Deep-learning-assisted composite polarized fringe projection profilometry for three-dimensional measurement of high-dynamic-range objects

The presence of highlights on the surfaces of high-dynamic-range objects poses significant challenges to their three-dimensional measurement. Achieving fast and high-precision reconstruction of high-dynamic-range objects has become a primary research focus in the field of optical measurement. Although the polarization coding technique can effectively mitigate the influence of highlights, it also introduces increased nonlinear errors and reduces the signal-to-noise ratio. To address these issues, we propose a deep-learning-assisted composite polarization fringe projection profile measurement method for the three-dimensional measurement of high-dynamic-range objects. This method combines data-driven deep learning techniques with physical-model-based polarization coding to eliminate highlights, reduce nonlinear errors, and mitigate the reduction in signal-to-noise ratio. Experimental validation across various scenes and different methods of highlight removal demonstrates that our proposed method significantly improves both measurement efficiency and accuracy compared to traditional physical models.