Robust active stereo three-dimensional reconstruction using RGB speckle projection

Stereo vision, as a critical part in the field of three-dimensional reconstruction, perceives and measures the 3D shape of objects by matching corresponding points between the left camera and the right camera views. The extremely challenging in stereo vision is to ensure the accuracy of matching areas and to obtain disparity robustly. However, passive stereo imaging has inherent problems for ill-posed areas such as complicated illumination, weak texture, and perspective occlusion in adverse environments such as industrial scenes. The active stereo technique designs and projects patterns to improve the accuracy of digital image correlation (DIC). Focusing on the intricate maintenance scenes of high-speed railways,we propose an active stereo strategy with RGB speckle projection patterns. Compared with previous speckle patterns, our RGB speckle pattern enhances more comprehensive texture information and demonstrates significant adaptability in diverse scenes. Comparative experiments were implemented on different stereo matching algorithms and projection strategies in industrial scenes,and the results demonstrate that the proposed approach achieves favorable reconstruction results and higher accuracy.