Phantom 4 pro and gravel pile mapping

[Dave Pitman]

Can you please indicate any links to that? I find something interesting.

I can't think of where I ran across that but I'll try and see if I can find it. When you think about it, it makes sense. We all know that we need GCPs dispersed around the entire area in x and y axis to be able to have good accuracy. If the terrain varies much in elevation, and all of your GCPs are all at relatively the same elevation, the software has nothing the tell it exact differences like it has to tell it exact differences in horizontal. There is no way that the scan of a bluff, for example, will be as accurate if the GCPs are all at the top of the bluff and none at the bottom. So if you have a relatively flat plane like the stock pile yard. And then you have piles that can be 20 meters high. If you don't tell the software exactly what the difference is in a couple of samples. There is no way it will be anywhere near as accurate in the z axis as it is in the x and y with the GCPs. I would think that the z axis is equally important for volumetrics.

 

[ArrUnTuS]

I can't think of where I ran across that but I'll try and see if I can find it. When you think about it, it makes sense. We all know that we need GCPs dispersed around the entire area in x and y axis to be able to have good accuracy. If the terrain varies much in elevation, and all of your GCPs are all at relatively the same elevation, the software has nothing the tell it exact differences like it has to tell it exact differences in horizontal. There is no way that the scan of a bluff, for example, will be as accurate if the GCPs are all at the top of the bluff and none at the bottom. So if you have a relatively flat plane like the stock pile yard. And then you have piles that can be 20 meters high. If you don't tell the software exactly what the difference is in a couple of samples. There is no way it will be anywhere near as accurate in the z axis as it is in the x and y with the GCPs. I would think that the z axis is equally important for volumetrics.

It is fundamental and I understand what you mean but knowing why will make us not forget it

If I have a little time tomorrow and I remember....... I will try to look for something on the subject as well.

 

[Dave Pitman]

If I have a little time tomorrow and I remember....... I will try to look for something on the subject as well.

Good!

 

[Dave Pitman]

@ArrUnTuS I've relocated one citation and I know I've read others. I'll post them if I find them and you can do the same.

Capturing Ground Control Points

Placing the GCP Markers
Placement of GCP points should be scattered throughout the area of interest being mapped. If the area being mapped has noticeable elevation changes, make sure your GCP placement (hills, mines, valleys etc.) accounts for this by placing them in the different elevations on a relatively flat surface.

 

[ArrUnTuS]

Yes, when the difference in height is great it is understandable. But how much is noticeable elevation changes when it comes to piles?

Or put it another way, quantitatively, 10,20,50 meters?

When the difference is high I already had it in mind but now I am not thinking about piles of 50 meters high, I don't know. The idea is to have a reference and then apply it to the real world. I was thinking that someone had done a real study where they give you recommendations based for example on the different types of GPS based on their accuracy.

 

[Jesse G]

I had never heard of putting GCPs at different heights to try to give more precision to the Z axis.

Can you please indicate any links to that? I find something interesting.

Here is a great read about accurately creating 3d models with nadir and oblique imagery...http://www.tandfonline.com/doi/full/10.1080/22797254.2017.1313097...while it doesn't talk about the importance of GCP placement at diffetent vertical levels it does explain and illustrate the importance of oblique imagery for accurate 3d reconstruction of surfaces that include complex vertical features. Thought you guys would enjoy it if you haven't read it already.

I can say from experience that it is definitely important to have GCPs placed at differing elevations to achieve accurate results. Last year I mapped a landfill that had a large pit in the center of it that was about 40 feet deep, this pit was small in comparison to the rest of the site as it only took about 1/8 of the area of the site with most of the site being farely level. I placed a grid of GCPs around the site and 1 GCP in the center of the pit, initially just planning on using it as a check shot. Upon processing I found the pit elevation to be off by over 3 feet, once I added in the GCP in the center of the pit that dropped down to just over a foot which was still not acceptable so I took a trip out to the site and shot the pit in manually with my RTK system and combined the data sets. Upon returning to the site a few months later for another flight I then used 3 targets placed in the pit arpind the bottom slopes and flew a small grid of photos at 40' lower than my main grid and results were perfect. In the end I learned that placing GCPs at slope limits really helped to nail down slopes and that even 40' in elevation change can have an impact in overlapping and lead to poor results. Both factors staked atop eachother really made a mess of a fairly simple and small pit.

 

[TAZimmer]

@TAZimmer Are you setting at least 1 or more gcps on the top of any piles? I've read that for GCPs to be able to dial in z axis on varied terrain (like stock piles) they must be placed dispersed both horizontally and vertically in order to calibrate the difference in elevation in addition to horizontal scale.

I can tell you that I have done some volumes for owners that even after explaining greatly reduced accuracy by not using GCPs have elected to not pay to use them. They said that what they use the data for, "it's close enough". I cannot tell you for what purpose "close enough" and to "what" is though as I have no way to quantify without truthing obviously.

Yeah, we always make sure at least one GCP is on top of each stockpile. We also make sure we have a few validation points registered as well along the slopes / top of a stockpile. We use a Spectrum RTK system to survey in the GCPs and validation points, so really it doesn't take very long to get a good set of points for the modelling.

I've done some comparisons between 3D meshes built using stereo imagery alone and stereo with GCPs. The difference can be pretty big (> 3m vertical on a 10m pile). But I've never compared meshes built with GCPs on ground only versus GCPs on ground and on stockpile tops. Should be easy enough comparison to run, though. I'll crack open one of our previous surveys and see if I can run the two scenarios, and then let you know what I find.

 

[ArrUnTuS]

Interesting reading jesse G and good TAZimmer appraisals but we are still in the same.

We depend on the accuracy of the coordinate system we use. A RTK system will always be better than a GNSS but the error rate remains at 1.5 times more error in Z than in XY.

I've been looking for information and can't find anything that specifically talks about GCP's recommendations when the height difference is X. We have to rely on our intuition and experience. Without a doubt, putting a target on top of the stack will improve the calculation but when it's advisable to do so is the question..

I'm gonna transfer the doubt to topographers. In principle, they have studied more on the subject and can still clarify it for us. If they can make it clear to me, I'll tell you.

 

[Dave Pitman]

I've done some comparisons between 3D meshes built using stereo imagery alone and stereo with GCPs. The difference can be pretty big (> 3m vertical on a 10m pile). But I've never compared meshes built with GCPs on ground only versus GCPs on ground and on stockpile tops. Should be easy enough comparison to run, though. I'll crack open one of our previous surveys and see if I can run the two scenarios, and then let you know what I find.

That would be valuable insight. I look forward to your results.

It is so interesting that both of the cloud services I have used, Maps Made Easy and Drone Deploy swear up and down that for relative accuracy, GCPs are unnecessary. I believe your results vs their insistence.

 

[Dave Pitman]

Thanks @Jesse G . for adding some data points to the discussion. and for the article.

 

[ArrUnTuS]

I already have the first answer and it leaves us in the same place. The recommendation is to distribute GCP's well over the entire terrain taking into account height changes and setting them at different heights as well. Same thing we knew until now. I keep asking to see if anyone can tell us something more specific.

It may be that no one has studied it and it is taken as a general rule to have the height as a parameter to take into account but without following a specific procedure. I don't know but since asking is free

 

[ArrUnTuS]

The answers I've come up with leave us the same. Try to disperse the GCPs well and have them at different heights if possible to achieve accuracy on all 3 axes.

Regarding the issue of calculation of volumes, this article of comparison of volume calculations using traditional techniques or photogrammetry is interesting.

How does photogrammetry measure up? A comparison - Pix4D

 

[Mike Nevins]

I think there is some very deep discussions on this forum about accuracy. I just saw that DD upped their price for Pro to $ 129.00 per month. Also, I would find it difficult to say they DD can achieve 98% accuracy without GCP. But if the client is happy, go for it. Also, you can use GCP in MME. One option they offer is not true GCP that are expensive, but using Geo referenced maps. I just went through their tutorial on this, but will have to actually ground truth this to make sure it works. Ideally, I would want to use LiDar and then compare to MME. Totally agree on the pay as you go through MME. You can also use their DEM to extrapolate into Topo (Contour) maps which we have been asked to do.

 

[TimmyD]

That is correct. The dji drone flies and captures images like normal. But the on-board gnss receiver also logs the position of the drone when the image is captured, only much more precisely.

Back at the office, the log file from the gnss receiver is processed with CORS logs to correct the positions for atmospheric anomalies. This is Post Process Kinematic (PPK) as opposed to Real Time Kinematic (RTK) which corrects the logs of a rover in real time. PPK does not require a live connection to a base or a CORS network making it an ideal tool for UAS work.

Then the corrected log is used to re-write the geo-tags in the images exif data. Then they are processed like normal only this time with centimeter accuracy, in theory anyway.

Hey Dave,
I know this post is old but wanted to ask you. Do you have a separate receiver mounted on your P4P? Or are you pulling log files from the P4P itself? Either way can you tell me your work flow? I use EZsurv to post process my GCP's but it can also post process all the images taken by the P4P but I need log files from the P4P. I am not sure if you are converting the flight log or using a different receiver all together. Thanks for any help!