Welcome Matrice Pilots!
Join our free DJI Matrice community today!
Sign up

Advice on Large Area Topo Map - GCPs, Overlap, ASL Elevation, etc.

Joined
Dec 15, 2017
Messages
8
Reaction score
0
Age
56
I'm looking to do a topographic survey of a 700 acre site where the vertical measurement accuracy needs to be within 12 inches. So dividing that by 3 means I can have a HSD of 4in/pixel or better (it will be better due to altitude limits.) So I have a few questions.

There seems to be no information anywhere regarding what determines the number of ground control points you need. I HAVE found conclusions that it's the law of FAST diminishing returns. Most example use 3-10 ground control points but those have been for site sizes in the 20-300 acre range. The site I am looking at doing is about 40% flat and 60% has hills with lots of elevation change up to about 500ft. It's a desert-like landscape so there are sparse buildings and vegetation, roads, but not forest or water.

The other question is, given the objectives, what's a good minimum frontlap and sidelap to shoot for? I was thinking 75/65.

And finally, I have a processing question. The elevation measurement of my GCPs is going to be more accurate than the drone's. I'm planning on processing with Photoscan. Wouldn't you need to basically launch from one of the GCPs and make a note of the ASL elevation of that GCP then modify the EXIF elevation data for all the photos so that the drone elevation has a more accurate ASL elevation that matches up with the GCP instead of using AGL, or an uncorrected ASL from the EXIF? Or would you use the option in Photoscan to just ignore the coordinates in the EXIF data and only use the GCPs?
 
First off let me apologize for what is going to be a lengthy post...you are correct in that your GSD will be well below 4". It will vary depending on what platform you're flying and what mp sensor your camera is...if you are flying an inspire 1 with x3, 400' AGL will put you closer to 2"/pixel. One problem you may encounter is maintaining a constant distance AGL with the varying elevations you describe. Some flight planning apps don't allow for terrain following and instead fly the path at the specified height from the take off point which can cause varying resolution and also could cause a crash if you take off at a lower elevation and the terrain is higher than the specified distance AGL. I use UGCS which allows you to plan flights that follow the terrain and keep your AGL and GSD consistent.

For overlap settings 60 side/80 forward is the minimum overlap required for 3d models as stated in photoscan documentation. I would suggest at least 70/80(I typically do 80/85 to achieve survey grade accuracy under 2") to get the best results. Keep in mind that overlap settings determine your accuracy and your processing times...the higher the overlap the higher the accuracy BUT the number of images increases drastically which means so does the processing time. For example if you use the minimum 60/80 you would end up with about 1100 images and if you did the higher end of 80/85 you would end up with more than triple that. This is where the machine you have to process on really matters since data sets this large take a significant time to process.

GCP placement varies and depends on the type of terrain, I think some people don't like to comment as it really is different for every scenario. If you're dealing with a flat landscape with minimal vertical features than 10 may be adequate but I've found that on large sites like this I usually put down around 20 and sometimes wish I would have done more after processing. Its important to have a decent grid throughout the project and around the outer limits, also if there are large vertical differences(like a cliff or pit) than placing a few GCPs along the ridge line on the lowest and highest points will make a huge difference. Remember that having some extras is never bad as they can be used as check points if not used in the solution.

As for the exif elevation data the only thing that photoscan is going to use the drone gps data for is for the photo alignment process which does not need to be more accurate, once you load the GCPs locations and optimize the project based on the GCPs the exif data becomes irrelevant.

One more thing to consider is that covering this much ground will take a considerable amount of flight time. If your flying at 400' AGL at a speed of 20mph and minimum overlap you'll be in the air around 1.5 hours which means you need at least 5-7 flights and as many batteries if you intend to fly the whole site in one outing. You'd want to fly it in the middle of the day so lighting and shadows are relatively consistant (say from 11am to 1pm) and if you have to fly on multiple days plan to go out at the same time of day.

Any other questions feel free to reach out, there's plenty more to discuss especially on the processing side of things.
 
Last edited:
For overlap settings 60 side/80 forward is the minimum overlap required for 3d models as stated in photoscan documentation. I would suggest at least 70/80(I typically do 80/85 to achieve survey grade accuracy under 2") to get the best results. Keep in mind that overlap settings determine your accuracy and your processing times...the higher the overlap the higher the accuracy BUT the number of images increases drastically which means so does the processing time. For example if you use the minimum 60/80 you would end up with about 1100 images and if you did the higher end of 80/85 you would end up with more than triple that. This is where the machine you have to process on really matters since data sets this large take a significant time to process.

This is a good write up summarizing much of what I've already gathered however in general I'm finding the amount of information (or number of people) that have "figured it out" is pretty small for these types of large land surveying applications. Recommendations in many cases are all over the map and very few people have done well controlled projects testing varying flight parameters and then doing accurate verification.

With that said, I would expect that increasing the overlap follows the general law of "diminishing returns" I've seen common claims that the relative measurement accuracy in the horizontal plan will be within 2-3 times the GSD. So if you need accuracy within 3 inches then you probably want 1in/pixel GSD. Vertical error is a bit more.

I'm wondering if you (or anyone else) has verified how the precision is actually affected by increasing the overlap to various different overlap settings? Example, 80/70 increased to 90/80 and the precisions went to max error of 3 inches to ?? inches?
 
For anyone interested in the topic of GCPs and mapping, you may be interested in this informative case study titled "Do more GCPs equal more accurate drone maps?" that is published by Pix4D and based on testing performed by the Nevada DOT in 2018.

Key takeaways:
1) Error in the vertical direction is most sensitive to the number of GCPs
2) 4 GCPs (and additional check points for evaluation of error) were deemed adequate in their testing scenario to achieve an acceptable accuracy

Here's a link to the article: https://www.pix4d.com/blog/GCP-accuracy-drone-maps
 

Members online

No members online now.

Forum statistics

Threads
2,727
Messages
25,387
Members
5,600
Latest member
RomanChrz