I recently was in Japan and went to Shirakawago. I am attempting to put together some photogrammetry models from photos I took. The camera was a Lumix GH5 with a Canon 28-100 L lens. In this example, I took roughly 850 images. I’ve now tried processing them 3 times and each time I cannot get them to align.
I have added control points but I still cannot get any alignment to happen. I’ve tried using RAW files, JPG versions that were processed to enhance their clarity as well as a combination of both. None of them will align. Instead I end up with the example image I’m attaching to this.
What can I do to fix this? I can’t return to Japan, so the photos I have are all I have. In the past, I haven’t run into this issue, so it seems really strange to me.
Here is another example of the same set of images. It’s actually closer to aligning but it’s still doing strange things - some of the images show the walls tilted back at a strange angle, some of them have a section of the roof in the sky.
I also should mention, I tried the standard troubleshooting of selecting all the cameras, disabling them, then selecting a small subset of them and aligning only those - then later tried to merge those components. It still did not work. Here’s another example from one of my attempts.
Hi Paul_Griswold,
what are your alignment settings? Are those images also georeferenced?
I see one issue there - the images captured from one spot, which is not ideal.
How many images are aligned after alignment? How many components is created?
I see that you are using control points. Are those placed over all component which you want to merge?
They were taken with a GH5, so there is no GPS data. They aren’t taken from one spot, I walked around the entire building while taking the photos. There are roughly 850 taken from around the house, including from a hill near the house to get shots of the roof.
The only thing I couldn’t really get is a shot from above the roof on the opposite side - there was a hill on one side that let me get above it, but not on the other side. Otherwise I was able to take pictures from around it entirely.
I’ve played with the aligment settings and tried things like turning rematch on and off. I’ve tried adding upwards of 15 control points, but even when I start to get it close, when I try to align all the images it just explodes.
I didn’t mean that all images were taken from one spot, but according to the alignment results it seems like some were:
It is not ideal for photogrammetry.
Is the 1920x1440 the real image resolution?
Otherwise the captures look good to me. Is it possible to contact our support and share your data with us? We can check what else could be wrong there.
The images are 5184 x 3888, shot with a GH5 and a Canon 28-105 L lens and a speedbooster adapter.
I have 3 locations in Shirakawa that I shot and I’m having the same issue with all of them. Walls end up curved and bent and sections of the buildings refuse to line up regardless of whether or not I add control points.
I sent in a support request. But yes, because of the location and the large number of tourists, I would have to stop and take photos from one spot to both get all sections of the building covered, and also to avoid getting people in the shots. Should I try to combine those into a single large image?
You may already know most of this but I recommend the Reality Capture photogrammetry course which might help you identify areas for improvement in the future
I’d be interested to see the full dataset and see if its possible to produce something that is useable for you
