On 3-D scene flow and structure recovery from multiview image sequences

Area of Science:

• Computer Vision
• Robotics
• 3D Reconstruction

Background:

• Estimating 3D scene flow and structure from image sequences is crucial for applications like robotics and autonomous navigation.
• Existing methods often assume rigid scene motion, limiting their applicability to dynamic environments.

Purpose of the Study:

• To develop and evaluate two novel systems for computing dense 3D scene flow and structure from multiview image sequences.
• To handle nonrigid motion within scenes, expanding the capabilities of 3D scene analysis.

Main Methods:

• Integrated Model-based System (IMS): Assumes local 3D affine motion, fitting non-linear models using optical flow and stereo constraints for simultaneous motion and structure estimation.
• Extended Gradient-based System (EGS): Extends 2D optical flow, employing a hierarchical stereo matching algorithm for disparity estimation and utilizing multiview constraints for motion estimation, incorporating image segmentation for discontinuities.

Main Results:

• Both IMS and EGS successfully compute dense 3D scene flow and structure from multiview image sequences.
• The systems effectively handle nonrigid motion, a significant advancement over rigid-motion assumptions.
• Experimental results on synthetic and real imagery validate the proposed 3D motion and structure recovery schemes.

Conclusions:

• The developed IMS and EGS provide effective solutions for dense 3D scene flow and structure computation from multiview imagery.
• The ability to handle nonrigid motion makes these systems suitable for complex, dynamic environments.
• Empirical comparison between IMS and EGS offers insights into their respective performance characteristics.