High dynamic range 3D reconstruction of milled surfaces using complementary exposures

Abstract When reconstructing the micro-topography of metallic workpieces with structured light, highly reflective and low-reflective regions in high-dynamic-range scenes often cause information loss or measurement errors, thereby degrading reconstruction accuracy. Conventional HDR approaches rely on capturing and fusing multiple sets of differently exposed images. Although these methods extend the imaging dynamic range, they markedly reduce capture and reconstruction efficiency. To address this issue, this study focuses on milled surfaces with micron-scale roughness and high dynamic range and proposes a structured-light 3D reconstruction method that incorporates complementary exposures. The method segments captured surface images into high-reflectance and low-reflectance regions based on grayscale values and, by maximizing regional information entropy, determines optimal complementary exposure times for each region type. A lightweight dual-exposure fusion model, PSA-CurveNet, performs feature-level complementary fusion of exposure pairs guided by fusion curves to generate high-quality fringe patterns for 3D reconstruction. Experimental results demonstrate that the proposed complementary-exposure approach can fully reconstruct reflective milled workpiece surfaces. Compared with other methods, it significantly reduces the number of captured images and improves reconstruction efficiency while maintaining comparable accuracy.