Line Matching in Wide-Baseline Stereo: A Top-Down Approach

Area of Science:

• Computer Vision
• Geometric Computer Vision

Background:

• Line matching is crucial for various computer vision tasks.
• Existing methods often struggle with incomplete line observations and wide-baseline images.

Purpose of the Study:

• To introduce a novel algorithm for robust line matching across images.
• To leverage epipolar geometry and coplanarity constraints for improved line correspondence.

Main Methods:

• Utilizing epipolar geometry to constrain coplanarity conditions between line pairs.
• Detecting candidate line-pair correspondences by addressing incomplete line observations.
• Verifying matches using local homography transforms derived from interest points.
• Processing a line affinity matrix to obtain final matching lines.

Main Results:

• The proposed algorithm effectively handles incomplete line observations and non-matching endpoints.
• Demonstrated superior performance compared to state-of-the-art methods, particularly for wide-baseline images.
• Achieved robust line matching without relying on appearance models.

Conclusions:

• The novel approach offers a significant advancement in line matching accuracy and robustness.
• The method's reliance on geometric constraints provides advantages over appearance-based techniques.
• This algorithm is highly effective for challenging scenarios like wide-baseline image pairs.