RTKLIB with P4RTK

[amicron]

PeteM, we have already discussed this earlier in this thread. It’s not just about accuracy and how much the equipment costs. Like I said before, repeatability and confidence play a big part too. And GCPs shouldn’t even be a factor because you have to have them no matter what you use.

You are obsessed with $3-4000 difference being a lot of money for survey equipment. To buy the survey equipment to do conventional or GPS surveying you would likely spend anywhere from $20-100k. And spend more time on the ground. There is already a huge savings using drones, so debating the last couple thousand is silly to me.

Bigriver, I don’t have any comparisons or hard numbers to answer your question but I have more confidence in the vertical accuracy and precision with the P4RTK. That’s what’s important to me.

And regarding the calibration of the camera, the cameras are the same and one is not “better calibrated.” The difference is with the P4RTK, you are able to use the original distorted pictures and DJI provides the calibration values to use in post processing. This is how photogrammetry has always been done. Photogrammetrists would regularly have to send out their camera to have it calibrated and they would get the calibration values specific to their camera to use on their pictures. On the P4P, the pictures are automatically undistorted with unknown values in the camera and the photogrammetry software will calculate a calibration based on those pictures.

 

[PeteM]

I do not agree.

When drone technology for topographic surveys was born, I was among the first to invest in it, because the objective advantages compared to the classic way of surveying the territory were HUGE.

Now, I would be ready to invest in an RTK drone, if and only if, it was possible to have a topographic product "that works" without GCPs. But considering that GCPs are always needed and given that using a Phantom and a couple of CGPs you can have products "that works", in my opinion it's not worth it.

And the total lack of scientific data to support drones with RTK on board does not help me change my mind...

 

[amicron]

I’m pretty sure these companies/websites have comparisons or white papers regarding RTK/PPK but not sure they qualify as “scientific.” I don’t think it will change your mind though.

Geocue
DroneDeploy
DroneDJ
Aerotas
Microdrones

 

[doktorinjh]

when DJI says that the camera on the RTK model is "better calibrated"

You get a much better product when you're using a calibrated camera, as opposed to using an uncalibrated camera. Even with precise geolocation information, your model can be skewed because without a camera calibration the photogrammetry solution will be influenced by lens distortion. We've had checkshots be feet off because we had the incorrect camera picked or had the camera settings wrong while in flight. The two really work hand in hand to produce an accurate model. I assume (I'm not an PhantomRTK user) that they are providing some sort of lab calibration file with the delivery of the drone? Ours was done in a lab and we have to add that file to our PhotoScan or Pix4D processing.

With better geolocation data, you should get a better overall model because the structure-from-motion processing is based on having a reasonable idea of where the photos were taken compared to each other. With GCPs, you're likely to see some vertical accuracy loss as you move away from your ground points, although it might be so minimal as to not be discernible from other noise.

 

[doktorinjh]

if and only if, it was possible to have a topographic product "that works" without GCPs

We are trying to streamline our workflow to a PPK enabled drone and a set of Aeropoints. I know that @amicron doesn't like the Aeropoints, but we've done enough field verification to trust that they are giving us an accurate location. We've run these against their Propeller Correction Network (basically a CORS solution) and our own local base (Trimble R7), and then compared the results to fast static, VRS, and Trimble's RTX service, and compared to known points. I trust the Aeropoint positions, our Professional Land Surveyor trusts the data, and all the solutions come out less than a tenth of a foot apart.

With that said, I always process directly from the PPK data (using a calibrated camera file) and use the Aeropoints as checkshots. On really important jobs, we'll still check topo with the VRS network and verify against known monuments, but we're starting to find that it's becoming almost too redundant. On our last job, our RMSE values were 0.14', 0.24', and 0.15' for X, Y, and Z, respectively. This is well within the project requirements and we could probably chase down the error in the Y value to get it even tighter, but there was no need. I know the RTK/PPK producers like to claim 3cm (0.10 feet) accuracies, but we're seeing RMSE errors typically under 0.15' in the horizontal and 0.25' in the vertical, on average.

 

[doktorinjh]

I find it very "suspected" that no one has compared the result obtained from a classic drone (with GCPs) with one obtained from an RTK drone (without GCPs). Maybe because the first solution continues to be better one ?

We've done this plenty of times and the answer is that it kind of depends. In essence, the GCPs are always going to drive the model to it's final location. They are usually given the higher accuracy setting in the processing software so you are telling the program that there is no room for movement at the GCP location; "make the model fit this exact spot and to hell with the rest." This makes sense because you can identify the exact location of the target and you (should) have a very precise and accurate location of that target. Of course, there's some thought that the model will deviate the farther you get from the GCPs, causing higher error values where there's low correlation. A precise geolocation of the imagery should help to reduce that error.

In a perfect world, at a perfect mapping site, I would fly with PPK and reference to a locally established control station. I would also have a dense collection of GCPs and also collect a ton of topographic control. This would give me and all of the surveyors the warmest of fuzzy feelings about the survey and we'd have so much great data to back it all up. I would process using the ground control points and use the geolocation data and orientation data to help with alignment and processing speed. I would then use the check shots to give me independent checks against the point cloud and bring them back into any additional model manipulation (post-ground classification or after mesh smoothing, etc.). This is (perhaps) the best case UAV mapping scenario, but it's not one we're always able to use.

The value in RTK/PPK processing is also the ability to map areas where we can't lay GCPs (private land, unsafe terrain, inaccessible, etc.) or mapping sections that are extremely linear where your GCP layout wouldn't create a good block (highway cooridors). We love having the PPK solution as our primary mapping technique and using aerial targets first as checkshots, but also as backups as GCPs, should anything go wrong.

 

[PeteM]

In this post I talk about relative accuracy.

In "commercial" drones the highest error does not come from the GPS installed inside the drone but from the optical technology.

Phantom 4 RTK does not mount any camera suitable for photogrammetric purposes but it has only a system that allows to calculate the (absolute) position in flight in a more precise way.

The photogrammetric output that comes from a Phantom 4 RTK is not a "survey grade" product. It becomes so only if the beautiful model that is obtained is corrected and scaled according to a skeleton with survey grade precision... like the past... like the old and cheap P4...

 

[doktorinjh]

This is a complicated process and I can't blame you for not understanding how the components work together, but I'm going to attempt to go through your statements so that others might learn from this. You've obviously written off aerial mapping using RTK/PPK enabled drones and that's your choice, but it doesn't mean that the thousands of others that use this on a daily basis are providing incorrect information.

In this post I talk about relative accuracy.

In surveying and mapping, no one cares about the relative position of data. That's not surveying, that's measuring. If I know that something is exactly ten feet from another point, that's a relative position, but it doesn't mean anything in the context of where it is in a geographic reference system. When we're performing a survey, we want to know an absolute position because that's what surveying is.

In "commercial" drones the highest error does not come from the GPS installed inside the drone but from the optical technology.

I'm assuming that the "commercial" drones that you're referring to are RTK/PPK enabled drones. You're halfway correct in the sense that an RTK/PPK enabled drone must work together with a lab calibrated camera to provide an accurate and precise topographic model. This is well known and it is why I have to enter lab calibration data into my photogrammetry model for my specific camera to get accurate results. This is why I have to fly with manual camera parameters so that the focal length does not change with each photo. This is the same thing that we do in manned aerial mapping. If you try and process your calibrated camera data using recreational GPS geolocation data, it won't be accurate. If you try and process your precision geolocation data without a calibrated camera file, then you won't get accurate topographic data.

Phantom 4 RTK does not mount any camera suitable for photogrammetric purposes but it has only a system that allows to calculate the (absolute) position in flight in a more precise way.

Any camera can be calibrated and used for precision mapping. Some are (much) better than others, but once their lens distortion errors are known, the software can "un-distort" the image and rectify the known stretching and warping that occurs. Even our $750k metric aerial camera has lens distortion.

The photogrammetric output that comes from a Phantom 4 RTK is not a "survey grade" product. It becomes so only if the beautiful model that is obtained is corrected and scaled according to a skeleton with survey grade precision... like the past... like the old and cheap P4...

Again, this is an incorrect statement. If it were, than there would be tens of thousands of Phantom 4RTK users scrambling to return their gear and there would be a dozen other RTK/PPK hardware companies getting sued for lying to their customers. It's not like we just get the data and say "OK, this is probably right, no need to check it." The methods are sound and the ability to create an accurate and precise model from nothing more than a lab calibrated camera and precise geolocation data is well established. They didn't invent this method out of thin air at the DJI factory, this is based on decades of photogrammetric work.

I've already responded to why we might want to use RTK/PPK, even when we have GCP coverage and I've already addressed why having all of them working together will give you the best product of all. Can you get an accurate and precise topographic model using only GCPs? Of course you can, that's not even a question. But, there are many circumstances where utilizing an RTK/PPK enabled drone will save you time, money, processing headaches, and has several additional advantages. So, feel free not to buy one, but don't spread misinformation about their abilities.

 

[PeteM]

Can you get an accurate and precise topographic model using only GCPs? Of course you can, that's not even a question.

Phantom 4 RTK was presented by marketing as the drone that can fly over an area and return the topographic model in a snap.

Who buys it must know that in any case it is necessary:

1) at least a couple GCPs
2) at least a couple of checkpoints
3) a lab calibrated camera

Moreover those who purchase it must know that without an expensive GNSS service and without internet connection the RTK does not work. Moreover those who purchase it must know that in case of RTK not working, the PPK alternative is not "natively supported" and goes through DIY solutions and additional equipment and programs to be purchased.

Finally, those who buy it must know that they will not get a very different result than an old Phantom with a dozen GCPs.

Now its up to you...

 

[doktorinjh]

Phantom 4 RTK was presented by marketing as the drone that can fly over an area and return the topographic model in a snap.

So are you only focusing now on the Phantom 4 RTK, or all RTK/PPk in general, because you're shifting directly to that model and not RTK/PPK as a whole? I have 3 RTK/PPK rigs and none of them are the Phantom 4 RTK. As of comment #42, you were referring to RTK/PPK drone solutions in general and now you're directly referring to one piece of equipment. Either way, I have no doubts that with the P4RTK in the hands of someone who has the correct setup, that they should be able to generate a "topographic model in a snap." I even believe that the model will stand up to rigorous map checks. I also have no doubt that someone with the P4RTK couldn't make a map. It's a tool and it's only as good as the user. But, it's a capable tool.

Who buys it must know that in any case it is necessary:

1) at least a couple GCPs
2) at least a couple of checkpoints
3) a lab calibrated camera

If used properly, GCPs are not necessary. Why do you keep considering that they are mandatory? An unused GCP is just a checkpoint. If anything, using "at least a couple GCPs" has the potential to ruin your model and is bad practice. Either use all GCPs or all RTK/PPK. Checkpoints are always a good idea, regardless of the mapping that you're doing. They are the only way to verify your data. That is not a by-product of RTK/PPK mapping, but for all mapping. A lab calibrated camera datasheet came with my setup but the camera calibration files that are included in the camera databases also count as optimized camera parameters. You can also create your own field calibration using lots of aerial targets and the photogrammetry software. There are several ways to achieve this, but the bottom line is that the software has to know what the lens distortion looks like.

Moreover those who purchase it must know that without an expensive GNSS service and without internet connection the RTK does not work. Moreover those who purchase it must know that in case of RTK not working, the PPK alternative is not "natively supported" and goes through DIY solutions and additional equipment and programs to be purchased.

Again, this is survey grade mapping and, as others have said, additional equipment and services may be needed. I have $25k-$100k of equipment with me every time I enter the field, and that's not including the truck. If you're in the precision mapping game, you will need some way to either get positions of your GCPs and your checkshots, regardless. Whether you pay someone else to do it or not, they are not an option for any kind of mapping. If you're working in areas that don't have cell coverage, then you must have a backup plan, that's not the RTK/PPK drone's fault. Learn how to look up your nearest CORS station and download data. Learn how to process with RTKLIB. Buy a different setup. It doesn't matter because the discussion was around if RTK/PPK provides you with an accurate map. You're deviating from the course. If you want a $25 setup, you're in the wrong business.

Finally, those who buy it must know that they will not get a very different result than an old Phantom with a dozen GCPs.

This has never been in question. Here, let me quote myself from the last comment:

I've already responded to why we might want to use RTK/PPK, even when we have GCP coverage and I've already addressed why having all of them working together will give you the best product of all. Can you get an accurate and precise topographic model using only GCPs? Of course you can, that's not even a question. But, there are many circumstances where utilizing an RTK/PPK enabled drone will save you time, money, processing headaches, and has several additional advantages.

I still use a Phantom 3 and some bucket lids at times. It's slower, produces poorer resolution images at similar heights to a P4P, has a reduced flight time, and looks cheap and outdated to the client. Will it give me a similar bare earth DTM with enough processing and GCPs? Yep. Is it the best tool in our fleet? Not even a chance.

 

[PeteM]

In my work I do NOT deal with georeferenced maps. My need is to measure distances and heights within construction sites and excavations in general, in order to obtain extensions and volumes. Its a job that 10 years ago I started doing with a simple total laser station using several whole days and at least 2 people even for small lands. Today, thanks to drones, I can do the same thing in like 2 hours. In these 2 hours, 20 minutes are dedicated to the flight and the remaining 100 minutes are dedicated to place and measure GCPs.

My dream would be to arrive on the site, fly over it for 20 minutes with a drone and return back home to work out relative measures. Moreover I don't need high precision, 10 cm of error (but even on Z) would be perfect.

With the P4 (normal) I discovered that this is possible if the drone stops in flight 3 seconds in each point of photo shooting. This is why I say that GPS is not the main cause of relative measurement errors. The model that comes out with this way of flying has correct and very little distorted measures. Unfortunately this way of flying kills the battery in minutes...

When the P4 RTK came out I was hoping to have found this solution... but unfortunately it is not so...

I also thought of using physical scale constraints... for example, lay down some colored bars of known length on the ground so as to correct the model with those certain measurements... but for the relative heights ?

If you have any ideas...

 

[doktorinjh]

In my work I do NOT deal with georeferenced maps.

For volume estimates and non-referenced work, an RTK/PPK enabled drone is not likely to be of benefit to you. A P4 flying with the regular, on-board GPS will suffice for almost all volumetric calculations that do not require detailed location information. I've run tests using GCPs and no-GCP models and there is no negligible differences between the models for volume measurements. You can try this yourself by running both models, taking the volume measurements, and comparing the results. I ran over 20 comparison measurements and the spread between all of the models (GCP and no GCP) was less than 2.5%. The models without GCPs were good relative to themselves.

We always fly without stopping and have never had issues, so you should probably experiment with that setting. If you want to slow the drone down for taking pictures without stopping, then you can increase your forward overlap.

 

[PeteM]

I've also done many experiments.

The problems are almost always on the Z axis.

If the area to be "measured" is quite large, the model tends to curve in the central part (like an extended sheet held only at the edges). If the drone flies at a height greater than 100 m and/or at a speed greater than 5 m/s the errors along the Z axis begin to become non-marginal. If the drone photographs reflective or too flat surfaces errors along the Z axis become serious.

The GCPs form a skeleton that straightens all this. Usually I put one in the middle (which is almost always the lowest point of the excavation), some at various heights and then near the four ends of the flight. I use some of these for verification.

I would not trust measures not verified in this double way.

 

[Nico Jaeger]

Hi there, if a smart and easy PPK solution is still of interest here (also to minimize GCP's/Check Points) I encourage to trying our ApplanixPPK Web App Cloud Solution. It's straight forward and provides excellent GNSS position performance.

Go to:Applanixppk.com