Reproduce “The Double Sphere Camera Model” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Vladyslav Usenko, Nikolaus Demmel, Daniel Cremers
The Double Sphere camera model offers a compact way to represent fisheye imagery with fields of view up to about 195°. Across six camera/lens configurations, its six-parameter formulation achieves reprojection errors within 1% of the best-performing eight-parameter Kannala–Brandt model while providing closed-form unprojection and substantially lower computational cost.
Key insights
- Adding a second sphere gives the model considerably more flexibility than the Unified Camera Model without the higher-order polynomial structure of Kannala–Brandt models.
- The model preserves nearly polynomial-level fitting accuracy while keeping inverse projection explicit, making it suitable for repeated computations in real-time estimation pipelines.
- A reparameterized UCM using a bounded alpha parameter reduces intrinsic-parameter coupling and produces more stable estimates across repeated calibration sequences.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.