Fast depth estimation with cost minimization for structured light field.

Area of Science:

• Computer Vision
• Optics
• Signal Processing

Background:

• Conventional light fields (LF) present challenges in depth estimation due to feature scarcity, leading to ambiguity and high computational demands.
• Existing methods often struggle with accuracy, particularly at object boundaries, and can be computationally intensive.

Purpose of the Study:

• To introduce a novel Phase Light Field (PLF) approach for improved depth estimation.
• To address the limitations of conventional LF depth estimation, including feature ambiguity and computational load.
• To enhance the accuracy and efficiency of depth map generation, especially for boundary regions.

Main Methods:

• Development of Phase Light Field (PLF) using sinusoidal fringes for latent phase coding.
• Utilizing re-formatted phase-epipolar-plane-images (phase EPIs) within a global cost minimization framework.
• Proposing a convex cost function based on weighted phase variance along candidate lines in phase EPIs.
• Employing the Beetle Antennae Search (BAS) optimization algorithm for efficient optimal line detection.
• Incorporating a bilateral filter for post-processing and refinement of depth maps.

Main Results:

• The proposed PLF method accurately estimates depth, showing significant improvements in boundary regions compared to conventional methods.
• Achieved a notable acceleration of approximately 5.9 times over state-of-the-art refocusing techniques.
• Demonstrated comparable depth quality to existing methods while offering substantial speed enhancements.
• Validated through both simulation and real experimental data, confirming its effectiveness.

Conclusions:

• Phase Light Field (PLF) offers a robust and efficient solution for depth estimation in light field applications.
• The developed method effectively mitigates depth ambiguity and reduces computational load.
• The enhanced accuracy and speed make PLF a promising technique for practical, real-world applications requiring precise depth information.