Real-Time Shadow Detection From Live Outdoor Videos for Augmented Reality

Area of Science:

• Computer Vision
• Augmented Reality

Background:

• Accurate shadow detection is vital for realistic augmented reality (AR).
• Existing methods often fail with moving viewpoints and dynamic scenes.

Purpose of the Study:

• To develop a robust shadow detection framework for live outdoor videos with moving viewpoints.
• To enhance shadow simulation in augmented reality.

Main Methods:

• A novel framework processing tracked and emerging regions using optical flow.
• Feature extraction from intensity profiles and Bayesian learning for shadow boundary detection.
• Incorporation of spatial layout constraints to eliminate spurious shadows.

Main Results:

• Outperforms state-of-the-art methods in real-time shadow detection under challenging conditions.
• Achieves a 33.3% increase in average F-measure on a custom dataset.
• Generates realistic shadow interactions when combined with image-based shadow-casting.

Conclusions:

• The proposed framework offers superior performance for real-time shadow detection in dynamic AR environments.
• Enables more immersive and visually accurate augmented reality experiences.