Terrain based navigation is certainly nothing new(1950s?), so there is probably a large pool of research into it, but I wonder what's been done with recent pattern recognition advances?
Terrain based navigation is certainly nothing new(1950s?), so there is probably a large pool of research into it, but I wonder what's been done with recent pattern recognition advances?
In areas with unique features (dense urban) localizing to within ~50ft has been demonstrated pretty robustly but more accurate methods of localization are a lot more variable and depend on the datasets, cameras (monocular vs stereo), compute, etc.
Aside from the fact that the amount of data required for that would be orders of magnitudes larger than you can reasonably store, that wouldn't work at all in areas with little terrain features like deserts or the open sea.
https://en.wikipedia.org/wiki/TERCOM
Here's a modern take on it: https://www.youtube.com/watch?v=U5Kr0YI3sec