The Fourth Industrial Revolution has accelerated the shift from traditional manufacturing to intelligent manufacturing. As the “eye of industry”, 3D measurement provides comprehensive, accurate, and reliable information for precision manufacturing. The traditional structured-light method is widely used for its high accuracy and flexibility. However, it relies on a point-to-point triangulation mechanism and is mainly applicable to single-reflection scenes. As application demands expand, measurement tasks increasingly involve mixed illumination conditions, such as interreflection, subsurface scattering, thin scattering media, and semitransparent layered scenes. In these cases, direct and global illumination become coupled, causing conventional point-to-point 3D reconstruction method to fail and greatly limiting its applicability. To address this challenge, researchers introduced light transport coefficients (LTC) to describe complex light propagation. As a representative approach, PSI reconstructs LTCs through a point-to-plane computational imaging framework. It enables effective separation and localization of complex illumination components and shows advantages in mixed-scene reconstruction. However, a complete theoretical model is still lacking to fully explain its imaging mechanism and quantitatively characterize its measurement performance, creating obstacles to the application of technology.
The research team led by Professor Zhang Qican and Associate Researcher Wu Zhoujie at Sichuan University established a comprehensive imaging and noise modeling framework for PSI. The proposed imaging model characterizes the distribution of LTC under complex illumination conditions. It reveals the mechanism of separating and localizing the direct-illumination components under global illumination (Fig. 1). The noise model quantitatively analyzes the influence of ambient noise on measurement accuracy (Fig. 2), providing a comprehensive theoretical framework for error propagation analysis in PSI.
To validate the proposed theoretical models, the team conducted a series of representative experiments under complex illumination conditions (Fig. 3). It covers two categories: complex reflection conditions, including overexposure, strong interreflection, subsurface scattering; and complex transmission conditions, including a structure viewed through a semitransparent layer and complex structures obscured by thin haze. The experimental results agree with the theoretical analysis, confirming the accuracy of the imaging model. In addition, the noise model was validated by simulating localization success rate of the direct component under different noise levels. Accuracy evaluation experiments were carried out under different signal-to-noise ratio (SNR) conditions (Fig. 4), showing PSI significantly outperforms conventional methods in low-SNR environments.
This work first establishes comprehensive theoretical framework for PSI, quantitatively characterizes its "point-to-plane" imaging capability and error propagation process. Based on this, it highlights key challenges and opportunities for future (Fig. 5), including efficient encoding and retrieval of LTC, decoupling and characterization of global components and utilization of direct components, global-component decoupling and characterization, and broader applications using LTC. The imaging paradigm based on light transport modeling is driving the evolution of structured light 3D imaging technology from traditional geometric triangulation toward a computational 3D imaging paradigm.
Advanced Devices & Instrumentation
Experimental study
Not applicable
Modeling of Parallel Single-Pixel Imaging for 3D Reconstruction: New Insights and Opportunities
6-Feb-2026