Computational framework for steady-state NLOS localization under changing ambient illumination conditions.

Area of Science:

• Computer Vision
• Computational Imaging
• Robotics

Background:

• Non-line-of-sight (NLOS) imaging is crucial for applications like rescue operations and autonomous driving.
• Existing NLOS localization methods struggle with accuracy and robustness under diverse ambient illumination conditions.
• Steady-state NLOS imaging requires advanced computational techniques for reliable object detection.

Purpose of the Study:

• To develop a computational steady-state NLOS localization framework.
• To achieve accurate and robust NLOS imaging under various illumination conditions.
• To enable practical, around-the-clock, and all-weather NLOS imaging solutions.

Main Methods:

• Developed a physical NLOS image acquisition system and a corresponding virtual setup.
• Acquired real-captured and simulated steady-state NLOS images under different ambient illuminations.
• Utilized a multi-task convolutional neural network (CNN) for simultaneous background illumination correction and NLOS object localization.

Main Results:

• The proposed method effectively suppresses disturbances caused by ambient light variations.
• Significantly improved accuracy and stability of steady-state NLOS localization using consumer-grade RGB cameras.
• Demonstrated robust performance on both simulated and real-world captured NLOS images.

Conclusions:

• The developed computational framework provides a robust solution for steady-state NLOS localization.
• The method enhances the reliability of NLOS imaging in dynamic and challenging environments.
• This work paves the way for practical, all-weather NLOS imaging applications.