Support Website Contact Support Blog

Calibrating uncalibrated cameras

Hi all

So I’ve read through everything I could find…even the step by step Pix4D official tutorial and can’t get cameras to calibrate.  I gave one 5 manual tie points with at least 4 tiep oints to each (three had between 10 and 20) and the CALIBRATE button is still greyed out.  This is one thing I really will need to work if I am to invest in this software.  Could someone maybe tell me what I am doing wrong because I am guessing it is user error but I have tried about half a dozen times now with different cameras and projects and just can’t seem to work it out.   Maybe if there is a video tutorial that I can’t find a url would be good…or any advice really.





I’m having the same problem but, I believe the issue comes that i don’t really get calibrated camera images to use as tie points.

I have one project where 60% of the cameras as uncalibrated.


Is there anyway to force loading other images from calibrated cameras which i could use for the tie points?


Manually calibrating images is a time-consuming task. Actually when many images are not calibrated, we suspect a problem with the following:

  • Poor quality image content (blurry images, no texture, etc.)
  • Insufficient image overlap
  • Incorrect initial internal camera model parameters
  • Inaccurate initial image position coordinates or incorrect coordinate system
  • Inaccurate ground control point coordinates or incorrect coordinate system

Solutions that we recommend are to:

  • Verify the image quality. For more information: 203756115.
  • Verify that there is sufficient image overlap. For more information: 203756125.
  • Verify the initial internal camera model parameters. For more information: 203032259.
  • Verify the initial image position coordinates or coordinate system. For more information: 203755985.
  • Verify the ground control point coordinate or coordinate system. For more information: 203032299.

You can learn more about interpreting your project’s Quality Report: 202558689

Best regards,

1 Like

Any resolution on this? I have images collected as r-jpegs with a Zenmuse XTR and I cannot get the Pix4D software to generate any ATPs? I tried the MCP and it did not work, I couldn’t apply the MCP…

The how-to documentation for how to process thermal imagery does not address any of these issues really. It seems that if you are flying a DJI drone with a DJI/FLIR camera and using the Pix4dCapture app and desktop software… it should all communicate and function properly. unfortunately, that is rarely the case. 

Hi Francis,

I trust that you are already familiar with the following, but I want to take the opportunity to mention a handful of leading reasons why Pix4D’s image processing engine may produce uncalibrated images, which include but are not limited to:

  • Blurry or severely distorted image content. It is relatively easy to identify when an image is blurry but consider that linear rolling shutter distortion is not visible to the naked eye in a typical drone flight.
  • Insufficient image overlap for the application. The recommended minimum amount of image overlap for any Pix4D project is 75% front and 60% side image overlap, but the minimum recommended amount of image overlap increases to as high as 90% for a low-resolution sensor like the DJI XT.
  • Inconsistent or dynamic image content. Low-resolution images are especially vulnerable to image content that is not consistent between overlapping images. Examples of dynamic image content include vegetation on a windy day or the direct reflection of the sun on a photovoltaic array.
  • Homogeneous image content. Pix4D’s image processing engine, like, practically speaking, all photogrammetric solutions, relies on unique textures in the image content to identify a common feature in overlapping images. Thankfully much of the world provides us with some texture to work with, but there are project areas that yield little to no usable texture because of the nature of the object.

My understanding is that your latest project results yielded no Automatic Tie Points after processing Step 1. If that is the case, I do not recommend that you incorporate Manual Tie Points as your first plan of attack. Manually calibrating images with Manual Tie Points is a painstaking process that is only possible if you already have a set of calibrated images to tie the uncalibrated images to.

Instead, I recommend that you tailor your project’s processing options to achieve a larger set of Automatic Tie Points and ultimately a robust image block of calibrated images. Considering the intricacies of processing low-resolution thermal images please contact us directly so that we can continue discussing your project’s specific needs.

Thank you for your understanding.

Hi Andrew, 

Thanks for addressing this, I currently have two thermal projects that have open support tickets. My rant was more aimed at the lack of communication from the drone, the sensor, and the app. It seems like Pix4D is the most user-friendly and most compatible app and software. Following all the collection and processing best-practices should yield good, if not perfect, results every time. It would be great if Pix4D could do a thermal camera specific training!


Hey Francis, I am glad if Andrew’s comments could help you :)
And I invite you to share your experience regarding the image acquisition in the dedicated forum topic here. You never know, some other users might be able to give you some tips and tricks.

I have also taken note of your wish for a training course about capturing and processing a thermal dataset. At the moment, there are no workshops planned with a focus on thermal imagery. However, we are looking to expand our offer and your feedback is very valuable. Thank you for sharing it!

@Francis , there is now a form online that you can fill with your request for a workshop. Please navigate to the bottom of this page. Looking forward to hearing from you!

Hello, I have a problem with my dataset. On pix4dmapper desktop I can generate mosaic, but it has some Grey areas, but when trying to do the same on pix4d fields, all what I get is camera calibration problem.

Hi Pedro,

I have answered to your question here:

Hi all,

I’m having many problems in calibrating a dataset of RGB images taken from DJI Phantom 4 Pro. It is the first time having this kind of strange problem: the images are good, without blurry or reflections, the overlap of the flight is very good, the height of the flight is the one I always choose, gps of the photos are OK, and I’m working with the same camera and drone in many projects (both before and after this issue) without any problem.
The images remain uncalibrated (70% calibrated best case) always in the centre of the project, while they are calibrated on the edge. I used all kinds of calibration method, any optimization of external parameter (all prior alternative was the best) and any kind of image scale (and I also did all the combinations).
Tell me if I can (or should) put some screenshots or the quality report here. Thank you.


The Quality Report would be very helpful! Could you post it here so we can review your project? :slight_smile:


alternative_RGB_new_report.pdf (1.1 MB) standard_RGB_report.pdf (1.2 MB)
These are 2 reports for standard and alternative calibration. Thank you for you help.


Thanks for the data. It seems like you are mapping solar panels and because of their highly reflective texture the software can’t extract many kepoints. So I would try the following:

Let me know if you obtain better results.


Unfortunately it was not successful. Attached report.

Carinola_RGB_new_report.pdf (1000 KB)


Thank you for your fast reply.

Definitely, so far, your first attempt is the best.

What is only positive about my try is the median of points per image which increased significantly.

I’d like to obtain a similar result in your first version as we’re lacking some points there.

Did you try Alternative pipeline, All prior but Full scale? I expect it gives us more points and calibrated cameras.

The unconventional method would be to use the Free Flight or Terrestial. This strategy tries to match more surrounding images in the dataset and searches for keypoints using more images.
This is why it can result in a higher number of matches, but this does not necessarily mean a higher accuracy of the results.

If you’re eligible for personal support please send us a ticket. I’d be interested in processing your dataset and give you the best options.