Achieving widely distributed feature matches using flattened-affine-SIFT algorithm for fisheye images

Area of Science:

• Computer Vision
• Image Processing
• Robotics

Background:

• Traditional fisheye image correction methods often lose details or cause distortion, hindering feature matching, especially in peripheral regions.
• Feature matching is crucial for applications like 3D reconstruction and simultaneous localization and mapping (SLAM).

Purpose of the Study:

• To develop a novel approach for robust feature matching in stereo fisheye images.
• To address the limitations of conventional correction methods by reducing image distortion and improving match distribution.

Main Methods:

• A new imaging model integrating scalable and hemisphere models was established.
• A flattened array model was designed to minimize fisheye image distortion.
• Affine transformation was applied to simulated flattened images for feature extraction and matching using differential expansion and optimal rigidity transformation.

Main Results:

• The flattened-affine-SIFT algorithm successfully identified a large number of reliable feature matches.
• Matches were found to be widely distributed across the entire effective image area.
• The method proved effective even in peripheral regions with significant distortion.

Conclusions:

• Flattened-affine-SIFT offers a significant improvement over traditional methods for feature matching in stereo fisheye imagery.
• The proposed approach enhances the reliability and distribution of feature matches, enabling better performance in computer vision tasks.
• This method effectively handles the challenges posed by fisheye lens distortion.