Thoughts on the Phantom 4 RTK

[Bigriver]

I was just wondering what some of your thoughts (or concerns) are on the new Phantom 4 RTK?

At this point there is little to gather on accuracy of the mapping other than what is posted by DJI advertising. As a surveyor, I am trying to determine the advantages of the system.

Concerning GCP’s and checkpoints… My normal procedure is to set twice the GCP’s than are needed for accurate georeferencing of the mapping. I use half the points as GCP’s and the other half as check points. I need these check points to validate my mapping. Does the RTK eliminate the need for check points? Not for me. I need that proof, that I am providing accurate mapping. Maybe it just eliminates the points I would use as GCP’s? But still, I must collect positions of check points, either conventionally or with GPS to verify my mapping. So there is still ground work I must complete.

Also, concerning RTK. In GPS RTK surveying, we normally collect topo quality points by observing the point for three seconds, maintaining a static position. “Control” quality observations require three-minute static observations. I wonder what quality positions are obtained from a Phantom moving at 10 mph.

I am not trying to shoot down the new technology. Actually, I am eager to justify the purchase of the new unit. I am just trying to understand and put a time/cost savings value of the RTK vs the normal mapping with GCP’s and check points.

 

[Aex_khab]

I half a year using phantom 4 ppk topodrone.ru. At the beginning of use puts a lot of control points. The accuracy was 1-10 cm depending on the conditions. After many completed projects, I began to trust the method, now I put 1-3 control points on one flight, this data is enough.
Completed Phantom 4 ppk projects www.instagram.com/737gnss/.
On all projects the error is not more than 10 cm.

 

[Aex_khab]

 

[vgo195]

I think the PPK solution is better than rtk. Rtk needs a good uhf connection that is not so easy with drone. I think also that you need gcp’s anyway to verify the model. When you say accuracy 1-10cm you mean horizontally or vertically?

 

[Aex_khab]

[QUOTE = "vgo195, post: 1405597, участник: 57517"] Я думаю, что решение PPK лучше, чем rtk. Rtk нуждается в хорошем подключении uhf, которое не так просто с drone. Я также думаю, что вам нужно gcp, чтобы проверить модель. Когда вы говорите о точности 1-10 см, вы имеете в виду горизонтально или вертикально? [/ QUOTE]
горизонтально и вертикально

 

[vgo195]

Sorry I don’t understand Russian but I see your point. How much the topodrone solution cost?

 

[Aex_khab]

[QUOTE = "vgo195, post: 1405614, участник: 57517"] Извините, я не понимаю русского, но вижу вашу точку зрения. Сколько стоит решение topodrone? [/ QUOTE]
Комплект обновления PPK для DJI Phantom 4pro - 2300 usd
Phantom 4pro - 4000 usd
TOPODRONE • Готовность к полетам БПЛА GNSS RTK
Файл из Облака Mail.Ru

 

[Aex_khab]

https://www.pix4d.com/blog/GCP-accuracy-drone-maps

An article about the number of control points without using RTK / PPK. It is said that only 4 control points are needed to achieve an accuracy of about 3 cm. I could not achieve such accuracy without using PPK.

 

[doktorinjh]

https://www.pix4d.com/blog/GCP-accuracy-drone-maps

An article about the number of control points without using RTK / PPK. It is said that only 4 control points are needed to achieve an accuracy of about 3 cm. I could not achieve such accuracy without using PPK.

The article says that it used a Phantom 4 Pro, so why don't you think that you'd be able to replicate the results?

 

[vgo195]

3cm is not difficult to achieve with traditional gcp’s as long as they are well spread and they are measured with rtk gps or total station. Mind that the rtk gps if used with COORS as network rover may have vertical errors more than 5cm between gcp’s. The best method for rtk gps is base-rover which keeps the baselines small and so the errors.

 

[Bigriver]

What about the movement of the aircraft?

Are you obtaining the photos from a stopped/hovering aircraft?

I don't see how we can expect to get sub-5cm solutions for the position with an aircraft moving at common mapping speeds (regardless of whether RTK, PPK, local base station, or CORS).

 

[Bigriver]

I found this article while researching the P4 RTK. Found it very interesting.

First look at the Phantom 4 RTK

 

[John Swinburn - Coptrz]

FYI DJI have just released a firmware update for the P4RTK https://dl.djicdn.com/downloads/phantom_4_rtk/20181205/Phantom_4_RTK_Release_Notes_en_20181205.pdf

 

[Apexpert]

Interesting thread but it would be nice to know when you are talking accuracies the x,y,z or horizontal and vertical.

 

[PeteM]

Unfortunately, today, there is no technology that allows the use of drones by totally eliminating GCPs (even only as a function of checkpoints). This is a fact.

Investing large amounts in RTK drones or working out complex solutions to make the (normal) drones compatible with PPK processing, just to reduce GCPs, doesn't make much sense to me.

If you still need to measure 3 points, it is not a problem to measure 6 or 7...

 

[doktorinjh]

Interesting thread but it would be nice to know when you are talking accuracies the x,y,z or horizontal and vertical.

When not explicitly stated, most errors refer to the vertical because it will always be the largest error for any survey. You can "typically" expect about 1-2x the GSD in horizontal error and 2-4x the GSD in vertical error.

doesn't make much sense to me.

There are a lot of situations where you don't have access to the land that you're trying to map (think highway corridors surrounded by private property) or times when a reduction of even a few GCPs will mean hours of time saved in the field. We have found that we can achieve RMSE values that are identical between complete RTK/PPK surveys and those using GCPs. I wasn't a believer at first, but we've done plenty of surveys that see no improvement between adding GCPs or just using them as checkshots. The technology is there and identical surveys can be performed without the use of GCPs. That is a fact.

 

[Dave Pitman]

You can "typically" expect about 1-2x the GSD in horizontal error and 2-4x the GSD in vertical error.

Howdy doktorinjh,

If the desired vertical error is 2" max., then the GSD would need to be ~.5 in/pixel, correct? For a P4P, that would equate to flying at about 150'.

Have you experimented with the effect of overlap on the GSD/ accuracy question? The folks at MME tell me that greater overlap (to a point) equates to better vertical accuracy of the point cloud. Have you found this to be true?

 

[doktorinjh]

Hi Dave!

If the desired vertical error is 2" max., then the GSD would need to be ~.5 in/pixel, correct? For a P4P, that would equate to flying at about 150'.

Yes, that would be where I would start from. Your overall accuracy is also going to be dependent on a lot of other factors, such as: the type of surface (gravel, grass, corn field, pavement), the quality of your imagery, the accuracy of your survey equipment, your processing software, etc. We're usually within that ±0.2' range, but we'll get outliers depending on the surface and processing (PhotoScan generates pavements surfaces better than Pix4D, etc.).

Have you experimented with the effect of overlap on the GSD/ accuracy question? The folks at MME tell me that greater overlap (to a point) equates to better vertical accuracy of the point cloud. Have you found this to be true?

While I believe that this is true to a point, there definitely is a cost/benefit ratio break-over limit. The more pictures you have, the more "looks" you get at a point, which is how "structure from motion" processing determines the point in the point cloud. Eventually, there won't be any improvement between 5 "looks" and 500 "looks." All you'll do is cost yourself more processing time, data storage, and mapping time.

The closest test that I have for you is when I took a model and ran it using 100% of the photos, then 80%, 75%, 66%, 50%, 33%, effectively reducing the overlap as I went. The removed photos were evenly spaced, or at least as close as I could get based on their locations. After reducing the photo count, I reran the model (using Pix4D), calibrated using GCPs, and compared my check shots to the model. Even using 33% of the original photo count, I did not see a noticeable reduction in the RMSE. The mission was flown using a standard overlap (~70%-80% forward and side, I'd have to dig for the actual values). The biggest problem that occurs with photo reduction is that there is not overlap to create a point, so I started getting small holes in the model that eventually grew to be large holes where there wasn't enough data. You can set the minimum limits to how many photos are needed to create a point and I think that I was using the minimum of 3.

I have not done an altitude (GSD) comparison to see how well the 2x-3x rule of thumb holds up, but I think that's more related to the type of terrain and quality of the imagery, as mentioned above. We're able to create large terrain models from our manned aircraft, which varies in GSD from 4cm (1.5") up to 30cm (12").

Hope that answers some of your questions!

 

[Dave Pitman]

It does, thank you.

<OT>
I'm not a surveyor and I color within the lines, so to speak. I have been providing orthophotos and pile volumes for a few years. Surveyors in my area are not interested in drone scanning for the most part. So it's great to hear experience from a surveyor. (I've also been following the conversation on this other thread)

I've recently had a meeting with an engineering firm with in-house PLS interested in looking into the process for their preliminary project topos. (Usually around 20 acres). They are looking for accuracy within 2/10ths of a foot. They are not interested in the uas end. And, they also want nothing to do with obtaining a surface from a point cloud. They would place GCPs and Check-shots and then want me to deliver a simple surface in an .xml format. They would then determine if it would be suitable for their needs. They are currently doing this with 2 men placing points at a rate of 3-5 acres a day. This is on newly cleared land.

I've indicated to them that it should be doable, with some trial and error and practice. I told them that I would be willing to target and scan parcels that they are already done with and work on providing them data that they can scrutinize and compare to their terrestrial survey. The engineer is very interested, but the surveyor doesn't seem to want to even try. And they are equals in the company so it's kind of a stand off. The engineer reached out to me, btw. I didn't go trying to sell them anything.

It is very interesting to me that the surveyor isn't even curious on what the data will look like when testing wouldn't require any extra work from him and if we were to get to actionable data it would save his firm quite a bit of money. Right now, I'm waiting for some requested info from the surveyor and he seems to have gone missing, so I don't know where they stand as of now.

If I could bother you with a specific question hear and there someday @doktorinjh that would be great. If so, you could let me know by pm and we then use email. I won't bother you on here. I'm in Washington state if that matters.

Anyway, sorry for the OT guys.
</OT>

 

[Bigriver]

Dave - As an engineer and surveyor, you won't catch me picking sides in that fight. But I have experienced much of the same here. Like you were considering, to prove your accuracy to the surveyor, see if you can fly a project they have already ground surveyed. Georeference it with photo-identifiable points from the survey. Then overlay their ground-surveyed points onto your orthomosaic. Hopefully that will help prove its value. It is also a great way for you to test your methods and the results you are obtaining from your processing.

Doktorinjh - I sure wish you would add a P4RTK to your arsenal! Then you could do all our beta-testing and answer all our questions before we make the leap. But your insights on the whole PPK/RTK and uav photogrammetry process are very helpful and much appreciated.?

Right now, I think the biggest drawback (for me) is having to use the DJI GSR app for the flight planning, without allowing for the use of third-party planning apps. It definitely has some possible advantages in my line of work.