Minimal Solvers

The purpose of the minimal solvers is to efficiently solve polynomial systems that arise in Computer Vision.

Minimal Solvers are cool! Minimal Solvers are cool!

Why it is Important

  • Polynomial systems arise various Computer Vision tasks, such as 3D reconstruction, Visual Localization, SLAM, and Camera Calibration.
  • Often, the input data is comtaminated with noise and wrong values, so we need a robust solution.
  • To achieve robustness, minimal solvers must efficiently process many samples, which demands speed and efficiency.
  • The pursuit of minimal solvers embodies an intriguing approach to these complex problems.
  • Minimal solvers are cool. ☺

Minimal Solvers are cool! Minimal Solvers are cool!

Key Problems

  1. Relative Pose Estimation: Finding the pose between two or more cameras.

  2. Absolute Pose Estimation: Finding the pose between a camera and a point-cloud.

  3. Applications: Minimal solvers are used in many computer vision tasks, such as:

    • 3D reconstruction
    • Visual Localization
    • SLAM (Simultaneous Localization and Mapping)
    • Camera (Auto-)calibration

Conclusion

Minimal solvers are a very cool approach to solve polynomial systems that arise in Computer Vision. They are present in various computer vision pipelines.

Minimal Solvers are cool! Minimal Solvers are cool!

Publications

  • Semicalibrated Relative Pose from an Affine Correspondence and Monodepth (ECCV 2024) [Paper]
  • Efficient Solution of Point-Line Absolute Pose (CVPR 2024) [Paper]
  • Handbook on Leveraging Lines for Two-View Relative Pose Estimation (3DV 2024) [Paper]
  • Vanishing Point Estimation in Uncalibrated Images with Prior Gravity Direction (ICCV 2023) [Project page]
  • Four-view geometry with unknown radial distortion (CVPR 2023) [Paper]
  • Learning to Solve Hard Minimal Problems (CVPR 2022) [Paper]