I am trying to reconstruct a 3D building model from 2D video frames using an open-source Structure-From-Motion(SFM) software (Bundler). The final reconstructed 3D model will be in Cartesian coordinates but ultimately it needs to be geo-registered so that I can overlay it on a virtual globe or in 3D analyst.
As far as I know, the typical SFM approach solves for relative camera poses and the 3D points. However, since I already have GPS/INS metadata for each video frame, is it possible to use those as an initial estimate for the extrinsic camera parameters (camera pose) to input into bundler, and get my 3D point cloud directly in the same world coordinate system directly?
I know a common approach (correct me if I am wrong) after the SFM process is to manually try to fit the model or find a rigid transformation to the world coordinate system, by using some marked ground control points on a reference imagery, but I just want to know whether this geo-registration can be done automatically instead by specifying known rough initial camera poses as input to Bundler (or any other similar sFm software) for subsequent refinement through bundle adjustment.