I use the drone for surveying areas of land that we're going to be using for the construction of "Integrated Constructed Wetlands" for wastewater treatment. So, I take whatever drone (previously a Mavic Pro, now a Phantom 4 RTK+D-RTK2), plan out a mission using Pix4D Mapper, then it flies an autonomous mission in a grid pattern. Each photo has georeference data relating to it, which also includes camera angle and other bits of info. Then I put them all into Pix4D and put it through 3 phases of processing (initialisation, Point Cloud generation, DSM rendering). From the Pix4D processing, I get a fully 3D model of the site, contour lines (for use by our CAD technician for design drawings) and a few other bits. The Mavic Pro was great for quickly surveying sites where I just needed to get a rough (+/- a few metres) layout of a site for design and planning, but the RTK unit can get me down to ~3-5cm accuracy in all 3 axis. There's a little bit of throw, since I have to convert everything to bloody Irish Transverse Mercator instead of GPS positioning, but it's not too bad. The main issue is elevation. GPS will log it at X, but ITM is usually ~56m below that value. The software does an awesome job of taking each of the photos, relative to its position in the grid that it flew, cross-referencing each image against others and finding keypoints in each photo. From this, the algorithms will use these key points to generate a 3D frame, then flesh it out with a point cloud to fill out the rest, then create a texture, DSM/DTM (digital terrain map - terrain without vegetation) and contours at whatever interval I want. What I surveyed in an hour of flight-time would have taken someone 1-2 days to do by hand, easily. The only shortfall I have, is that the software for the D-RTK2 base station does not allow for it to act as a mobile rover yet. Thankfully, for this particular site, I had an old survey of a yard that is part of the survey area, from which I got accurate and meaningful elevation data for 5 of my 7 control points (markers laid out on the ground by me), which in turn gave me a correction value to adjust the other 2 with confidence. All in all, I think I managed to get this down to 3cm accuracy, and this is only for machine work in a field, so that's an order of magnitude more than I needed.