Accurate depth estimation in structured light fields.

Area of Science:

• Optics and Photonics
• Computer Vision
• 3D Reconstruction

Background:

• Passive light field imaging struggles with depth estimation in complex scenes due to reliance on image structure.
• Existing methods using defocus and correspondence cues face robustness and accuracy limitations.

Purpose of the Study:

• To develop a novel active light field depth estimation method.
• To overcome the limitations of passive methods in complex environments.
• To enhance the accuracy and robustness of depth estimation.

Main Methods:

• Analyzed defocus and correspondence depth cues using phase encoding instead of image structure.
• Investigated spatial variance for defocus and angular variance for correspondence.
• Developed an active method utilizing the correspondence cue in a structured light field for direct depth search.

Main Results:

• The proposed active method directly searches for non-ambiguous depths without optimization.
• Angular variance weighting reduces depth estimation uncertainty using phase encoding.
• Experimental results show clearer depth region distinction and improved phase consistency compared to passive methods.

Conclusions:

• The novel active light field depth estimation method offers superior robustness and accuracy.
• Phase encoding with structured light fields provides a reliable approach for 3D scene reconstruction.
• The method effectively addresses challenges in complex scenes with rich colors.