Ground Surface Accuracy

[Jason Frehner]

Hi,
I am new to photogrammetry and UAV's. I wish to understand if I want to obtain high resolution 3D models of the ground surface within a few mm of accuracy will photogrammetry suit or would LIDAR be more appropriate. The ground I wish to assess is free from objects such as trees and grass.

Thanks

 

[Richard R]

' Bare earth' gives you the best change of getting good results regardless of the technique used. Give us some more details. Are you looking for mm level elevation accuracies, ground location, both? What level of ground resolution (GSD) do you want. And how big is the area that you want to map? What bird do you plan on flying? The 'right' answer greatly depends on questions like those above. In any case, mm levels are going to be very difficult (translate that to expensive) to obtain.

 

[Richard R]

You might want to read this Drone LiDAR vs Photogrammetry: A Technical Guide | SPAR 3D

 

[doktorinjh]

within a few mm of accuracy

We control our projects with a reasonably high level of accuracy and we're getting around 0.05'-0.1' (15mm-30mm) in horizontal accuracy and 0.3'-0.5' (100mm-150mm) in vertical accuracy. We can do a little better than this, but I would say that this is reasonable to expect. Getting horizontal and vertical accuracies better than that is going to be pretty tough. LiDAR is likely to be a better solution. What are you trying to measure with those tolerances?

 

[Richard R]

Actually, airborne lidars aren't going to do much if any better than what you're quoting! And I bet you have to work pretty hard and be very careful to get down to those numbers. The other issue with lidar is gsd. 10-20 points per square meter is about the level that you will get with lidar. Actually, that gives you a lot of useful info, but there are limitations. I've built and flown a number of lidar systems for Air Force and other agencies on all kinds of aircraft and you are always going to be trading off gsd for area coverage rates. Be interesting to hear more from the OP about just what he wants to do and why he thjnks he needs such hi red.

 

[Richard R]

We control our projects with a reasonably high level of accuracy and we're getting around 0.05'-0.1' (15mm-30mm) in horizontal accuracy and 0.3'-0.5' (100mm-150mm) in vertical accuracy. We can do a little better than this, but I would say that this is reasonable to expect. Getting horizontal and vertical accuracies better than that is going to be pretty tough. LiDAR is likely to be a better solution. What are you trying to measure with those tolerances?

Are those numbers relative or absolute?

 

[Jason Frehner]

Ok I am needing quite accurate resolutions 1-2mm, the uav can be close to the surface if needed.

 

[GVDD scout]

Hi,
I am new to photogrammetry and UAV's. I wish to understand if I want to obtain high resolution 3D models of the ground surface within a few mm of accuracy will photogrammetry suit or would LIDAR be more appropriate. The ground I wish to assess is free from objects such as trees and grass.

Thanks

I can not speak to the accuracy/repeatability of LIDAR, as as we are aiming for accurate orthomosaics, and to a lesser degree, DSM generated contours and volumes. My experience with P4P and various flight planning and processing packages, (finally settled on Pix4D Mapper and Pix4D Capture), and consultation with our surveyor and others that get the type of accuracy you request only from survey grade GPS points, would lead me to tell you that drone accuracy and more importantly, repeatability, require that you have both VERY accurate Ground Control Points that can be located in the imagery in x,y, and have little to no gradient in the ground near them, and that your UAS has RTK GPS/GNSS. The tutorials on the Pix4D, Drone Deploy and Drone 2 Map websites will all reinforce this. Also, the site size and thereby the number of images used will need to be such that image overlap results in a resulting ground sampling distance at or under the desired accuracy. A mentor shared with me an example of the above done to the high limit of attainable precision: GSD=1.1cm, GCP's RMSe=0.085m , the resulting model only has meter range absolute accuracy.

 

[Richard R]

Ok I am needing quite accurate resolutions 1-2mm, the uav can be close to the surface if needed.

Not impossible, but definitely not easy or cheap. Also, what kind of areas and coverage rates are you looking for? higher resolution usually means lower coverage rates. a whole bunch of variables that have to be juggled.

 

[doktorinjh]

Are those numbers relative or absolute?

Those are the RMSE values when compared to check shots. When we're in the field, we survey certain points for use as GCPs and mark others to be used as checks. We process the imagery using the GCPs and then compare the point cloud to the check shots to see what our accuracies are. The check shots are not used in the calculation of the model and the number of checks varies by the project size. We use an RTK GPS that has been calibrated to existing transportation monuments near the project site.

 

[Jason Frehner]

The areas are typically a thin strip of 5m at a length of a a few Km.

 

[Bigriver]

Wow! 1-2 mm accuracy? Like millimeter? I would not guarantee that kind of accuracy using conventional or GPS survey equipment, much less via photogrammetry or lidar. A US penny is only 1.5 mm thick.

That said, Pix4D claims 1-2 GSD horizontal and 2-3 GSD vertical accuracy (I think?). So with a P4 and12MP camera, you would need to fly at something like 4m or less altitude, providing a 7m or so image width and 5m length. With the recommended overlap, that would be a LOT of images for a "few Km" project. (1000 images per KM per flight path)

Maybe not impossible, but not very probable. UAV photogrammetry has many benefits, but one must understand it's limitations too.

 

[Richard R]

Wow! 1-2 mm accuracy? Like millimeter? I would not guarantee that kind of accuracy using conventional or GPS survey equipment, much less via photogrammetry or lidar. A US penny is only 1.5 mm thick.

That said, Pix4D claims 1-2 GSD horizontal and 2-3 GSD vertical accuracy (I think?). So with a P4 and12MP camera, you would need to fly at something like 4m or less altitude, providing a 7m or so image width and 5m length. With the recommended overlap, that would be a LOT of images for a "few Km" project. (1000 images per KM per flight path)

Maybe not impossible, but not very probable. UAV photogrammetry has many benefits, but one must understand it's limitations too.

And flying in that manner is going to give you little or no side to side overlap. Unless the mission is over flat, tree and shrub less terrain, you are going to quickly loose sight of it, making things just that much more difficult. And a Lidar isn't going to fare much better. Be interesting to get a bit more info from the OP. When I was still developing Lidars for DOD, customers often came in with unrealistic requirements that, after some discussion about how they intended to use the data, were eventually relaxed but still did what they needed.