I used to work for a company that did GIS mapping of guard rails, lines, and road signs. Part of what we always ended up doing was massaging the data we received from the Lidar mapping to line things up to exact locations for clients. If this map is accurate, I salute whoever spent the time fixing the details.
I don't have much experience with lidar, but I do with machine learning based approaches with photographic data. There has been a massive explosion in the last 5-10 years in filtering and processing algorithm advances that can more readily clean up false positives and insignificant noise. I would assume the barriers to high quality data are quite a bit lower now.
