Road blockages caused by landslides and debris flows obstruct emergency vehicles and delay rescue operations. Rapid identification of blocked roads is therefore essential for effective disaster response. This paper proposes a road blockage estimation method using terrain data obtained by remote sensing and interpolation method based on roadside regions. First, vegetation is removed from orthoimages to reduce the influence of trees, and terrain changes are extracted by calculating the difference between pre- and post-disaster Digital Elevation Model (DEM) generated from 3D point cloud data. Candidate sediment areas are then obtained by combining the vegetation mask with the DEM difference. To compensate for areas where sediment is difficult to detect directly, sediment information detected in roadside regions is interpolated into the road area, enabling more robust road blockage detection. Finally, blockage conditions are determined for each road segment and visualized on a road network. The proposed method was evaluated using post-disaster sediment maps and visually interpreted orthoimages. Experimental results demonstrate that the proposed roadside interpolation improves the detection of sediment-related road blockages and confirms the effectiveness of integrating terrain data with road information.

In Japan, the deterioration of infrastructure constructed during the period of rapid economic growth has led to serious issues. In response, the Ministry of Land, Infrastructure, Transport and Tourism has adopted the application of BIM/CIM principles, aiming to improve productivity through the utilization of 3D models. For existing structures, it is necessary to create 3D models based on measurement data, and LOD 300 is considered the standard for such 3D models. However, complex structures such as bridges require specialized knowledge and significant effort. Therefore, methods have been developed to automatically classify acquired point cloud data as preprocessing to reduce work costs. Nevertheless, these approaches are insufficient for reproducing the external geometry of objects as defined by LOD 300 and require significant manual work to create training data. Accordingly, this research proposes an efficient training data generation method utilizing 3D models to construct general semantic segmentation models on bridge components.