Automatic driving image matching via Random Sample Consensus (RANSAC) and Spectral Clustering (SC) with monocular...

Area of Science:

• Computer Vision
• Data Science
• Robotics

Background:

• The proliferation of Internet of Things (IoT) and autonomous driving generates vast visual data.
• Existing image matching algorithms struggle with low accuracy and inlier rates, hindering effective data utilization.
• Accurate visual information is critical for reliable autonomous driving systems.

Purpose of the Study:

• To develop an advanced image matching algorithm for autonomous driving applications.
• To address the limitations of current methods in achieving high accuracy and inlier rates.
• To enhance the reliability and robustness of visual perception in autonomous vehicles.

Main Methods:

• Proposed a novel SC-RANSAC algorithm integrating spectral clustering (SC) with Random Sample Consensus (RANSAC).
• Utilized monocular camera datasets from autonomous driving scenarios.
• Employed RANSAC for initial inlier identification, followed by SC for outlier filtering to refine the inlier set.

Main Results:

• SC-RANSAC demonstrated effective and reliable elimination of mismatches from initial image matching results.
• The algorithm achieved a high inlier rate, real-time performance, and robustness.
• Evaluated performance across camera translation, rotation, and combined transformations, outperforming existing methods like RANSAC and graph-cut RANSAC.

Conclusions:

• SC-RANSAC offers a significant improvement for image matching in autonomous driving contexts.
• The algorithm's robustness and accuracy make it suitable for real-world autonomous driving environments.
• Spectral clustering integration enhances the reliability of consensus-based matching techniques.