Adaptive High-Precision 3D Reconstruction of Highly Reflective Mechanical Parts Based on Optimization of Exposure Time and Projection Intensity

This article is used to reconstruct mechanical parts with highly reflective surfaces. Three-dimensional reconstruction based on Phase Measuring Profilometry (PMP) is a key technology in non-contact optical measurement and is widely applied in the intelligent inspection of mechanical components. Due to the high reflectivity of metallic parts, direct utilization of the captured high-dynamic-range images often results in significant information loss in the oversaturated areas and excessive noise in the dark regions, leading to geometric defects and reduced accuracy in the reconstructed point clouds. Many image-fusion-based solutions have been proposed to solve these problems. However, unknown geometric structures and reflection characteristics of mechanical parts lead to the lack of effective guidance for the design of important imaging parameters. Therefore, an adaptive high-precision 3D reconstruction method of highly reflective mechanical parts based on optimization of exposure time and projection intensity is proposed in this article. The projection intensity is optimized to adapt the captured images to the linear dynamic range of the hardware. Image sequence under the obtained optimal intensities is fused using an integration of Genetic Algorithm and Stochastic Adam optimizer to maximize the image information entropy. Then, histogram-based analysis is employed to segment regions with similar reflective properties and determine the optimal exposure time. Experimental validation was carried out on three sets of typical mechanical components with diverse geometric characteristics and varying complexity. Compared with both non-saturated single-exposure techniques and conventional image fusion methods employing fixed attenuation steps, the proposed method reduced the average whisker range of reconstruction error by 51.18% and 25.09%, and decreased the median error by 42.48% and 25.42%, respectively. These experimental results verified the effectiveness and precision performance of the proposed method.