Localizing a camera in a point cloud

[Ashwin Kannan]

The basic problem that I am trying to solve is localizing a camera in a point cloud. Using Structure from Motion, we have built a 3D point cloud offline, and now with an image at test time, are trying to localize where the camera was located. Multiple research papers suggested using P3P in RANSAC to get rid of outliers and then to use PnP to get an estimate of where the camera is within the 3D environment.

Is there a more direct or better approach, specifically within theia that could be used to solve this? Any topics to read on that might be helpful?

Sincerely,

Ashwin

[Chris Sweeney]

Hi Ashwin,

You're right that this is the recommended approach in the literature. If you'd like to read more on this topic try a google scholar search for "image based localization." In particular, the work from Torsten Sattler have been the state of the art for localization.

Theia does not have a simple out of the box solution, but does indeed have the P3P+Ransac method (and a whole bunch of others) conveniently wrapped in //theia/sfm/estimators/ directory.

--
You received this message because you are subscribed to the Google Groups "Theia Vision Library" group.
To unsubscribe from this group and stop receiving emails from it, send an email to theia-vision-lib...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Anirudh Koul]

Thank you Chris for the helpful information! Appreciate the reply. 

Anirudh Koul

--

Cheers!

Anirudh Koul

Schedule a meeting: https://calendly.com/koul