3D Point Cloud and BIM Component Retrieval for Subway Stations via Deep Learning

Yongfei Shi, Wei Ye, Bo Qu, Hong Jia, Zhanying Fu, Xinming Lv, Chenglu Wen, Weiquan Liu

Abstract: It is urgent to digitize the subway equipment to detect the changes in the components in the subway station in time. In this paper, we use the 3D point cloud of the subway station as a benchmark, compare it with the completed BIM model of the subway station, find out the changes in the components of the subway station, and retrieve 3D point cloud and BIM model components. First, we obtained 3D point cloud and BIM model components in Xiamen Metro Station. Second, we labeled 140 pairs of matched 3D point cloud components and BIM model component point cloud. Third, we constructed a Siamese network which embedded triplet loss to learn the feature descriptors of 3D point cloud and BIM model components, and then retrieve 3D point cloud components to BIM model components. Experimental results show that our proposed method realize the retrieval of 3D point cloud components to BIM model components in the subway station environment.