Just wanted to share my experience of vision on InMoov. Some time ago I subscribed to a kickstarter with a camera called JeVois. They received their funding and sent me one of their cameras. I think that this camera is a game changer for vision on InMoov.
From their website :-
JeVois = video sensor + quad-core CPU + USB video + serial port, all in a tiny, self-contained package (28 cc or 1.7 cubic inches, 17 grams or 0.6 oz). Insert a microSD card loaded with the provided open-source computer vision algorithms (including OpenCV 3.4 and many others), connect to your desktop, laptop, and/or Arduino, and give your projects the sense of sight immediately.

I’m running a deep neural network using TensorFlow which can recognise up to 1000 everyday objects. You can of course train it for your own requirements. When video output is connected it is showing predictions of around 25ms. This is ALL within the camera. No external net connection is required. I’ve just shown my InMoov at makerfaireUK.com this weekend and even the hard core techies found it difficult to believe that there was no net connection. I was showing different pics of different breeds of dogs and it could tell the breed from the pic. These were random pics from the net, not preselected. It could even, fairly reliably, tell the difference between an Indian elephant and an African elephant.

It can be used with no video output (so no PC required) and serial data is output from the on board hardware serial port to an Arduino, RPi or whatever.

I haven’t begun to explain what this device is capable of, I suggest you go to jevois.org and see for yourself. And best of all, it’s £57 from Amazon.