Multi-Camera Rig for scanning Caves

[johnkendall]

Hi All,

I'm experimenting again with a rigid, Multi-camera rig for scanning in flooded caves. The camera rig is 12 4K video cameras. I have created scripts to pull frames from each video, label them sequentially, and put them in a sub-folder.

What I'd like to be able to do is import all of the images, but constrain Metashape by feeding it the relative positions of all the cameras. Is there an easy way to do this?

I'm currently pulling the images in 100 per folder at a time, and incrementally aligning them, but the model is shrinking as it progresses (The tunnel size in the model is getting smaller, rather than staying approx the same size)

Thanks
John

[chrisd]

If the camerars are all parallel I suspect that the effect is because the camera positions are observed from multiple locations.
Ideally, the cameras would be facing approx. normal to the wall of the passage, rather than all facing down the length of the passage.

[Alexey Pasumansky]

Hello John,

You can follow a multi-camera system approach (described on pages 21-23 of Metashape Pro manual: https://www.agisoft.com/pdf/metashape-pro_1_6_en.pdf). This approach requires to have all the frames synchronized and be put to the sub-folder structure, either using each sub-folder for the camera or for the frame: i.e. you can have 12 sub-folders and add the frames from each camera to the certain folder, or to have hundreds of sub-folders with 12 images inside every one. Note that the order of the images in each folder is important.

Having the images added as a multi-camera system, you can enable Adjust Location option in the Camera Calibration dialog for all the Slave sensors, if you would like to let Metashape to estimate the relative shift between the cameras in the rig (by default only relative orientation will be estimated).

[johnkendall]

Hi Alexey,

I tried doing that, (12 folders, one from each camera), but unfortunately didn't align even slightly correctly.
I got OK results just pulling the images in normally and aligning them, but there seems to be an issue where scale errors seem to compound as it progresses.

Thanks
John