Non-line-of-sight imaging with arbitrary illumination and detection pattern.

Area of Science:

• Computational imaging
• Optics and photonics

Background:

• Non-line-of-sight (NLOS) imaging reconstructs objects hidden from direct view.
• Current NLOS methods demand extensive, regularly gridded measurements, limiting practical use in dynamic environments like robotics and autonomous driving.

Purpose of the Study:

• To develop a flexible Bayesian framework for NLOS imaging.
• To overcome the limitations of dense measurement requirements in existing algorithms.
• To enable high-quality reconstruction of hidden object albedo and surface normals under general relay conditions.

Main Methods:

• A Bayesian framework is proposed for NLOS imaging, removing constraints on illumination and detection point patterns.
• A novel algorithm, confocal complemented signal-object collaborative regularization (CC-SOCR), is introduced using virtual confocal signals.
• The method achieves high-fidelity reconstruction of object properties like albedo and surface normals.

Main Results:

• The CC-SOCR algorithm successfully reconstructs hidden object details (albedo and surface normals) in diverse relay scenarios.
• The framework significantly reduces measurement density requirements on regular relay surfaces, decreasing acquisition time.
• Experimental validation demonstrates the broad applicability and effectiveness of the proposed NLOS imaging approach.

Conclusions:

• The developed Bayesian framework and CC-SOCR algorithm offer a flexible and efficient solution for NLOS imaging.
• This approach broadens the practical application scope of NLOS imaging in fields requiring vision in occluded environments.
• The method advances NLOS imaging by enabling detailed reconstruction with reduced measurement complexity.