Welcome, Commercial Drone Pilots!
Join our growing community today!
Sign up

How to achieve sub-centimeter accuracy

dang113616

New Member
Joined
Feb 22, 2022
Messages
1
Reaction score
0
Age
25
I'm looking for a drone that can measure volumes with .1 - .3 centimeter accuracy for cut and fill operations. I was looking at the Phantom 4 RTK drone but it seems the vertical accuracy is about 2 cm. Are there any drone that can achieve the accuracy I need, or is it simply not possible?
 
A company I contract to has a Phantom RTK, and they aren't getting that kind of accuracy. I was told their vertical accuracy was greater than 2cm. Have you checked with DJI ?
 
I'm looking for a drone that can measure volumes with .1 - .3 centimeter accuracy for cut and fill operations. I was looking at the Phantom 4 RTK drone but it seems the vertical accuracy is about 2 cm. Are there any drone that can achieve the accuracy I need, or is it simply not possible?
Not likely. A total station would have a hard time repeating those values and a survey-grade GNSS receiver is more like 1.5-2cm. If you can hit 2cm with an RTK drone you are doing great. It also depends if you are talking RMSE's or stakeout/checkpoint values. We have found that an acceptable control point error is about 4-5cm or 0.15ft. We can hit 2cm on something small but the bigger the project gets the more the error trickles up. Most of our sites are right around 100ac with the biggest ones being between 200-300.
 
  • Like
Reactions: Dave Pitman
I'm looking for a drone that can measure volumes with .1 - .3 centimeter accuracy for cut and fill operations. I was looking at the Phantom 4 RTK drone but it seems the vertical accuracy is about 2 cm. Are there any drone that can achieve the accuracy I need, or is it simply not possible?
Not happening. What are they building, the mirror for the next space telescope, lol ?

I don't think that kind of accuracy is achievable at scale except maybe with a fine-tuned LIDAR workflow.

Are you sure you have the decimal in the correct place? Or maybe this is a tiny area?
 
Are you sure you have the decimal in the correct place? Or maybe this is a tiny area?
That's a good point!

I don't think that kind of accuracy is achievable at scale except maybe with a fine-tuned LIDAR workflow.
Maybe a terrestrial scanner...

4mm is the tolerance we had for the kerb (yes it was spelled that way - German Engineers) and that was a PITA with a 1-second gun with a mm-laser attachment. I guess when you're going 200mph 3 inches off the deck it better be smooth.
 
Welcome to the forum mate. I have never heard of anyone getting a 2cm vertical accuracy with an RTK drone. If anyone has I’d like to hear about it.
Regards
 
Welcome to the forum mate. I have never heard of anyone getting a 2cm vertical accuracy with an RTK drone. If anyone has I’d like to hear about it.
Regards
Sounds to good to be true right? Exactly why we will never buy a Phantom 4 Pro. Not because they can't do a good job but for the fact that DJI took an RMSE value and fooled everyone that doesn't understand the context with technical marketing lingo... and they never explained it. You know and I both know that when we hear accuracy of the product is 2cm to a Surveyor (and most people) that means that I can take a device out into the field, hit an elevation and it would be within 2cm of the actual point. Reality is that it is an averaged best case scenario especially when using GCP's because of warping in between them because of the GCP's localized effect. In reality 4-6cm is the acceptable range. Anything better than that just means that everything went right. Another thing to think about with photogrammetry that accuracy depends on the material as well. We typically see 5cm at best in rougher terrain whereas I have hit reports within 0.05ft on parking lots. So they market to a blind masses (no surprise) numbers that practically mean nothing to a person who understands actual +/- error when they take their survey rover into the field and stakeout the drone surface.

At the end of the day we are getting 1-2cm but that is an RMSE value which is a data calculation and not ground. One last note on that is when you use a physical marker in the field that point is obviously going to get more recognition in processing so checkpoints will always be more accurate than the mean.

While we are on the topic something that has been coming up a lot lately is the RTK drones and how they relate to GCP's. Our experience is that it is best not to use GCP's with RTK imagery especially if you have a localized site. Reason being that in processing you have to weight the picture against the GCP to determine which one is more right. Most processing software doesn't account for this well and gives the majority of the weight to the GCP's and makes the images do what they want. You can go in manually (or setup a template I guess) which basically tells the machine that the images need more weight but by default it's not and none of the cloud processing solutions even had the option. Most recently we have been working with DroneDeploy along with several others with different models of drones to determine their actual accuracy as they are not created equal. By default they use 32ft (10m) expected accuracy for the images which is what uncorrected GNSS gets so the GCP's can do whatever they want. Now we have it tuned to 3-5cm (drone dependent) per the hardware of the RTK solution. You would think this would fix it but in reality it only makes it a little better. Think of your map as a pin cushion. What happens when you push a pin? It sucks down right? Because of the way that photogrammetry creates accuracy from matching visual points the effect of the GCP is localized and the further you get away from it the worse the error is going to be - just like anything else. So if you have two GCP's the midpoint of the line between them is where you are going to get the most calculated error. In general averages each image covers about 300-400ft in width but you are overlapping 75% so the 9 or so images that were tagged with the GCP is actually only about 600ft on the ground so at 200-250ft your GCP shouldn't be any further apart than about 1000-1200ft apart at most. It would be better for them to be 600ft or less so that they can feed off each other but on 100ac+ construction sites where the terrain is constantly disturbed it is just not feasible in time or money to set and maintain 10-15 GCP's every time you got out and that's a minimum.
 
Reason being that in processing you have to weight the picture against the GCP to determine which one is more right. Most processing software doesn't account for this well and gives the majority of the weight to the GCP's and makes the images do what they want.
Metashape makes it pretty simple to assign weighting however the user wants. I have not done nearly enough with the Evo2 RTK to make a stab at what the weightings should be, but I'll get there.

It does seem to me that a GCP would still be more reliable than any "air point" simply by the nature of how they are collected. But, for now, I have to admit that I have not determined it for myself. There are so many factors in play.
 
Metashape makes it pretty simple to assign weighting however the user wants. I have not done nearly enough with the Evo2 RTK to make a stab at what the weightings should be, but I'll get there.

It does seem to me that a GCP would still be more reliable than any "air point" simply by the nature of how they are collected. But, for now, I have to admit that I have not determined it for myself. There are so many factors in play.
Too true Dave, I always use GCPs, and the error in the vertical height is always greater than the horizontal. The error is usually at a minimum of a third of the horizontal measurement.
Regards
 
Metashape makes it pretty simple to assign weighting however the user wants. I have not done nearly enough with the Evo2 RTK to make a stab at what the weightings should be, but I'll get there.

It does seem to me that a GCP would still be more reliable than any "air point" simply by the nature of how they are collected. But, for now, I have to admit that I have not determined it for myself. There are so many factors in play.
This would be great! I would start at 2cm. Although I have been seeing better with the AE2 Pro RTK.

The point was that we cannot do this with the GCP's so just changing the image prediction may not be enough or even the right side of the coin to focus on. The issue with GCP's is that I do not know of very many if any that use geographic datum in their GCP's. By that fact you are forcing the drone imagery to a CRS that is almost certainly not a 1:1 transformation so it is better to let the RTK do what it does in geographic datum to maintain relativity and then transform the entire map.

This is especially problematic in construction because our machine control base stations acquire a coordinate by a "I'm Here" method which with an uncorrected GNSS is at best 3m error from the actual global location. The coordinates that are on the survey data collectors of the layout crews is not from a known site coordinate established by a Survey or from an RTK network. This is still a feature request with the Emlid group because their software will not localize.

At the end of the day we were seeing a possible two-fold problem. One with geographic vs grid coordinates and another with image predicted accuracy.
 
The point was that we cannot do this with the GCP's so just changing the image prediction may not be enough or even the right side of the coin to focus on

In MS, you have full control of precision assigned to both camera positions and control points. In fact, you can assign precision to the entire set or individual cameras or GCPs.

a.JPG

The issue with GCP's is that I do not know of very many if any that use geographic datum in their GCP's. By that fact you are forcing the drone imagery to a CRS that is almost certainly not a 1:1 transformation so it is better to let the RTK do what it does in geographic datum to maintain relativity and then transform the entire map.

This is very true. Offsetting the modeled surface by a determined value based upon its relationship to the GCPs rather than trying to force the correction in processing may be the best way to handle it.
 
  • Love
Reactions: adm_geomatics
In MS, you have full control of precision assigned to both camera positions and control points. In fact, you can assign precision to the entire set or individual cameras or GCPs.

View attachment 3390



This is very true. Offsetting the modeled surface by a determined value based upon its relationship to the GCPs rather than trying to force the correction in processing may be the best way to handle it.
Very cool. Every local software I have seen has the image accuracy but it's been so long since I processed locally that I don't recall if they have the control points option. Hopefully this will become a more common discussion as more people get RTK because normal people (not us, lol) don't even know to look for it much less find it. With the options you have shown there I would bring the Camera accuracy down to 0.15ft and the Markers up to 0.05ft. Degrees kind of is what it is because RTK doesn't change the function of the camera.
 
  • Like
Reactions: Dave Pitman
With the options you have shown there I would bring the Camera accuracy down to 0.15ft and the Markers up to 0.05ft. Degrees kind of is what it is because RTK doesn't change the function of the camera.
The option values in the screen grab are the defaults which assume non-precision gnss for aerials. MS also has the option to pull the precision for camera positions directly from the image exif if it exists. It does exist in the images on the Autel RTK. Then you just set your desired GCP precision and go from there.

eg.

c.JPG
 

Attachments

  • c.JPG
    c.JPG
    12.7 KB · Views: 2
Last edited:
  • Like
Reactions: adm_geomatics
The option values in the screen grab are the defaults which assume non-precision gnss for aerials. MS also has the option to pull the precision for camera positions directly from the image exif if it exists. It does exist in the images on the Autel RTK. Then you just set your desired GCP precision and go from there.

eg.

View attachment 3392
This is specifically what doesn't exist in any cloud processing software right now. We got DroneDeploy to adopt the P4RTK, Evo II Pro 6K RTK and the Yuneec H520E RTK. Now we're all dialed in. I know the Yuneec doesn't have those values.
 

Members online

Forum statistics

Threads
4,277
Messages
37,605
Members
5,969
Latest member
KC5JIM