Deep diffusion model-based dual structured light for large-format complex surface imaging

Abstract Phase shifting-based structured light is widely used in three-dimensional (3D) topographic measurements due to its high accuracy. However, the typical single projector–camera configuration faces challenges such as blind spots, limited measurement range, and slow speed when reconstructing large-format complex objects. To address these limitations, we propose a dual-head structured light prototype system based on generative diffusion models. The system comprises two synchronized projector–camera units with minimal overlap to fuse 3D data. The diffusion model framework resolves fringe pattern aliasing in the overlap region, enabling the dual-head system to capture surfaces with high accuracy. We validate the proposed method through two experiments: one focusing on fringe pattern recovery and the other on 3D reconstruction. The results demonstrate that the method achieves high-quality fringe pattern recovery with a peak signal-to-noise ratio of up to 45.80 dB and a 3D reconstruction accuracy of up to 8 µ m.