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

Large-format structured-light measurement has a practical trade-off: a single projector-camera unit is accurate but limited in field of view, while dual or multi-head systems introduce overlapping projected fringes that can alias and corrupt phase recovery.

Algorithm principle

The paper builds a dual-head structured-light prototype with two synchronized projector-camera units. The overlap between the two projection fields creates aliased fringes. A generative diffusion model is trained to recover clean fringe patterns from these aliased observations, allowing phase reconstruction and fusion across the two heads.

Measurement setting

The design targets large complex surfaces where moving the object, scanning repeatedly from multiple viewpoints, or relying on a single narrow field of view is inefficient. The dual-head setup expands coverage while the learned recovery model handles the interference produced by simultaneous projection.

Representative figures

Results and impact

The paper reports two validation tracks: fringe-pattern recovery and 3D reconstruction. The diffusion model achieves fringe recovery PSNR up to 45.80 dB, and the reconstructed geometry reaches accuracy up to 8 um. The result is relevant for high-accuracy industrial measurement of large parts and surfaces with complex curvature.