Common line-structured light center extraction algorithms have limitations in accuracy and robustness when extracting the center of line-structured light obtained from objects with complex surface structures. In this study, we propose a high-precision center extraction algorithm for multi-curvature line-structured light. This algorithm improves the enhanced parallel thinning algorithm to eliminate redundant points and burrs and fills for missing center points at the end of the light stripe via the improved internal propulsion center extraction algorithm. Finally, sub-pixel centers are recalculated in the direction of the normal vector. A comparative experiment using three classical algorithms was conducted based on a tooth model with a complex curved surface. The experimental results show that the mean absolute error of this algorithm is <0.1, which is only half that of the best-performing dual-threshold grayscale center-of-gravity method in the classical method. Extraction of the center can be performed for multi-curvature line-structured light over 20 curvatures in the range of ( 0.0001 , 4.1439 ) / pixel − 1; moreover, the end of the light stripe and the identification of the truncation can be improved. Therefore, the proposed algorithm is suitable for contour detection of multi-curvature targets and effectively improves the accuracy of high-precision object detection.

中文翻译：

---

基于牙齿模型的高精度多曲率线结构光中心提取算法

常见的线结构光中心提取算法在提取具有复杂表面结构的物体获得的线结构光中心时，在准确性和鲁棒性方面存在局限性。在本研究中，我们提出了一种多曲率线结构光的高精度中心提取算法。该算法改进了增强型并行细化算法，通过改进的内部推进中心提取算法消除冗余点和毛刺，并填充光带末端缺失的中心点。最后，在法向量方向上重新计算子像素中心。基于复杂曲面牙齿模型，采用三种经典算法进行对比实验。实验结果表明，该算法的平均绝对误差<0.1，这只是经典方法中性能最好的双阈值灰度重心法的一半。对于(0.0001,4.1439)/pixel-1范围内超过20个曲率的多曲率线结构光可以进行中心提取；并且可以提高光带的末端和截断的识别性。因此，该算法适用于多曲率目标的轮廓检测，有效提高了高精度目标检测的精度。可以提高光带的末端和截断的识别性。因此，该算法适用于多曲率目标的轮廓检测，有效提高了高精度目标检测的精度。可以提高光带的末端和截断的识别性。因此，该算法适用于多曲率目标的轮廓检测，有效提高了高精度目标检测的精度。