Calibrating a lens Fisheye or Perspective

We use an 8mm Sigma lens on a 5DMkiii and I think we have been having some problems with the calibration lately. A couple things could be the problem, but I’d like to run this by the community to get a feel for if I’m doing this right.

Problem : The biggest issue has been with alignment and getting a lot more artifacts in the reconstruction and a lot less calibration between images even though we are having quite a bit more overlap than with our other lens a 17-40MM.

From a set of 300 photos we’ll have about 60-70% calibration with the fisheye lens. This usually comes from an interior pass and exterior pass being combined. But with a similar set of photos with the 17-40mm lens we have almost 95-100%.

The last problem we’re having is when we combine the image data sets, that the fisheye photos are often misplaced after combining, and require a lot of hammer and mallet to get positioned. We sometimes double the amount of MTPs from the original project just to get the remaining photos to align properly.

Thanks in advance.

  1. I think I have the camera model set correctly, but I’m concerned that I’m using the wrong intial settings. Here’s my settings I am currently using. The image stats I’m using are 16bit TIFF files at 5760x3840.
  • Fisheye Lens
  • Shutter Model: Global / Fast Readout
  • Sensor Width: 36
  • Sensor Height: 24
  • Pixel Size: 6.25
  • Principal Point x: 18
  • Principal Point y: 12
  • Polynomial Coeficients: 0-1-x-x-x
  • Afine Transofmations: C-1820, D-0, E-0, F-1820
  1. I am adding a vignette in post processing to make sure there’s no extra data from the barrel of the lens see the images below. These images are both the same size. Does adding a vignette have any effect on the reconstruction? Because it cuts out about 1/2 of the size of the image. I did my best in adding the vignette right at the edge of the circular image.

  1. We are making sure to have plenty of overlap between images, and we are getting, I think about 80%-90%, but would taking more images be better for reconstruction?

Lastly. We have been considering just using a wide angle lens at about 14mm and just taking more images because we’ve been having such problems with the fish eye.

 

Thanks!

You stated that you are having problems, but did not list any of them.

I would not change the photos. Pix4D has a annotate function that is easier to use. The “Global Mask” needs only to be applied to one photo and this will remove the inside of the lens for all the photos.

Read this:

https://support.pix4d.com/hc/en-us/articles/202560549-How-to-Annotate-Images-in-the-rayCloud

1 Like

Hi Tommy. I updated my post about the problems I’m experiencing.

 

Thanks for the note. I do annotations, but worry about using the global masks when combining projects from multiple sources. Do you think this would be a problem?

I agree with Tommy, I would not do post editing on the images before importing them in Pix4D software. Have you tried using the normal images? If the camera model is good, it should not create too much issues to have these borders. 

As far as I can tell, the camera model looks ok, but if you want to be sure and make a new camera model, you can follow the instructions in this article: https://support.pix4d.com/hc/en-us/articles/202557009-How-to-calibrate-a-Fisheye-Lens-Camera

As for the Global Mask, it applies to outputs from step 2. Point Cloud and Mesh. However, if you merge projects, you would typically only run step 1. Initial Processing and then process step 2 in the merged project. Hence, I would expect that the Global Mask is applied to the entire project. 

For the image acquisition and overlap question, it would be easier to see a project. Maybe the image acquisition plan can be improved. Comparing one successful (95%-100% calibrated) with a less successful one (about 65% calibrated images) would be ideal to better understand what is happening.