Line-Constrained Camera Location Estimation in Multi-Image Stereomatching.

Area of Science:

• Computer Vision
• Photogrammetry
• Robotics

Background:

• Stereomatching provides dense depth data when active sensing is infeasible.
• Lightfield methods capture multiple views for depth estimation but require precise camera pose information.
• Current camera pose estimation relies on sparse feature matching, which can be less robust.

Purpose of the Study:

• To develop a novel method for estimating camera locations using dense correspondences within a lightfield.
• To leverage an existing depth estimation framework for simultaneous camera self-calibration.
• To demonstrate the effectiveness of dense correspondences for accurate camera pose determination in lightfield imaging.

Main Methods:

• Utilizing dense correspondences derived from lightfield data for camera self-calibration.
• Integrating camera location estimation within an established depth estimation pipeline.
• Employing a graphics processing unit (GPU) for efficient, parallelizable computation.

Main Results:

• Visual validation of camera location estimation through epipolar plane image rectification.
• Quantitative assessment of improved depth estimation accuracy due to precise camera poses.
• Demonstration of the proposed method as a viable alternative to sparse feature matching techniques.

Conclusions:

• Dense correspondences offer an effective approach for camera self-calibration in lightfield applications.
• The proposed method enhances depth estimation accuracy by providing accurate camera trajectory information.
• The parallelizable nature of the algorithm ensures efficient computation on modern hardware.