Please find hereafter a presentation of my project aiming at monitoring the parameters of a swimming pool through a set of regulators and sensors.
The key point of this post is about collecting the information (chlore and pH) provided by the regulators through an electronic display.
The platform is a raspberry pi 5 running Home Assistant 2025.4.x
So collecting the value of chlore and pH and have them available in Home Assistant was my hot topic.
Knowing that I have no expertise on the field of OCR and AI, I found that many of the OCR solutions available require training the data and complex parametrization.
I turned to EasyOCR and made a quick test of a set of images : excellent hit rate. And very easy to use (python in my case).
The only problem to be addressed is the incompatibility between pytorch and alpine linux (largely explained on internet).
To workaround this, I considered creating an addon running in a separate docker container. This new container is based on Debian. So installation of python, pytorch and easyocr are successful.
For performance reason, it is important that the easyocr instance is not restarted with each and every image to be analyzed. So I created a small http service that will receive the image to be analyzed and that will return the discovered text. The http requests are made from Home Assistant via a curl command.
At this stage I’m struggling with the integration of the docker image into Home Assistant. I miss the visibility of my easyocr service as a custom addon into the Home Assistant sphere (stop/restart, logs access, config).
If any of you has experience in this specific field, please let me know.
I’m also available for deeper information.