An Interactive System for Set Reconstruction from Multiple Input Sources

Mikhail Smirnov, Eugene Vendrovsky, Joe Mancewicz, Jonathan Meier, Cyrus A. Wilson


Visual effects pipelines typically require accurate set and environment geometry. However, the type and quality of data obtained on set during production are typically outside the control of the post-production visual effects crew. We present a photogrammetry system which is flexible with respect to input data, and with respect to the specific needs of a shot. Geometry can be reconstructed from a set of input images, which can include a combination of perspective images and spherical images. Direct geometry inputs (e.g. LIDAR scans or Kinect depth maps) can also be integrated into the reconstruction. Alternatively, a pre-existing model can be refined to conform to input imagery, preserving mesh topology if desired, or adaptively remeshed according to local detail. This process can be guided interactively, to efficiently produce satisfactory geometry though the form and quality of input data may vary.