Monocular Depth Estimation for Glass Walls With Context: A New Dataset and Method.

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• Traditional monocular depth estimation fails with transparent surfaces like glass walls.
• Annotating glass wall depth is difficult as sensors and prior methods cannot handle transparency.

Purpose of the Study:

• To develop the first dataset (GW-Depth) and method for monocular depth estimation with transparent glass walls.
• To enable accurate depth perception in environments with challenging transparent surfaces.

Main Methods:

• Introduced the GW-Depth dataset with annotated glass wall depths, segmentation masks, and edge line segments.
• Proposed a novel monocular depth estimation method leveraging glass structure context (boundary lines) and reflection context (depth variance).

Main Results:

• The proposed dual context method significantly improves depth estimation accuracy in scenes with glass walls.
• Experimental results demonstrate superior performance compared to existing state-of-the-art methods.

Conclusions:

• Presents the first viable solution for monocular depth estimation in the presence of glass walls.
• The method's adaptability and accuracy make it suitable for applications like autonomous navigation.