Support Website Contact Support Blog

Seagrass stitching issues

I am trying to use the software to stitch seagrass beds. The images are georeferenced, very high overlap, I have measuring tapes and marker panels down. The software refuses to calibrate any of the images for a point cloud or ortho. For some reason it cannot match the images. I am uncertain what else to do, I can very clearly see matching points in each image and I can manually stitch the images into an ortho using PTGUI and then georeference and scale in QGIS, but it takes forever to complete. I would love if Pix4d could do it for me. 

Any help would be great!

Thanks

Cory

Hey Cory,

It is difficult to calibrate images that have complex geometries such as snow, bare soil and dense vegetation. Try Alternative calibration method and 0.5 image scale, you may have have more images calibrated. Alternatively if those methods don’t work change the median number of keypoints from automatic to manual. That value would be based on your quality report. If the number of automatic median number of keypoints is something like 40,000 I would put 25,000. 

1 Like

Hi Cory, 

Selim has some good tips, I’d suggest you try them. 

If this does not work, more information on your project could help understand the issue.

How much frontal and side overlap do you have between your images? What is the resolution of the images? How high did you fly?

If you share a few sample images or even the entire project with a link to a file sharing service (Google Drive, Dropbox,…) I can have a closer look and run the images on my computer.

Thanks for the help guys!

I probably should have mentioned that I shot the images underwater using 2 Gopros. There is lots of side and frontlap. I am loading the images on dropbox now and will post the link when finished. I will also try Selims ideas.

I did read the couple articles that Pix4d has up about stitching underwater, but I did not see any technical information about settings.

Thanks again for the help!!!

Cory

You’re welcome! Did you have the chance to try Selim’s tips?

Sorry it has taken so long to respond!

I have tried the tips but still no luck. I am trying slight modifications to his tips now as well. I wonder if I should break it into smaller projects and merge them at the end?

I know it can be quite difficult to stitch this type of imagery, I had hoped that laying down measuring and caution tapes in grid lines might help. Does not seem to have!

I am open to ideas for any types of markers that I could lay on the seagrass beds to help the software stitch.

Could you share the images? I could take a shot at it and try to see if I can figure something out

I am uploading them to dropbox now and will post the link when they are finished.

I am trying to figure out the best image acquisition process for this type of underwater imagery as well. This is only my 3rd try. 

Thanks!!

He is the dropbox link to the photos. I have tried about every option I can and nothing seems to work. Even around my numbered marker panels! 

https://www.dropbox.com/sh/kxj4ivvhwc8b1fr/AABCsy3y7dCeQd9u6NN4xrDia?dl=0

 

After looking at your images, i’m guessing there isn’t enough contrast/clarity for Pix4d to match pixels with, combined with slight motion of the sea grass, and changing light pasterns from the water surface and that’s a big ask of the software. Pix4d had a hard time with vegetation and water and you’re combining both. Hope you’re able to get some results!

1 Like

If I were you I would start with smaller datasets (100-200 images max). This should enable you to make tests faster and incrementally improve the image acquisition and reconstruction of seagrass areas. The project you shared has about 2600 images, which makes it much harder to investigate. 

In this project, the images were acquired in several long lines. I would rather do smaller zig zags, so that you cover more of the ground that is close to each other. Here is a rough drawing (not at scale) to illustrate what I mean. Below the current image acquisition and at the top the one I would test: 

This might help to find more matches between images, as you cover more of the same area closer in time. Also, it might be worth to use only one camera as a start and then add complexity when you do more testing. Keep us posted with your tests! 

Well I have tried everything so far and still no luck. My GPS is not super accurate so that has been proving to be a problem, the points are quite close so sometimes pix4d dumps the gps data. GoPros picture quality is not super awesome and he water clarity has been less than stellar lately.

Anyone have ideas? All I have left to try is: 

  1. Better camera: I would guess better resolution would make it easier 

  2. centimeter level GPS: more accurate georef. might make it easier as well??

 

Anyone know if Pix4D is working to get the software better at underwater ortho’s? Factoring out the flickering light would be awesome.