A way to apply SfM from frames belonging to Multi-camera with shared FoV?

[Sentient07]

Hi,

I am trying to understand whether it is possible to run SfM for a scene involving moving targets, but static multi-view cameras. (Initial pose of the camera wouldn't change with time). I don't plan to use it for real-time but as a post-processing step. How can I pass multiple-images from the same camera, which are temporally aligned? (in other words, a video stream from multiple cameras).

Furthermore, once the a) Focal length of the camera, b) Relative pose of the cameras are obtained, I would like to use this information to sequentially compute feature, map them and obtain depth maps. I am only interested in the depth maps and hence would prefer to not save the features, matches, etc. locally. Is there a stand-alone script that does it?

Thank you in advance for your time and response.

Regards,

[YanNoun]

Hi @Sentient07 ,

OpenSfM doesn't support such setups for moving targets.

If you already have image poses and cameras intrinsics can just construct a reconstruction.json file and start the process from the undistort_images step.

Yann