Support Website Contact Support Blog

Processing DJI Phantom 4 RTK datasets with Pix4D

Hi Guillermo,

Pix4Dmapper reads information from the image EXIF data. It reads both standard EXIF tags and Pix4D defined camera XMP tags.

In the XMP tags, you will find the orientation and accuracy-related information written in the parameters Xmp.Camera.Yaw, Xmp.Camera.Pitch, Xmp.Camera.Roll, Xmp.Camera.GPSXYAccuracy and Xmp.Camera.GPSZAccuracy: Specifications of xmp.camera tags.

To make sure your images contain such information, you can use this tool by dragging and dropping one sample image and looking for the necessary parameters: metapicz.

Cheers,
Teodora

Thanx for reply. I did what you said, and the images has the information. So, i dont understand why pix4d is not reading it

Hello again,

Indeed, the accuracy values are in the metadata. Have you also tried processing the project with the calibration method Accurate Geolocation and Orientation? I think the software should be able to pick up the accuracy of the geotags. More details can be found here: FAQ on the accurate geolocation pipeline.

Cheers,
Teodora

I cannot select that option

image

Hi,

This means that the images do not have accurate geotags. Do you have a sample of the image geolocation file? The values should indicate very accurate geolocation and orientation, as shown in the screenshot below:

Cheers,
Teodora

Hello,
I am having similar issues, where Pix4D is shifting the images irrespective of the high-accuracy PPK geolocaiton through PropellerAero.

We are working in an Australia co-ordinate system (GDA94) and an Australian Geoid grid (AHD09). The input file is already converted to the co-ordinate system and the geoid. I select the co-ordinate system we are using (GDA 94 MGA Zone 54), and currently select “Arbitrary” as the vertical geolocation.

I find that these settings are producing poor accuracy compared to the cloud processing through PropellerAero. The reason we are not cloud processing is that we want to be able to crop the data and edit the point cloud, but data accuracy takes precidence.

The data MUST be in the aforementioned coordinate systems to meet our clients requirements. What settings do you recommend in Pix4D to achieve the proper map accuracies from PPK?

Hi Daniel,

The processing workflow for RTK should be similar to PPK. In the end, you are dealing with image datasets that have highly accurate geotags, only their origin differs.

If you are getting some shifts, I would suggest first double-checking the PPK correction process and make sure the appropriate data was chosen:

  • the rover file (rinex data from the drone) - ensure there was no strong wind that induced vibrations to your drone.
  • the base station file (rinex data from a fixed base station) - this should be as close to your study area as possible.
  • broadcast ephemeris data (precise satellite navigation data) - this acounts for orbital errors and should be reliable.

The last resort would be to correct the shift by including GCPs in the project. RTK/PPK geotags in consumer drones is a relatively new technology and the workflows for using them have not yet been clearly defined in the industry. To what extent they can replace GCPs as a means for high accuracy in mapping is not thoroughly understood. In general, the best practices would still require the use of GCPs and checkpoints to meet proven absolute accuracy standards.

However, the benefit of RTK/PPK image geotags is that you can reduce the number of GCPs collected for a given project.

Hopefully this makes sense.

Best,
Teodora

Hi Teodora,
Thanks, could you please clarify a couple of points though:

  1. is there a way to import the co-ordinates and their accuracies, but not the camera orientation? So far, I can only find a way to import co-ordinates
  2. If I am importing the PPK data in a local co-ordinate system and a geoid, will that effect the accuracy of the processing?

Thank you

@Daniel_Ierodiaconou,

  1. In order to import image geolocation with the accuracy values, the orientation angles need to be determined in the file as well. If the orientation angles are written in the EXIF of the images you can create a project in Pix4Dmapper and use the To File… option to export them in a text file. Then include the exported omega, phi, kappa angles in file with PPK geotags and use the From File… to import everything.

  2. If your final goal is to deliver accurate results based on the already transformed values of image geotags then I would recommend changing the vertical coordinate system to Arbitrary in the image geolocation and outputs coordinate system menu. This way no transformations will be done during processing and results will be in the same reference frame as PPK geotags.

Let us know how it goes,

Hello
I’ve just processed images collected by a DJI P4RTK drone.
Indeed after the first step of the processing, I noticed the following problem: model divided into two blocks (see capture).
Have you ever had such a problem. I request your help to solve this problem.

Yours sincerely
Doc1.pdf (201.5 KB)

Hello @alkassoumoutari,

I would recommend you to follow the processing steps provided by Blaz at the beginning of this post and see if that helps to solve your issue.

Regards,

Good morning, dear @Kapil_Khanal .
All right, I’ll try to see if the problem will be solved that way.
Thank you very much!

Good morning, dear @Kapil_Khanal .
What seems a little blurry to me is that I had the same problem before when I used a non RTK drone. I ended up with two blocks almost perpendicular to each other.
Which exit for me dear community?

Thank you very much!

Hi @alkassoumoutari,

A block is a set of images that were calibrated together. Multiple blocks indicate that there were not enough matches between groups of images to provide global optimization. When a project contains multiple different blocks they may not be accurately georeferenced relative to each other. A project should ideally contain only 1 block.

Cause:

  1. The terrain is flat with homogeneous visual content such as agriculture fields or snow, sand, and water.
  1. The project area consists of forest and dense vegetation.

Additionally:

We recommend adding common Manual Tie Points between the blocks in all the mentioned cases above. For more information: How to add / import and mark manual tie points (MTPs) in the rayCloud:

Regards,

Hi@Kapil
Thank you very much !!

Good morning, dear Kapil

Now when is the problem with optimizing the camera settings?
Often when I process some projects, I end up with a camera settings optimization ratio in the order of 100/100 or more times, when all the images are calibrated.

So what’s the way out for this kind of problem my dear brother?

Yours sincerely!

Hi @alkassoumoutari,
These are very interesting questions, I will be glad to give some hints/help so that you can move on with your project.
However, you may also want to look into one of our paid training services. Our trainers can help develop your workflow and answer any questions you may have about complex topics related to processing.
Learn more about our workshops and personal training offering here.
If you have a valid license, you could also reach out to the personal support here. It would be easier to communicate and deal with your issue. Make sure to provide quality reports, few screenshots of the issue while reaching out to personal support. If there is anything useful, I would share the result on this page later.

Regards,

Hi thanks for your answer. I did the process as you mentioned but when I upload come check points to verify my process I have 1.5m of difference in height (z) between the the check points and the original measure.