In the transportation sector there is lots of interest in using cameras to quantify origins, destinations and flows on different roads by recognising either vehicles by their size/shape or by reading their license plate and referencing a database.
Of course, there are issues: poor light conditions, partial image blockage etc.
I know Deepmind, other google projects and a whole branch of academia are working on this. What is the state of the art accuracy for either a. identifying vehicle type or b. identifying the license plate?
For example, Google say here[http://www.techworld.com/personal-tech/google-deepmind-what-is-it-how-it-works-should-you-be-scared-3615354/] their accuracy for static image recognition is ~95%. There is also tensorflow and other open source tools