Conference Agenda

The regulation of urban advertising signage is critical for preserving visual harmony and ensuring regulatory compliance in modern cities. This study presents a novel pipeline for the automatic detection, tracking, geolocation, and textual identification of storefront signboards from 360° street-level imagery acquired via Mobile Mapping Systems (MMS). We first fine-tune a YOLOv11 object detection model on a custom-labeled dataset of urban scenes, enabling robust identification of signboards across varied viewing angles. To associate detections across consecutive frames, we leverage the integrated YOLOv11 tracking mode, which assigns consistent object IDs based on motion and appearance features. Each tracked instance is then localized in 3D space using a photogrammetric Line-of-Bearing (LoB) method, relying on known camera poses and pixel coordinates. In parallel, we extract the textual content from each detected sign using advanced GPT-4o Vision, which has demonstrated improved performance in complex visual environment. The proposed pipeline offers a scalable alternative to manual inspection, providing precise spatial and semantic information about urban signage. The internal geometric precision, quantified by a RMSE of 0.25 m derived from LoB intersection consistency, confirms the pipeline’s reliability for automated urban inventory systems and smart city applications.

Beside smart Information Management, Building Information Modeling (BIM) focuses on three-dimensional planning and model coordination. However, 3D CAD/BIM software does not conceptualize geodetic coordinates, which causes systematic deviations between geospatial surveys and 3D BIM Model. In local projects, the use of Cartesian coordinates (not curved geodetic coordinates!) is indispensable, due to the high precision demands in computational geometry. The solution is to convert geodetic coordinates into an optimized coordinate reference system. This procedure has been standardized and implemented for 6,326 traffic stations of the German Railway, minimizing systematic deviations and using the DB REF geodetic datum for consistent referencing the alignment of the railway
tracks. The developed approach enables the precise use of geometric models (CAD/BIM, high quality 3D point clouds, measured surveys) less than 5ppm, facilitates conversion to other coordinate reference systems using GIS-standard tools, and allows construction projects to be directly staked out from the
BIM-model.

This study quantitatively analyzes the status of three-dimensional attributes of building data in OpenStreetMap (OSM), a prominent example of Volunteered Geographic Information (VGI), and the characteristics of editing activities across Japan. Recently, demand for 3D city models has increased in various fields including urban planning, disaster prevention, and environmental simulation, but their development incurs substantial costs. OSM provides freely available open data with the capability to record 3D attributes such as building height and number of floors. However, existing research has primarily focused on city-level or 2D data quality assessment (Labetski et al., 2023; Lei et al., 2023). This study constructed a pipeline to extract and analyze building data from OSM, evaluating the spatial coverage of 3D attributes and the continuity and concentration of editing activities across Japan.

This work presents a method for automatically reconstructing point clouds into an IFC-compatible model based on CSG, aiming to facilitate a more efficient and effective BIM reconstruction process. The proposed workflow starts with segmented point clouds corresponding to rooms, doors, and windows, and automatically reconstructs an IFC model that contains key building components such as floors, ceilings, walls, openings, doors, and windows, with the advantage of preserving topology.