Robust and Globally Optimal Manhattan Frame Estimation in Near Real Time

Area of Science:

• Computer Vision
• Robotics
• Geometric Modeling

Background:

• Man-made environments often feature parallel and orthogonal planar structures.
• The Manhattan world assumption simplifies scene understanding by modeling these structures as Manhattan Frames (MFs).
• Existing MF estimation methods, while optimal, are computationally expensive and not real-time.

Purpose of the Study:

• To develop a computationally efficient method for Manhattan Frame (MF) estimation.
• To achieve global optimality in MF estimation with reduced computational complexity.
• To demonstrate the method's effectiveness and versatility in various computer vision applications.

Main Methods:

• Formulated MF estimation as a consensus set maximization problem.
• Introduced a novel bound computation method using the extended Gaussian image (EGI) domain.
• Relaxed the original problem to achieve constant complexity while preserving global optimality.

Main Results:

• The proposed method achieves constant complexity for bound computation.
• Global optimality is preserved through problem relaxation.
• Demonstrated strong performance on synthetic and real-world data, outperforming conventional methods.

Conclusions:

• The novel bound computation method significantly improves the efficiency of Manhattan Frame estimation.
• The approach maintains global optimality, making it suitable for real-time applications.
• The method's versatility is shown through applications in multiple MF estimation, video stabilization, and vanishing point estimation.