Drone Deploy changes my camera setting


#21

Huh? I was just responding to you talking about “consistency of writes”. I don’t agree that adding a couple of mb changes that consistency. Are we back on topic? :wink:


#22

Hopefully I can clear this up!

We capture 4:3 images because historically the DJI SDK has been more reliable in this mode, and we’ve seen errors when capturing 3:2 images on the P4P.

Capturing 4:3 is not “better/worse for photogrammetry”, at least not that we know of. The P4P has a really solid sensor and lens, even at the extents.

Capturing 3:2 would give you about 10% fewer legs, resulting in fewer battery swaps. It would also provide slightly better views of the sides of buildings, but only in the direction orthogonal to the direction of drone travel. It would result in slightly poorer orthomosaics as you would collect fewer nadir pixel views overall.

Capturing 4:3 results in more images, but approximately the same total MB of data - because the overlap results in an equal number of pixel views of every point on the ground (~9 on default settings of 75/65).

Summary:

  • We set 4:3 for historic reliability of capture, as that is our number 1 priority, but recent DJI SDK updates may have resolved the original issues.
  • 4:3 is not significantly worse/better for photogrammetry, but probably slightly better for orthomosaics.
  • 4:3 results in more images, but ~equal MB of data to upload.
  • In 2019 we may try allowing 3:2 capture again, and monitor capture reliability. Overall the battery savings (10%) are worth it assuming reliability is no longer affected.

#24

Thank you @Jamespipe
That makes a lot more sense!

@MichaelL Guess I wasn’t too far out in left field after all!


#25

Thanks @Jamespipe. Good one @Dave.


#26

It was Drone Base client that flagged it. They accepted the 4:3 sent feedback that the pictures we’re suboptimal because the pro shoots 3:2. 4:3 is fine if all you have is a P4 or earlier. We are looking into other apps that shoots 3:2. For now they’ll accept 4:3

Thanks


#27

Correct, I just have never seen that message in DroneBase - only that they require 4:3. I haven’t flown for them in about a month so maybe something new? Or have you gotten that for a while? Let us know what you find. I was hoping to test today, but it is ugly in Austin!


#28

Unfortunately since you have done this I have no longer been able to use your APP. All our camera calibrations are done using 3:2
Why not leave it up to the user who can set it up in GO. You could still recommend using 4:3 for your processing.
Its a shame you have done this because I love your App but can not use it now


#29

When was the last time you are able to use it? For my knowledge it has never officially supported 3:2 and if it did it was not for very long. How does your camera calibration rely upon the aspect ratio of the image taken? I’m just curious, I’ve never heard of that type of calibration.


#30

OK We originally used the M100 with no issues, then upgraded our Platform to M200 were I saw the issue and discontinued using the APP. We have now created a PPK module for the P4P which has the same issue.

So maybe it has always been there except the M100.

I always thought it was suppose to keep the camera settings you set up in GO.

I used to always recommend Drone Deploy to all our clients but can no longer do so.

The issue is we use a camera calibration done with 3:2 so therefor need to use 3:2

Our system actually plugs into the camera to measure the exact time the shutter is opened and a PPK GPS to achieve the best results.

I thought it was a bug until I read that post

I always believe the user/pilot should make the decisions with maybe a warning why he should not use the setting.

In this case a warning would be adequate to allow the user to make an informed decision

Anyway maybe let the relevant people know and see what decision they make

Thanks very much for the reply

Your support has always been excellent another reason why I like Drone Deploy

Regards,

Alan Bullock

B.App.Sc.Surv

Survey Support

Position Partners Pty Ltd

(ACN 130 367 065)


#31

The shutter/gps timing alignment has nothing to do with the aspect ratio. I have only been using DD since April 2017 so may not be privy to 3:2 compatibility, but it has always forced 4:3 from what I know. Are you getting bad results? I just purchased an Emlid Reach M+ for my P4P to compliment my Reach RS+ GPS receivers and am definitely interested in hiccups. I’ll start testing later this week or next and will definitely report some results. Un(calibrated) camera…


#32

Each individual camera must be calibrated for the highest accuracy ppk equipped systems. Notice even DJI is saying that the P4 RTK units are individually calibrated. The processors (incl. Drone Deploy) do a “pseudo- calibration” based upon image sets but is not the same as an actual lens calibration.

Alan is saying that if the lens is calibrated to the 3:2 aspect ratio, it is not necessarily calibrated if using a crop (4:3) instead. Make sense?

To ensure each Phantom 4 RTK offers unparalleled accuracy, every single camera lens goes through a rigorous calibration process where radial and tangential lens distortions are measured. The distortion parameters gathered are saved into each image’s metadata, letting post-processing software adjust uniquely for every user.

https://www.dji.com/phantom-4-rtk?pbc=IN4xQqec?pm=link


#33

Until someone can show me on paper a dramatic increase in accuracy then no, I don’t believe you. Real lens calibrations are for DSLR cameras only. If a Phantom 4 Pro camera isn’t focus optimally the you need to replace it. The checkerboard method is a joke as it was not meant for mapping. Everything more that 100ft or so is infinity.
As for 3:2 vs 4:3 a focal difference between the two is virtually nothing considering you will be at infinity 90% of the time.
The only real value that anyone using RTK/PPK on a Phantom 4 Pro should pay much attention to is the timing. This is already taken care of on the P4RTK, so it is what it is. Additionally, if you think you are not going to need GCPs with the P4RTK then you are going to be sadly mistaken.
Show me the paper!


#34

It’s not about focusing. It’s about lens distortion. Every single lens is slightly different. Look around a little. You will find that every precision system requires individual lens calibration.

For example, Geo Cue calibrates the cameras on the DJI drones they sell with their Loki system, or you can send them your drone and they will calibrate it for you. The process is not that difficult and you can do it if you have Photoscan or other software. But it is necessary to achieve the finest results, believe it or not!

not the devil’s advocate; just the bearer of truth :wink:


#35

It’s still done with images and totally focus dependent. Show me the paper with numbers. Marketing fools allot of people. If you need that accuracy you better move beyond a Phantom or Mavic.


#36

You are certainly stubborn with things you haven’t learned yet. Skepticism is fine. But denial, not so much. Good luck!


#37

Dave,
I agree with you that lens calibration is essential for highly accurate results. The angle, and thus distance, to off axis points is strongly impacted by lens distortion. Thus is true both when the object is close and when it is far away (at infinity focus). That being said, I do not know the typical lens distortion numbers, for say, the popular P4P, and thus how much error they introduce into the results. What is true is that DJI felt they were large enough for the P4P RTK version that they went to the trouble of providing lens calibration data for every P4P RTK that they make.

Please don’t say: Good luck! We are all on a journey and seek a harmonious destination where truth and good will prevail.

Regards,
Terry.


#38

Has anyone on this forum actually done any testing on a stock P4P vs a calibrated P4P on the same site and same conditions both with and without GCPs? @Jamespipe, how about your team?
@dave, I apologize if you don’t like my point of view. I didn’t intend to offend you, but to express a point of view. I know all about the calibrations that DJI does to the RTK and I have actually had the discussion with other manufacturers that tried it and did not see substantial benefits. I am going to stand to my opinion that until I see practical evidence that calibrating the camera on the P4P makes a positive difference in the photogrammetry output that is field verified then the process is null and void to me.


#39

Michael, I’m not offended in any way. And you are of course welcome to your point of view, as are we all. I don’t have any comparison data with and without calibration to share with you, if I come across any I will post it. I suspect maybe someone like Geo Cue may have the data you seek.

Keep in mind that photogrammetry using SFM has been around a lot longer than uas and a calibrated lens (on very expensive equipment I might add) has been part of the recipe. I’m not sure why you would come to the conclusion that now it is not something that adds to accuracy.

@Jamespipe can contribute to what DD does under the hood to help in this regard. I suspect that currently, DD doesn’t have a way to input or take advantage of a calibrated lens anyway. But that may change if they want to take advantage of the calibration on the p4 rtk’s lenses which are supposed to hold the corrections in the exif data.

Terry doesn’t like “good luck”, so how about Cheers!


#40

Cheers for sure! My skepticism is because I know the limitations of the P4P camera and have been dealing with photography my whole life. From what I understand DroneDeploy has distortion built into their algorithms and such a minor adjustment to the camera, in the order of 0.1-0.2 mm/pixel at the heights that we fly just do not add up to much improvement in my opinion with the current tolerances. This is why we use GCPs. As I stated before, anyone who thinks we will not use GCPs with RTK/PPK is fooling themselves, you just won’t need near as many. If I use 12-15 on 50 acres I expect to drop that down to 4-6 depending on the shape of the site. The only UAV scenario that I think qualifies to dismiss GCPs is Lidar with RTK/PPK. I am putting together a P4P now with Emlid RTK (which can be taken off) so maybe I will get that camera calibrated if noone else has verifiable data. When I say verifiable I mean taking the point cloud, generating a TIN file and staking that out in the field. My RMSE values are nearly non-existent and my stakeout tests have been proving +/- 2-3 inches. In dirt no less.

*edit - All that said, I will say that it is a big deal on larger lenses and DSLRs and wider angle lenses. My Father would get his gear calibrated every 6 months and you could easily see the difference in the last picture versus the new calibration.


#41

Sorry for the brief response- on my phone:

P4RTK has three massive wins for photogrammetry bs P4Pro (&v2)

  1. absolute positioning of each photo in the realm of 2cm XYZ in RTK mode
  2. accurate pose (direction) of the camera in good conditions.
  3. factory calibrated camera dewarp parameters

We can reconstruct the 3D model
A) using these things as gospel truth
B) using these as a starting point for the optimisation
C) some statistical balance of these being “mostly true” but assuming there is still error introduced afterwards - lens smudges, knocks, vibration etc.
D) ignoring these entirely and using the GPS only at the end for scale.
E) ignoring everything and using GCPs for scale

We have been testing the P4RTK using all the above.

Bottom line is our photogrammetry engine does an excellent job if estimating lens intrinsics, and is robust to intrinsic and extrinsic errors. Flying at <200ft, you should expect accuracy of <2cm XY and <5cm Z on check point positions.

Using PPK and RTK correctly, you can think of each camera position a bit like a GCP (a known true position in the reconstruction). Thus you have effectively 100s per map. Check points or ground reference measurements should be used to verify/prove your map accuracy though, as then you are utilising two isolated systems to confirm your results. Measure twice, cut once.