Image Recognition Software to run code

Sorry for the vague title

​

I don't usually do these sorts of projects but a client wants me to come up with a concept device/software that reads traffic signs and reads out what it means. The way of doing this I thought would be to use a camera that would recognise the roadsign (using a database of images I already have) and then read out the description from a table.

I've used Unity/Vuforia for an AR application, it recognises each image from a database and then displays the AR content so I was wondering if anyone know or had ideas as to how I could have a local database that would then read out the text rather than the AR image if you get what I mean

​

Cheers