Visual recognition of weld beads is essential for post-weld robotic grinding. The recognition of thin-walled weld bead boundary, especially the backside boundary, remains challenging due to the diverse features such as debris, misalignment, and deformation. Based on point cloud from a laser scanner, we present a robust and accurate backside boundary recognition method for girth weld beads of thin-walled pipes. A boundary point extraction method is designed based on an adaptive sliding window model. Without prior morphology features, the influence of misalignment and deformation on the accuracy of boundary point recognition is greatly reduced by the local model matching strategy. Leveraging the correlation among overall weld bead features, an anomalous boundary point recognition and correction method based on DBSCAN clustering is proposed to further enhance robustness. A series of validation experiments were conducted by the obtained backside point cloud data inside a girth weld pipe, and our proposed method showed a high accuracy and a high robustness to misalignment, deformation and debris features.

中文翻译：

---

准确识别环焊缝的背面边界


焊缝的视觉识别对于焊后机器人打磨至关重要。由于碎片、错位和变形等多种特征，识别薄壁焊道边界，尤其是背面边界，仍然具有挑战性。基于激光扫描仪的点云，我们提出了一种稳健而准确的薄壁管道环焊缝背面边界识别方法。基于自适应滑动窗口模型设计了一种边界点提取方法。在没有先验形态特征的情况下，局部模型匹配策略大大降低了错位和变形对边界点识别精度的影响。利用焊道整体特征之间的相关性，提出了一种基于 DBSCAN 聚类的异常边界点识别和校正方法，以进一步增强鲁棒性。利用获得的背面点云数据在环焊管内进行了一系列验证实验，所提方法对错位、变形和碎片特征表现出高精度和高鲁棒性。