Hi. I am confused about which altitude to fly at when mapping tall buildings, say 20m high. If I set for 80% overlaps, and an altitude of 40m, I am presumably only getting the 80% overlap on the ground. What overlap am I getting at roof level (20m above ground)? Hence, what GSD would I get at roof level relative to the ground GSD?
This is a common issue that many people never think about. Instead of thinking in terms of overlaps think about how many images per pixel you are getting. A really good map on average is going to have 12-15 images per pixel which happens to correlate to about 65-80% overlaps. For example, 75/75 should get you a subject in the frame about 4 times X and 4 times Y which would be 16 images per subject (pixel), but very rarely does everything actually get captured 16 times and much of it that is at the far sides of the frame is all but garbage because of the worse focus and distortion.
So by cutting your GSD in half you cut your images per pixel in half and if we need at least 6-8 images per pixel for a confident reconstruction then you are cutting close to that edge. This is why buildings always require additional imagery to stitch successfully. Running an additional nadir flight over just the building will help ensure that the ortho view is accurate. Otherwise the roof may not look centered over the floor or it just may not stitch correctly at all. Here is an example of the same mission. The left was nadir images for the ground only. The right is with additional nadirs to make sure the roof received the same coverage.
This can also be accomplished with oblique imagery, but I would not recommend that unless you are not so concerned with terrain accuracy and just want a pretty model. It is also not recommended to use GCP’s with oblique imagery. When the structure is the primary focus of the oblique image then the data in the background of the terrain can be pretty low quality and will affect it adversely. When we want the most accurate ground combined with the most accurate structure model we run two separate processings and then clip and merge the point clouds.
Thanks Michael. That’s pretty comprehensive. My projects are normally roofing focussed and I usually have the Roof Report produced also, to get CAD plans & measures.
Most buildings I fly are relatively low-rise (4-10m high) and I plan flights at 80% overlap, and flying at 25-30m above the building. I think I have been getting reasonable results more by luck than planning (I often do obliques also).
My new project is 20m high. Are you recommending two nadir flights, maybe criss-crossing the paths? What settings would you suggest for overlap and altitude above ground? All assistance gratefully received.
If you’re really only concerned with the roof then don’t worry too much about the ground data. We can fly the equivalent of about 80m without too much degradation. I think keeping it about 20-30m from the roof with one flight is fine until the structure gets to about 50m high. If you do larger commercial roofs then I would not get any closer than 30m if you want to run one flight. The more tiles you create across a larger area will increase the probability of warping and the overall dimensional accuracy will suffer unless you use control points. If the building has a seamed roof like above then you definitely don’t want to get too close.
I haven’t experienced any improvement in results above 80/65 overlaps. The only noticeable difference has been an increase in processing time and storage use.
Thanks again Michael. That’s all very useful.
Overlap on tall buildings is only partially intuitive. I built an app for dronedeploy that will calculate it for you, taking into account altitude.
Here is an illustration of what Michael is describing:
And here is the math to work this out by yourself if desired!
Photogrammetry at its best. Thank you @Jamespipe!
Thanks @Jamespipe. I need to study this further, but I get the principle now.