Phase unwrapping based on channel transformer U-Net for single-shot fringe projection profilometry

Abstract

Single-shot fringe projection profilometry (FPP) has become a more prevalently adopted technique for retrieving the absolute phase values of the objects in intelligent manufacturing, defect detection, and some other important applications. In FPP, phase unwrapping plays a decisive role as the quality of final three-dimensional reconstructions relies on how accurately the phase values have been calculated. However, noise, shadow, discontinuity, and aliasing often exist in the original wrapped phase patterns which increase the complexity and difficulty of phase unwrapping, even lead to the unwrapping failure. To deal with phase unwrapping problems, we propose a phase unwrapping method based on deep learning, called channel transformer U-net, for directly extracting absolute phase value from the wrapping phase patterns. In the proposed method, the advanced channel-wise cross-fusion transformer module is integrated into the design of deep U-Net architecture. And a new loss function by combining the Smooth L1 loss and multi-scale structural similarity function is proposed. The effectiveness of the proposed method has been verified by real dynamic FPP measurements. The experimental results demonstrate that the proposed channel transformer U-Net can obtain accurate, complete, and smooth absolute phase values in real FPP measurement.