The title reads a little disingenuous. Aren’t they using deep learning and Google Street View to predict the make, model, and year of a vehicle? Then, separately, they’re using predicted vehicle information to guess the demographic traits of the person who lives there.

Still, applying image recognition on Google Street View is a pretty cool idea. The location information gives you a whole other set of data to join with. For example, you could recognize flags hanging on people’s houses and join that with immigration data (there are a lot of Italian flags where I live), or recognize species of trees and join that with climate data.