The current issue of MagPi covers the topic of running AI on a Raspberry Pi.
Unfortunately taking a huge topic like AI and Machine Learning and squashing it into a Pi stretches many boundaries to the limits. Compounding this are the constraints of the magazine format and keeping things interesting for a general audience.
Image recognition is probably a subject which has general use for many small Pi projects, be they a paint-ball sentry , capturing smiles or train recognition.
Installing TensorFlow (the AI engine) is best covered with this link.
However if you want to get down to the Google Foo of how to actually set up image recognition, then you should probably start with this article, and follow with this how-to blog. Even if you are not going to play with a Pi, these articles will give you enough background to successfully bluff your way on at least a small but important subset of AI. Although they assume a moderate understanding of Linux you can probably skim through the Python bits by Googling things like Python pip, git, and curl – they are just ways of installing or downloading code, the other thing to google is ‘convolution’. It will be a bit easier if you have a spare Pi on which you can set up TensorFlow then just copy and paste the commands from the ‘how-to’ – understanding comes later!
Given the strain this places on a small Pi, I think I’d use a Pi 3B+ fitted with heat sinks as a minimum hardware setup.
.
