Efficient Recovery of Essential Matrix From Two Affine Correspondences.

Area of Science:

• Computer Vision
• Geometric Computer Vision

Background:

• Essential matrix estimation is crucial for 3D reconstruction and camera pose determination.
• Existing methods often require numerous point correspondences, impacting computational efficiency.

Purpose of the Study:

• To develop a novel method for essential matrix estimation using minimal affine correspondences.
• To enhance robustness and efficiency in computer vision applications.

Main Methods:

• Derivation of two novel linear constraints between the essential matrix and local affine transformations.
• Extension of Hartley's normalization technique to local affinities, considering intrinsic camera matrices.
• Generalization of constraints for arbitrary camera models beyond perspective cameras.

Main Results:

• The proposed method accurately estimates the essential matrix with only two affine correspondences.
• It demonstrates applicability to over-determined cases and robust estimators like RANSAC.
• Validation on synthetic and real datasets shows competitive performance against state-of-the-art methods.

Conclusions:

• The novel method provides an efficient and robust approach to essential matrix estimation.
• Its low point requirement makes it suitable for real-time applications and challenging datasets.
• The technique is extendable to various camera models and advanced applications like multi-motion fitting.