Deep Learning-Based Half-Cycle Order Complementary Coding for Phase Unwrapping in Structured Light 3-D Profilometry

Phase unwrapping is a critical step in fringe projection profilometry (FPP), yet achieving a balance between robustness and efficiency remains challenging. In this paper, we propose a novel phase unwrapping method that integrates half-cycle order complementary coding with deep learning. First, we design a half-cycle order encoding fringe pattern (HCOEP) and train a Res-UNet model to predict half-cycle orders from HCOEP features. Second, we generate a pair of spatially complementary HCOEPs via phase analysis, enabling efficient derivation of complementary fringe orders without requiring additional projected patterns. Finally, we introduce a dual-interval phase unwrapping strategy to improve robustness in static reconstruction, and develop an image reuse strategy combined with phase weighting to enhance reconstruction efficiency in dynamic scenarios. Experiments conducted on a binocular structured light system demonstrate that our method outperforms existing deep learning–based and conventional approaches in both efficiency and robustness, while effectively avoiding order-jump errors caused by order ambiguity.