Point cloud semantic segmentation has been widely applied for bridge inverse modeling. However, existing methods are either labor-intensive or exhibit poor generality for real-world bridges. To address these limitations, this paper presents a bridge semantic segmentation method based on a pre-trained visual model. A viewpoint selection method based on view consensus is proposed to evaluate and optimize the viewpoints. The key insight is ensuring two adjacent viewpoints share a substantial consensus with high component visibility. The proposed self-prompt augmentation strategy enhances the performance of the multi-view image segmentation by fusing the initial prompt with cross view 2D masks. Bridge components are extracted through hard voting and further refined via post-processing. Experimental results demonstrate our method achieves state-of-the-art performance on real-world bridges. It provides reliable semantic information of bridge point cloud data for bridge inspection and maintenance applications.

中文翻译：

---

基于视图一致性和跨视图自提示融合的桥接点云语义分割


点云语义分割已广泛应用于桥梁逆建模。然而，现有的方法要么是劳动密集型的，要么对于现实世界的桥梁来说表现出较差的通用性。为了解决这些限制，本文提出了一种基于预训练视觉模型的桥梁语义分割方法。提出了一种基于视图一致性的视点选择方法，对视点进行评估和优化。关键的见解是确保两个相邻的视点共享实质性共识，并具有很高的组件可见性。所提出的自我提示增强策略通过将初始提示与交叉视图 2D 蒙版融合，增强了多视图图像分割的性能。Bridge 组件通过硬投票提取，并通过后处理进一步细化。实验结果表明，我们的方法在真实世界的桥梁上实现了最先进的性能。它为桥梁检查和维护应用提供了可靠的桥梁点云数据的语义信息。