Implementation of an automated georeferencing workflow for architectural elements in GIS using ML and Cloud Computing

Abstract. This research presents a scalable, cloud-based workflow integrating Machine Learning (ML) and 3D Geographic Information Systems (GIS) to support the automated detection of architectural elements and urban management. Via Unmanned Aerial Vehicle (UAV) georeferenced images, the system enables an automated and scheduled detection, geolocation, and import of architectural elements (e.g., domes, photovoltaics panels, tanks) data and metadata into a 3D GIS environment. A validated urban case study was conducted using UAV-acquired georeferenced images processed through a Structure-from-Motion (SfM) pipeline. Orthoimage chunks and dataset were uploaded to Google Cloud Storage, triggering an event-driven architecture built on a Cloud Computing Infrastructure. The pipeline leverages Vertex AI object detection via AutoML, the predictions of which are subsequently enriched with geospatial metadata. The output data is stored in BigQuery and Cloud Storage for urban GIS integration and analysis. Results confirm the viability of the pipeline for repeatable, and automated urban monitoring, reducing manual labour and improving safety for building maintenance workers. This approach is focused on the use of mobile mapping data processing, 3D reconstruction of urban areas, AI process for detection and urban maintenance and to develop smart city applications.