Depth acquisition in single-pixel imaging with multiplexed illumination.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Machine Learning

Background:

• Single-pixel imaging (SPI) offers high signal-to-noise ratio and broad spectral range, making it suitable for scenarios where array sensors are costly or unavailable.
• Conventional SPI systems lose target depth information due to 3D-to-1D projection during acquisition.
• Existing SPI methods require specialized hardware or complex setups for depth information retrieval.

Purpose of the Study:

• To develop an efficient depth acquisition method for existing single-pixel imaging systems.
• To enable the simultaneous recovery of reflectance and depth information without additional hardware.
• To reduce sampling ratios and computational complexity in single-pixel imaging.

Main Methods:

• A multiplexed illumination strategy using both random and sinusoidal codes to encode spatial and depth information.
• Development of a convolutional neural network (CNN) for reconstructing spatial and depth data from 1D measurements.
• End-to-end deep learning reconstruction for efficient decoding of encoded information.

Main Results:

• The proposed method successfully acquires both reflectance and depth information from existing SPI systems.
• The technique achieves a 30% reduction in sampling ratio compared to conventional methods.
• Computational complexity is reduced by two orders of magnitude through deep learning reconstruction.
• Simulations and experiments validate the effectiveness and efficiency of the depth acquisition method.

Conclusions:

• This work presents an efficient and hardware-free approach for depth acquisition in single-pixel imaging.
• The combined illumination strategy and deep learning reconstruction significantly enhance SPI capabilities.
• The method offers a practical solution for applications requiring 3D information from cost-effective imaging systems.