Continuous intensity-polarization fusion for depth estimation with DoFP polarimetric stereo

Area of Science:

• Computer Vision
• Robotics
• Optical Engineering

Background:

• Conventional stereo vision struggles with photometric consistency, especially in scenes with specular reflections or sparse textures.
• Polarization offers geometric cues but can be unreliable in high dynamic range (HDR) scenes due to sensor clipping, mimicking depolarization.
• Existing methods lack robustness in complex visual environments.

Purpose of the Study:

• To address the limitations of conventional stereo vision in challenging scenes.
• To develop a robust depth estimation method leveraging polarization information.
• To improve the accuracy and reliability of 3D reconstruction in complex environments.

Main Methods:

• A novel depth estimation method utilizing a division-of-focal-plane polarization camera.
• Simultaneous consideration of polarization reliability and intensity validity for depth calculation.
• Modeling polarization reliability as a continuous function for smooth fusion with intensity-based depth.

Main Results:

• The proposed method demonstrates improved depth completeness and consistency in complex scenes.
• Effective handling of issues arising from sensor clipping in high dynamic range scenarios.
• Successful fusion of polarization and intensity data for enhanced 3D perception.

Conclusions:

• The developed method enhances depth estimation accuracy by integrating polarization reliability.
• This approach offers a more robust solution for 3D scene understanding in challenging visual conditions.
• The findings contribute to advancements in computer vision and robotics requiring precise depth information.