Spatial Clustering Guided Two-View Multi-Structural Deterministic Geometric Model Fitting.

Area of Science:

• Computer Vision
• Geometric Modeling
• Data Science

Background:

• Geometric model fitting is crucial for computer vision tasks.
• Handling multi-structural data with severe outliers presents significant challenges.
• Existing methods often struggle with accuracy and speed in complex scenarios.

Purpose of the Study:

• To develop a reliable and consistent method for two-view geometric model fitting on multi-structural data with outliers.
• To enhance the quality of sampled minimum subsets through improved spatial clustering.
• To ensure comprehensive coverage of all model instances within the data.

Main Methods:

• Utilizing spatial clustering with enhanced neighborhood preservation to deterministically sample minimum subsets.
• Implementing a multi-scale fusion strategy to increase high-quality subset generation and model instance coverage.
• Proposing a simple yet effective model selection algorithm for parameter estimation.

Main Results:

• The proposed method achieves fast, accurate, and stable model fitting results.
• Experimental results demonstrate significant superiority in accuracy and speed compared to state-of-the-art methods.
• A performance boost of over three times was observed on segmentation error, parameter error, and CPU time for specific datasets.

Conclusions:

• The novel approach effectively addresses the challenges of geometric model fitting in the presence of severe outliers.
• The method offers substantial improvements in computational efficiency and fitting accuracy.
• The developed datasets facilitate further research in homography and fundamental matrix estimation.