Yeah I think hdcp and reprocessing would be most difficult. There are hdmi splitter devices like those used for coloured bias lighting that I think could be used….similarly I think the processing actually isn’t unsolvable, it’s not much different than object detection from a live camera stream. I agree re-encoding the stream would be too hardware intensive for anything “cheap” like a pi, hence the secondary device control alternative initially, but analyzing the stream should be possible.
I suspect we got off on the wrong foot when I called it a product instead of just a project idea, but I don’t actually disagree with that stance. Cheers.