Rob has created a custom component for Home Assistant that allows us to use image classification via Tagbox, on a camera feed that Home Assistant has access to.
Following the instructions on github was pretty straight forward, the only thing I changed was to reference an external image_procesing.yaml rather than having it all in my configuration.yaml as I’m trying to be a bit cleaner.
In this case, my docker is running on Windows, the Tagbox instructions were fine for this, I didn’t realise you can just run the image and if its not installed, it will download it 🙂
As you can see, the component creates a new entity called image_processing.tagbox_name, with a state of the most likely item in the image, and then some attributes of the next most likely, and a count of any specific tags from the config.
I’ll play around with it to see what I can use it for more practically, I can think of all sorts of things this component will be useful for! From identifying if a car spot is free or taken, to seeing if the bins were already taken out!
I just wish I could specify a region within a camera feed for it to analyse.