Agisoft Metashape

Agisoft Metashape => General => Topic started by: geomaticist on September 22, 2021, 09:45:33 PM

Post by: geomaticist on September 22, 2021, 09:45:33 PM
I am trying to pin down a best practice or some good answers to a question about 'consumer grade camera' and the phantom 4 rtk camera in typical aerial mapping and processing in metashape. First of all, is the P4RTK camera considered consumer grade? I would assume it is.

Second question is about what camera calibration coefficients are appropriate to model. f, k1, k2, k3, cx cy, p1, p2 of course are.

Guides such as the USGS 2020 paper say b1, b2, and k4 are no appropriate to model for most consumer grade cameras. So .. what about the P4RTK one?

obviously you can just model it anyways. The real issue is does it make any noticeable difference to compute k4, b1, b2 on this camera? Will it just always get an unstable solution and will that adversely affect final product precisions? Does anyone actually know or have checked.

In context i am typically ppk processing P4RTk gnss positions with Trimble Business Center. I just use a script to insert events into the rinex file. Then I take the positions and adjust each by the yaw,pitch, roll and offsets to camera (which for P4RTK is very easy since it is in the .mrk file anyways and when I hand compute I get same answer.) So I then load those vey good positions into the Metashape.

Works great. There is nothing more I can do there. The rest is in fine tuning the process in Metashape. Which I have down pretty well. But the one thing I cannot get a good answer on is if k4, b1, and b2 should even be included in the optimization.  And is it only theory or has anyone really seen a difference skipping them or adding them in anyways. i notice the values for k4, b1, b2 vary a lot between different missions using the same drone. but the other values tend to be somewhat similar (the adjusted values) So I am suspicious that computing skew and k4 is FAKE.


Ok I might as well ask couple other questions.

Anyone doing a slight offset from full nadir, like 95? Just wondering if that is still a thing. I have not heard of it in a couple years now.

Finally, how many of you are consistently able to get sub pixel final precisions (except for tiny pixels of course, how would you even know.) Please dont claim this unless you KNOW for sure. By checking gcp outside of metashape.

I am successfully getting sub pixel final precisions both in horiztonal and vertical when flying at 400ft down to 200ft. And only use ppk corrected 'camera' reference positions and NO INPUT GROUND CONTROL during processing. However ... keep in mind I will adjust bias out after the products leave metashape. I am a land surveyor and can get high quality control points.

Hopefully some of the knowledgeable users will reply.

If this was already answered in depth I apologize as I could not find it.
Post by: MaciekK on September 23, 2021, 12:12:13 AM
I have completed dozens of P4RTK missions, and my experience shows that you should not use K4. I tested various variants and I am inclined to say that where you do not have to use K3 also. The K2, K3 and K4 coefficients are strongly correlated. The P4RTK camera does not give 100% repeatability - probably high depends on the temperature humidity ... So I tested how much I can reduce the K-factor. The best results were achieved using only K2 and very precise control points - a medium-sized mission about 50ha, flight at an altitude of 70m. Now I fly P1 and also most of the missions are only K2
Post by: geomaticist on September 23, 2021, 12:25:34 AM
MaciekK are you saying you do not select k4, and even ONLY select k2? Obviously we are talking about each dataset computing distortion coefficients per mission and not holding to a predetermined calibration (as we all know the environmental conditions are changing the answer too much.)

I just want to confirm you are NOT talking about holding a fixed value for each mission, but rather only solving for k2 uniquely each mission. Is that right?

by the way when a person is getting sub pixel precisions it can get tricky to know if any setting tweaks even matter at that point.

Thank you for your reply.
Post by: geomaticist on September 23, 2021, 12:31:14 AM
MaciekK I also have the ability to switch to the p1 and m300. But have wondered to what big benefit. For a few bucks I can deploy P4RTK to a survey crew who uses it for a couple years before it wears out. And get one pixel or less precision fairly easily. I so far am not sure what extra value will be gained by deploying to them a setup (P1) that costs 5 times as much. if it is not going to produce better results anyways.

But it very well might be time to move on to better equipment and cameras. Are you finding it to be worth it for the better equipment? You probably use it yourself. I dont, I train crews and send them out and they only do photo missions as one of many things. I am thinking maybe just keep one or two p1 around for when I need the best possible results. I dont know. if they make a mistake it cost way more with P1. ha ha.

Rest of you ... I am still looking for some feed back to original post!
Post by: MaciekK on September 23, 2021, 01:43:24 AM
P1 is primarily about greater efficiency. But also quality. Don't get me wrong, the accuracy is similar, but the stability of the solution and the geometry of the photos is a class higher. For example, during the P4RTK mission I had a point multiplier of 2.5 - 3, on P1 I have 4.5 - 5.5, it changes a lot. The camera is less susceptible to the environment, the calibration parameters are very repeatable, etc. It also has its drawbacks - size, weight, but I prefer m300 and P1. As for your earlier questions - I confirm for P1 I only use K2, for P4RTK usually K3 but sometimes only K2. B1 and B2 always off in the case of P4RTK. For P1 and large areas, e.g. 600 ha, the inclusion of B1 and B2 gave better results.
Post by: MaciekK on September 23, 2021, 09:48:19 AM
maybe I put it wrong in the previous post - using the B1, B2 coefficients improved the alignment at higher flight altitude for mapping larger areas