Dense stereo fish-eye images using a modified hemispherical ASW algorithm.

Area of Science:

• Computer Vision
• Robotics
• 3D Reconstruction

Background:

• Traditional stereo matching methods like adaptive support weight (ASW) are ineffective for distorted fish-eye images.
• Fish-eye image distortions prevent direct application of classical methods designed for perspective images.

Purpose of the Study:

• To develop a robust dense disparity estimation method for uncorrected fish-eye images.
• To adapt the adaptive support weight (ASW) method for the unique challenges posed by hemispherical imaging.

Main Methods:

• Proposed a modified hemispherical ASW method within a hemispherical framework.
• Calculated 3D epipolar curves on a hemispherical model to encompass the entire image disc.
• Introduced a modified ASW with hemispherical support windows and geodesic distance.
• Developed a three-dimensional epipolar distance transform (3DEDT) to address textureless regions.

Main Results:

• Successfully demonstrated dense stereo matching for fish-eye images.
• Achieved objectively evaluated and visually convincing results on a public fish-eye dataset.
• The modified hemispherical ASW method effectively handles fish-eye image distortions.

Conclusions:

• The proposed hemispherical framework and modified ASW method provide an effective solution for dense disparity estimation in fish-eye images.
• The integration of 3D epipolar curves and 3DEDT enhances matching accuracy, particularly in challenging areas.
• This work advances stereo vision techniques for wide-angle imaging applications.