When I think about Computer vision I think about when you have to verify you are a human now days. It normally asks you to identify something. It can be a bike, a car, a stop light, or a train. You must use your human ability to detect what is what.
Below is a good example my teacher came up with. A computer can learn to recognize things with our help. This is what I learned while doing the excercise in Fast.AI
Lets say you have a bunch of images of animals and you want a computer to be able to classify the images. When we say classify images we mean group them or cluster them, the images without a label. In order for a computer to do this the computer needs to to "learn" some general understanding of what an animal is lets think specifically about dogs and cats. We as humana able to interact with computers generally understand the differences between dogs and cats. Dogs are typically larger and have different shaped traits like faces and paws. A computer can learn these differences as well by looking through labeled examples of dogs and cats. If we create an image classification "model" or learning algorithm to classify images this model can start to understand the differences between dogs and cats by looking at say 50 images label cat and 50 images labaled dog. We can then show the model that we have just trained on a new unlabeled image and ask it to guess what kind of animal it is and apply a new label either a dog label or a cat label. The computer will then use the "learning" that it did with the labeled images to make a guess about what label most accurately fits the new image either a dog or a cat.
This is a very high level explanation of computer vision, but it is general enough that we understand how this might be valuable. In our NBA analytics project we might want to identify when a player is taking a shot. When a player is taking a shot we want to record this and analyze whether the player makes the shot or misses the host and we can do this with computer vision. It is up to use to "model" the computer vision problem because we don't yet have fully generalizable models that can tell us everything about an image at one point in time, but our modeling techniques are gettings better and we can segment images and understand a LOT about what is happening in an image. There are a lot of corporate research groups working on problems in this domains and they use these techniques in their products. I think that creating a few posts about company spotlights might be useful in this respect.