assistant, will give that one a try. But (apart from the fact it doesnt read my meter) I liked your setup a lot - very userfriendly.