Three-dimensional array imaging of asynchronous structured light through diffusion model-based fringe pattern separation

Three-dimensional (3D) reconstruction using structured light is a fundamental technique for capturing the geometric shape of objects through the projection of fringe patterns. However, most of the existing structured light systems rely on a synchronization mechanism between the projector and camera components to image objects from a single viewpoint, limiting their capability to scale to full surface or format measurement. To this end, we propose to engineer a 3D imaging array system prototype by integrating multiple structured light units without imposing the synchronization requirement on them to provide a scalable and flexible solution for full surface 3D high-fidelity measurement. We demonstrate this approach with four asynchronous projector and camera units by analyzing and solving the fringe mixing imaging problem within each unit and between them. Specifically, we introduce a diffusion model-based generative learning framework, with a U-Net-like encoder-decoder architecture, for separating the fringe patterns from the mixed fringe observations within and across overlapping fields of view (FOVs) of adjacent units. Experimental results show that the proposed method is capable of generating high-quality fringe patterns under challenging mixing conditions and reconstructing object surfaces with remarkable geometric fidelity. It also generalizes well to complex object geometries, demonstrating strong potential for asynchronous, multi-view structured light 3D reconstruction.