Why my model is curved?

[ronativ]

Hi all,

I am using 5 cameras, each shooting at the same time, to photo a topographic model. The cameras are located on a designed steel structure, just 1 meter above the model. I am using a scaled cartoon box with known dimensions to set a scale to my model (see attached photo). After loading the photos and aligning them, I place several markers and denote their relative coordinates in the reference pane.

After generating a dense cloud, there are two things in the resultant model that bother me:

1. The model is curved, i.e., it is inclined (see DEM photo) even though I used a rotation script to align it.
2. Some parts of the model seem to have "split" - see the attached photo of dense cloud where the red sheet looks as if it is split into two parts (in reality it's one sheet).

Can anyone share insights about these issues? why they are happening?

Thanks!
Ron

[chrisd]

Hi Ron,

Can you upload the pictures?
Are the cameras all parallel?

[ronativ]

Hi chrisd,

I uploaded 3 photos in the original post.

I attach a picture of the model set up. Please see the cameras on the top of it.

Thanks,
Ron

[chrisd]

Hi Ron,

I saw the pics in your post, I meant the actual images from the cameras you are using to build the model.

[ronativ]

Thanks again for your answer.

Here are the 5 pictures.

Best,
Ron

[chrisd]

Are these the full resolution images, not cropped or reduced?

[ronativ]

The images are not cropped, but the sized are reduced.

Here is a link to the full-resolution pictures:

https://drive.google.com/drive/folders/17ybWc8ZzdlYkvZfGpg6sFr2ZE2nmGi0V?usp=sharing

[Mak11]

Unfortunately only 5 Photos of this (low) quality isn't going to cut it for Metashape (or any other photogrammetry software). You would need more photos with more overlap and of higher quality.

Mak

[chrisd]

... and you would get best results from using RAW image format, not JPG. You would need to look at what cameras are available within your budget that can save this format.

CHDK can be used with some of the Canon point and shoot to get RAW images from the camera so you wouldn't necessarily need to spend $ on a pricier DSLR model.

[ronativ]

Hi,

Actually, we already made some nice results in previous experiments. Even in the last experiment (see attached photos to the original message) where the model output is curved, the resolution is quite good (around 2.5 million points in the cloud, and a 0.5 mm pixel resolution in DEM).

The problem not seems to be inability to generate a model, but rather something else.

Ron

[chrisd]

If the cameras were all mounted as in my markup of your image, the ambiguity (180 deg rotation) of the cameras in relation to the subject would be eliminated. I think some of the warping could be attributed to this issue.

[chrisd]

I threw out the 2 images that had the 180 deg rotation and there is still significant warping, so the lens specs may not be interpreted correctly from the images.
There is no EXIF data recorded with the image for comparison.

You probably will need to know the spec on the camera lens and use the AGIsoft Lens application (Tools menu) to do the calibration. I don't have any experience with calibrating the lens by this method so perhaps someone else could jump in here.

I understand these cameras were relatively low cost as they came with a video surveillance system, however, I think you would have a greater opportunity for a quality result with a different camera system.