ABSTRACT
Scan-to-BIM is precise method for BIM (Building Information Modeling) by measuring existing structures with LiDAR (Light Detection And Ranging), but has a limitation in that it consumes a lot of manpower, time, and cost. In order to overcome this limitation, studies on automating the semantic segmentation of 3D pc (point cloud) using deep learning are being conducted, but they have been actively applied to the architecture field. Research in the railway field is about applying deep learning to simple bridge data, and there are very few studies on railway tunnels in which various objects. In this study, using deep learning, semantic segmentation of the actual measurement 3D pc data of the railway tunnel was performed and its applicability was reviewed. By applying representative deep learning algorithms, we examined which algorithms are suitable for tunnel pc segmentation. Also, the effect of changing the hyperparameters of the training data on accuracy and IoU (Intersection over Union) was reviewed.
