High-accuracy stereo matching based on adaptive ground control points.

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Traditional stereo matching relies on fixed ground control points (GCPs), limiting accuracy.
• Ill-posed matching problems require robust regularization techniques.

Purpose of the Study:

• To develop a high-accuracy stereo matching scheme using adaptive ground control points (AdaptGCP).
• To improve disparity map generation by adaptively selecting GCPs based on multiple criteria.

Main Methods:

• Adaptive GCP selection considering color dissimilarity, spatial relation, and pixel reliability.
• Alternating updating scheme for disparity and confidence maps to handle unreliable pixels.
• Local plane model fitting with GCPs for disparity reassignment and confidence map updates.
• Multistep filtering for final disparity map refinement.

Main Results:

• The AdaptGCP scheme effectively regularizes ill-posed stereo matching problems.
• Achieved state-of-the-art performance on the Middlebury benchmark.
• Demonstrated superior accuracy compared to existing stereo matching algorithms.

Conclusions:

• The proposed AdaptGCP method offers a novel approach to high-accuracy stereo matching.
• Adaptive GCP selection and robust refinement strategies enhance disparity map quality.
• This work provides significant insights for advancing stereo matching technology.