You could also decouple image capture and packet processing, so that if the former starts to run ahead, you can buffer until the latter catches up. NICs do this a lot, for much the same benefit.

I'm not great with Python, but your code seems quite legible, and it doesn't look like there's any buffering or backpressure in place right now - which is perfectly reasonable for a fantastic proof of concept like this! Just a thought for when you next get around to hacking on it.