Non-line-of-sight imaging under white-light illumination using physics-enhanced deep learning

Area of Science:

• Computational imaging
• Optics
• Artificial intelligence

Background:

• Non-line-of-sight (NLOS) imaging enables seeing hidden objects via indirect views, with applications in medicine, security, and autonomous driving.
• Current NLOS methods often rely on expensive coherent illumination and specialized equipment, limiting widespread adoption.
• There is a need for practical NLOS imaging solutions that are cost-effective and accessible.

Purpose of the Study:

• To develop a novel NLOS imaging technique that overcomes the limitations of existing methods.
• To demonstrate the feasibility of NLOS imaging using readily available equipment and white-light illumination.
• To enhance the performance and scalability of NLOS imaging for real-world applications.

Main Methods:

• A new NLOS imaging method based on speckle correlation under white-light illumination was developed.
• A physics-enhanced deep learning approach was employed, integrating a speckle correlation model and denoising prior into a neural network.
• The system utilizes an ordinary camera for data acquisition.

Main Results:

• The proposed method successfully reconstructed object images from indirect views.
• Experimental results showed excellent performance in terms of recovered image quality.
• The technique demonstrated good scalability for practical deployment.

Conclusions:

• The developed NLOS imaging method offers a practical and cost-effective alternative to existing techniques.
• This approach removes the need for active coherent illumination and expensive hardware.
• The findings pave the way for broader applications of NLOS imaging in various fields.