Multibody structure-and-motion segmentation by branch-and-bound model selection.

Area of Science:

• Computer Vision
• Robotics
• Machine Learning

Background:

• Multiple Structure-and-Motion (SfM) segmentation is crucial for understanding complex scenes.
• Existing methods often struggle with an unknown number of objects and outliers.

Purpose of the Study:

• To develop an efficient and robust framework for two-view multiple SfM segmentation.
• To address the unknowns of object memberships, fundamental matrices, and object count.

Main Methods:

• Hypotheses for fundamental matrices generated via local sampling.
• Combinatorial selection problem formulated to optimize model selection cost.
• Branch-and-bound technique used for robust segmentation and outlier handling.

Main Results:

• Efficiently searches solution space, guaranteeing optimality over current hypotheses.
• Rejects solutions without explicit evaluation, enhancing efficiency.
• Validated with synthetic data and demonstrated on real images.

Conclusions:

• The proposed branch-and-bound approach offers an efficient and optimal solution for multi-object SfM segmentation.
• The framework effectively handles unknown object numbers and outliers in real-world scenarios.