About

Background

  • Multiface dataset is captured by dozens of cameras at many frames.
  • Calibration of the cameras in the dome is time-consuming.
  • Extrinsics don’t change over a short period (e.g. 1 day ~ 1 week).
  • Intrinsics fluctuate frequently (e.g. for each capture).

  • Calibration based on regular patterns needs to be re-performed before each capture and is computationally expensive.

Overview

  • Build an auto-calibration system and an efficient Structure from Motion pipeline to obtain the intrinsic camera parameters.
  • Enhance parts of SfM pipeline using different learning-based feature extraction/matching algorithms.
  • Apply featuremetric refinement to detect keypoints and triangulated points to improve accuracy.