Long ago, it was hard to do computer vision -- like PhD hard. Then, about ten years ago, it started to get easier with the OpenCV libraries, which did most of the hard work for you. But even that needed you to program and build a toolchain and it really only ran well on beefy PCs. But about two years ago, a series of small, cheap computer vision modules came out with camera and processor combined, making it possible to do advanced computer vision at the same cost and format of an Arduino.
That include Jevois and PixyCam, but my favorite of them is OpenMV, which combines built-in MicroPython (super easy to program) with a powerful and easy to use IDE and a number of plug-in boards that allowe it to control everything from motors and servos to screens and WiFi, with no additional computer needed. As an example of how easy it is to use, over at our sister site, DIY Robocars, we show how to put together a full computer vision racing car using OpenMV for less than $100.
Now OpenMV is getting even better, with a new processor with twice the speed and memory, along with interchangeable camera modules that allow you to use a global shutter sensor (best for moving robots like drones and cars) and thermal vision, with the FLIR Lepton module. And now it's cheaper, too, starting at $49!
You can back it on Kickstarter here. Unlike many other Kickstarter projects, which are new products from new teams, this is a mature product from a team that has made and shipped thousands of the previous version and the Kickstarter is just to allow them to purchase components at volume to keep the price down. I've been using an early production version of the OpenMV H7 for a few months already, and I can confirm that it works great.