Non-line-of-sight reconstruction with signal-object collaborative regularization.

Area of Science:

• Computational imaging
• Computer vision
• Optics

Background:

• Non-line-of-sight (NLOS) imaging recovers objects from scattered light.
• High measurement noise challenges NLOS reconstruction quality.
• Applications include autonomous driving, rescue operations, and remote sensing.

Purpose of the Study:

• Develop a unified regularization framework for robust NLOS imaging.
• Address challenges posed by high measurement noise in diverse environments.
• Improve reconstruction fidelity and visual quality of hidden objects.

Main Methods:

• Established a unified regularization framework incorporating sparseness and non-local self-similarity.
• Integrated signal smoothness into the regularization.
• Tailored the framework for various indoor/outdoor scenes and noise levels (confocal/non-confocal).

Main Results:

• Demonstrated robust reconstruction of hidden object signals, albedo, and surface normals.
• Achieved high-quality reconstructions even with substantial measurement noise.
• Outperformed state-of-the-art algorithms in quantitative and visual assessments.

Conclusions:

• The proposed regularization framework significantly enhances NLOS imaging performance.
• Robustness to noise is achieved through incorporated object properties and signal smoothness.
• The method offers faithful and high-quality reconstruction of obscured objects.