Activity Feed › Discussion Forums › Strictly Surveying › Best Procedure for Surveying UAS Targets for High Accuracy
Best Procedure for Surveying UAS Targets for High Accuracy
spmpls replied 2 years, 1 month ago 16 Members · 64 Replies
I spoke with my local Trimble sales rep today. Although he’s not a PLS, he has a lot of surveying experience prior to becoming a sales rep. Not only will he loan me any equipment I need (SX10, R12s, dini level), but he also wants to help with the project. He is good friends with a highly respected PLS in the area, that I’m also acquainted with. I plan on using the ideas from this post and consult this other PLS and develop a good plan.
Since this is a master’s project and not a thesis, the work won’t be published. I am using this as an opportunity to learn more about the specific equipment I already own (WingtraOne and Pix4D). If this were a thesis or dissertation, I would probably need to research more variables, such as photogrammetry software and drone hardware. I’m going to eliminate those variables since I own one type of drone and photogrammetry software package. Aside from fulfilling the requirements for my degree, my other main goal is to become more knowledgeable about my hardware to benefit my business.
I already know that I can pretty easily get +-0.10-0.15′ accuracy. But if I can consistently, and demonstrably, get closer to 0.05′, that will be extremely useful.
A few general points to bear in mind.
GSD is not the same as accuracy.
I’m assuming that you are using the boards to check the “accuracy”, so you are acutally conducting an exercise in getting consistency on what are effectively identical surfaces (great for use as primary control for photography, but biased in terms of determining what accuracy can be general;ly obtained).
Just as a comment, my experience with a Sony on boards is that better results tend to be obtained using a cross rather than a chequer board and we tend to use boards of around 18″.
Surface texture is important: if you get a surface with an extremely regular pattern the software can get confused and end up with mismatches that throw in large distortions.
“Checking accuracies” – conventional survey tends to give slightly low results, as the point on the rod sinks in to soft surfaces: aerial work tends to give slightly high results (unless you spend $100,000 on LIDDAR). We find that a correction is nearly always needed on vegetation: nothing fancy, just a simple classification of the vegetation into main types in the survey area which are then corrected from the results of a few ground surveyed checks in each type.
From experience, a flying height of around 320 ft. works out at a good compromise between achievable accuracies and the time needed to fly an area (and to maintain visual contact of the aircraft).
If the boards are being used to check on the achieved accuracies then a few right out on the edge would be useful as that is where any distortions are most likely to occur. With that perimeter road you have a good chance of getting areas where there is little detail even at pixel level, so the software may get quite a bit of mismatch.
Final thought: if you look at something on the ground and can’t exactly see where it starts and ends (ie. irregular edging to something) don’t expect the software to do any better.
I have lots of thoughts based on experiences…But this is your project and you will learn a good bit. S Series total stations are much better at traverse than SX-10. Run Static and Traverse and Levels, adjust in TBC. I think the targets may end up being too small to really zoom in to the exact pixel. Once P4D spits out the final product test it, you will probably find that P4D will be the weak link. Good luck, good project.
I’m wondering if a much smaller experiment would let you learn a lot about techniques, before doing the big one to get the accuracy results that are the end goal.
.- Posted by: @chris-mills
my experience with a Sony on boards is that better results tend to be obtained using a cross rather than a chequer board
This is counterintuitive for me. Is it a software effect or would it be true of visual interpretation also?
. Visually I find it better to “see” the chequer board, but using both Pix4D and Photoscan results seem to favour the cross. I guess it is because the cross is defined equally by being surrounded by the background colour, whereas the chequer board is trying to define the point by intersection of the four squares. Fixels having a finite size might be something to do with it.
40 acres is a small experiment. The difficulty with a much smaller site is that there wouldn’t be much room for the software to make errors, so you wouldn’t learn a lot. (Think measuring a distance less than the length of a pocket tape – you should struggle to get it wrong. try and measure 50 yards and you have ample opportunity to generate errors due to the technique you might choose.) A Wingtra has ample range to cover 80 acres a flight. When flying a fixed wing you need a big enough area to get the turns in at the end of each line and settle down onto the new heading.
In the early days when we tested various options we tried 8 inch targets: they did work but not as well as the larger ones. With the Sony you still get many pixels on the target; in vegetation of any length they are just a job to find and easy to get shadows across them, which can cause problems (and they can look like discarded litter).
@chris-mills yes, and I have even found two footers hard to see sometimes at 400′ AGL.
Levels, and RTK, if you want to do this for academic purposes, leveling the targets and then surveying them with RTK for horizontal is the best way to do it. Vertical is always the control and it’s more difficult to acquire precisely than horizontals. Levels are the key. The R12 is more than capable of getting it tight horizontally but it’s restricted by the Geoid Model for vertical. If for some reason you wish very tight horizontals, do it static, with redundant occupations.
less than .05′ with drones,,,,,,,,, haven’t seen that.
if the boards need to stay down for a while, rather than just for the day, then the 18 inch boards stand up better to wind, etc. than 2 ft. ones. If there’s a lot of vegetation it’s also easier to get the smaller size board to sit flat. Same comment applies if the ground has quite a slope on it.
We are using the boards for main control, but when we have been working with a manufacturer on test surveys for a PPK aircraft we found that the boards do improve the overall results. More importantly, if anything goes adrift with GPS signals then the boards provide a fall back so the project can be recovered.
The gsd of the Wingtra / Sony Rx1RII at 400′ agl is 0.05ft. I’d like to see how close I can get to that. This will be an interesting exercise for sure.
I’m only using the targets as independent check points. All of the processing will be performed with PPK processed geotags only (x, y, z, omega, phi, kappa). That is part of the study, to see what parameters affect accuracy, and ultimately how accurate is PPK-only for this particular UAS.
Guess that drone figure is a “mathematical” achievement, rather than a practical one. To check it on the ground every reading taken would have to be to a bipod or tripod – nobody does that.
Just for general interest, when we do tests on our aircraft after any repairs we will often include a check survey on the photography. Results from a couple of weeks ago, based on around 120 ground check points (gps on pole, three x 5 second readings at each location (locations selected at random whilst walking site, so the comparison with the air survey is onto the ground model and not an indicated and marked point):
Surface, mean cm, standard dev. cm.
stubble, -1.00, 5.68
track, -3.25, 3.28
new crop, 0.11, 1.83
rushes, -6.00, 5.29
bog grass, -5.00, 3.87
In P4D, if you process Step 1 first, then it makes identifying the targets in Ray Cloud extremely easy. With PPK, the cross-hair is within a pixel or two before manually identifying the center. So with a 12″ target, it is extremely easy.
Here is a target with the GCP location estimated by P4D. I have not manually selected the GCP center location yet. This is typical results with the Wingtra.
Certainly the accuracy of the point cloud varies greatly with terrain type. That is not what I’m trying to test in this project. I’m mainly concerned with the best case scenario: no vegetation, highly textured surface. Primarily land development, where the vegetation has been stripped and they want a pre-construction topo. Then progress and final topo. In all of these scenarios, the terrain is only bare-earth. There is a good market for this type of work.
Yes,for that type of work we expect 1-2cm. accuracy in height – plan accuracy for that type of work isn’t so iimportant. Of course, since you will be putting down a model with huge numbers of points the overall accuracy will be far higher than a traditiional one done on foot -especially since the rod point would sink in to any soft areas (or the rod-man might not want to walk there!).
I’ve done extensive testing of flown topos.
In fact to stake out many of our reclamation sites the procedure would be to stake the work limits which were placed on actual generated points from the photogrammetry company, get a correction then lower or raise the control points and stake the site.
Usually there was a “miss” high, normally about .5′ we would raise the control .5′ and proceed. These were optical photogrammetry sites with the flights covering large mapping areas.
Now that LIDAR is available our checks are more random and often quite good. One project was a 200 sq mile area that was going to have a railroad loop placed in it. We checked actual points near the loop and they fit randomly +/- .2′. Everyone was quite happy and I was pleasantly surprised.
For small drone flights with lots of ground control .15′ seems to be about what it can do. Maybe a tad better.
I figure surveyors always check photo control, just part of the job.
Using RTK for check points, I always get 0.15′ or better (usually <0.10′) with 0 ground control with the Wingtra. It’s quite impressive. This is why I want a more rigorously surveyed test grid to see how good it really is.
Wingra published a white paper claiming 1cm absolute horizontal accuracy with no GCPs. But they surveyed the check points using RTK, so I’m not sure about that.
That is impressive and not what I’ve seen, the ground control targets we’ve set for drone surveys are always very extensive. Time consuming and dense. Mostly I could have driven a 4-wheeler around the site in less time than flying it and controlling it.
Log in to reply.