Point cloud generation from UAV images

May 31, 2019
Reaction score
Can one of you who works in the surveying and mapping field with your Phantom 4 (Adv/Pro) help me understand how any of the softwares and cloud services (I'm especially interested in MapsMadeEasy) generate a point cloud from a set of images, WITHOUT control points, and why we should regard them as accurate (either horizontally or vertically)? I know, for example that the photomosaic produced by MapsMadeEasy is offset by a couple of meters (as I would expect given the GPS uncertainty of the Phantom 4)---so what kind of accuracy should I expect with the point cloud data? I'm not trying to avoid using control points; I'm just astounded at the ability to generate them without having used any, and am trying to understand what has been done...THX in advance, Phil.
Mar 4, 2019
Reaction score
If I'm understanding your question correctly (andI admit I may not be) I think you're confusing two very separate things. A point cloud is a set of data points in space, nothing more. The SFM software takes the images and puts them together as they are, based on positionality to one another. Think of is as a plastic sheet being assembled in the air that makes one big picture out of lots of little ones. If there is no ground control, or no RTK or PPK to adjust the imagery to align with exact points on the ground it is still just a big plastic sheet in the air. It will use the drone's geo-tag in the xref header to approximate the location to the real world, and therefore your error will be based solely on the accuracy of the drone saying "I think I was here when the photo was taken". GPS on a moving drone coupled with the IMU gets you to within ~6 to 30 feet of the real world (usually) but it is subject to a multitude of inherent errors. Elevation is almost always the worst one, but your X and Y can be several feet off as well.
The point cloud gets created in SFM software AFTER any geo-location inputs and bundle adjustments are made. So to answer the question (the way I think you meant it) the point cloud without GCP or at least RTK or PPK is not going to be accurate. The model itself can be accurate as to the internal measurements point to point (although it probably won't be), but it needs the GCP (or alternative RTK/PPK) to ""nail it to the ground" in the right places to stretch or shrink it to the lumpy old Earth to fit correctly. Think of a tarp stretched over a lumpy dirt pile - GCP/RTK/PPK would be your nails holding it taught so it outlines the pile correctly. So no, without GCP or at least RTK/PPK all you have is a pretty picture and some points in space, but nothing you can say has "survey accuracy".

New Threads

Members online

Forum statistics

Latest member