Deep-learning blurring correction of images obtained from NIR single-pixel imaging.

Area of Science:

• Optics and Photonics
• Computer Vision
• Remote Sensing

Background:

• Conventional cameras struggle in low-light and scattering conditions (rain, fog, smoke), reducing visibility.
• Near-infrared (NIR) light (850-1550 nm) offers reduced scattering, improving imaging in adverse conditions.
• Existing NIR cameras are often costly, necessitating alternative solutions.

Purpose of the Study:

• To develop and evaluate a cost-effective vision system for enhanced imaging in challenging outdoor environments, specifically targeting rainy conditions.
• To combine NIR active illumination single-pixel imaging (SPI) with time-of-flight (ToF) for 2D image reconstruction.
• To leverage diffusion models for improving the quality of reconstructed NIR-SPI images.

Main Methods:

• Proposed a vision system integrating 1550 nm NIR active illumination SPI with 850 nm time-of-flight (ToF) for 2D image reconstruction.
• Incorporated diffusion models to enhance the quality of the NIR-SPI images.
• Simulated outdoor laboratory conditions with varying background illumination and droplet sizes to assess system performance.

Main Results:

• Demonstrated the feasibility of the proposed NIR-SPI and ToF system for 2D image reconstruction in simulated rainy conditions.
• Showcased the effectiveness of diffusion models in improving the clarity and contrast of NIR-SPI images.
• Evaluated the system's performance under different environmental parameters, including droplet size and background light.

Conclusions:

• The proposed NIR-SPI system combined with ToF and diffusion models shows promise as a viable and potentially cost-effective solution for low-visibility imaging.
• This approach offers a pathway to overcome the limitations of conventional cameras in adverse weather conditions.
• Further research can explore real-world deployment and optimization for various challenging environments.