Accurate 3D Structured-Light Measurement of Highly Reflective Objects

Achieving high-precision 3D reconstruction of highly reflective objects remains a major challenge in structured light measurement due to local overexposure and fringe degradation caused by specular reflections, which destabilize phase retrieval and reduce reconstruction accuracy. To address this problem, we propose an enhanced structured-light reconstruction network, FaNIC-Net, enabling robust feature extraction and fringe restoration under strong reflective interference. FaNIC-Net comprises two complementary modules: a Frequency-aware Multi-scale Convolution (FaMC) module that embeds DFT/IDFT operations into a multi-scale convolution pipeline to enhance critical frequency components and preserve fringe periodicity, and a Nearest-Neighbor Interpolation Convolution (NIC) module that decouples resolution enhancement from convolution, effectively mitigating checkerboard artifacts and improving high-frequency texture continuity. Experiments on real and synthetic datasets demonstrate that FaNIC-Net outperforms state-of-the-art methods in terms of MAE and SSIM, achieving superior depth recovery particularly in severely reflective regions, thus providing a robust and generalizable solution for fringe degradation on high-reflective surfaces.