This study presents a novel learning-guided point cloud vectorization to form the vector models of building components. To this end, two learning-based models are modified to realize feature detection and vectorization. The learning-guided scheme can comprehend the definition and mutual relationships of object vertices by learning through existing BIM models and thus predict the vector model of newly given point clouds consequently. Moreover, the effectiveness was verified by using point clouds under different quality levels. The quantitative indices showed promising results, in which the accuracy of object vertex positions achieved 10 cm in beam and column categories and less than 25 cm for all building components. On the other hand, the vertex connections of the vector models reported accuracy above 70%. Therefore, the results can be deemed as fundamental models to improve the automation performance of further refinements or subsequent value-added applications.

本研究提出了一种新颖的学习引导点云矢量化，以形成建筑组件的矢量模型。为此，修改了两个基于学习的模型以实现特征检测和矢量化。学习引导方案可以通过现有的BIM模型学习来理解对象顶点的定义和相互关系，从而预测新给定的点云的矢量模型。此外，通过使用不同质量级别的点云验证了有效性。定量指标显示出有希望的结果，其中对象顶点位置的精度在梁和柱类别中达到 10 厘米，在所有建筑组件中均小于 25 厘米。另一方面，矢量模型的顶点连接报告的准确度超过 70%。所以，