Depth estimation method based on adaptive occlusion handling for light-field imaging systems.

Area of Science:

• Computer Vision
• Optical Engineering
• Image Processing

Background:

• Depth estimation in light-field imaging is challenged by occlusions, limiting accuracy in many existing frameworks.
• Current methods often struggle with computational efficiency and robustness against occluded regions.
• Accurate depth perception is crucial for applications like augmented reality and robotics.

Purpose of the Study:

• To propose an embeddable adaptive occlusion-aware module (AOAM) for robust depth estimation in light-field imaging.
• To develop an efficient occlusion handling strategy with low computational overhead.
• To improve the overall accuracy and reliability of depth maps generated from light-field data.

Main Methods:

• An adaptive occlusion-aware module (AOAM) was developed, incorporating a voting strategy for occlusion optimization.
• Beam propagation characteristics were analyzed to effectively filter disparity values.
• Adaptive voting cost was employed for regional partitioning and noise reduction in the global domain.
• The module was designed for embeddability and low computational resource consumption.

Main Results:

• The proposed AOAM effectively compensates for deficiencies caused by occlusion in depth estimation.
• The adaptive occlusion optimization mode demonstrated efficient performance.
• Validated superiority of the method on a common light-field dataset, showing improved depth map accuracy.
• The technique achieved regional partitioning and noise reduction effectively.

Conclusions:

• The AOAM provides a significant advancement in addressing occlusion challenges for light-field depth estimation.
• The method offers a computationally efficient and accurate solution for real-world applications.
• This work contributes to more reliable depth perception in complex imaging scenarios.