Single-shot multi-line structured light stripe recognition based on deep learning

Multi-line structured light measurement, as a high-speed and high-precision 3D surface profiling technique, has been widely adopted in reverse engineering, artifact restoration, and industrial metrology. However, the captured multi-line stripes image often exhibits cracks and misalignments, which pose significant challenges to the sequential recognition and numbering of stripes. Traditional methods typically rely on complex auxiliary coded patterns for stripe numbering, which reduces measurement efficiency. To overcome these challenges, this paper proposes a multi-line structured light stripe numbering method based on deep learning. This method performs semantic segmentation of multi-line structured light stripes without requiring projected auxiliary encoding patterns. Subsequently, the center lines of the stripes are applied to the semantic segmentation results to determine the ordering of the center lines, thereby enabling the numbering of multi-line structured light stripes. Because only one multi-line stripes pattern is necessary for stripe numbering, it requires no additional hardware setup. The proposed method is efficient and flexible for various applications. This experiment verified the feasibility of this method and validated it in complex measurement scenarios with high reflectivity and scattering.