(English: AI for building detection: Deep learning and official geodata for recording undocumented buildings)

Funded by the Bavarian State Office for Digitization, Broadband and Surveying (LDBV)

Project Leader
Xiaoxiang Zhu

Project Scientist
Qingyu Li

Cooperation Partners
German Aerospace Center (DLR)

Funded by the Bavarian State Office for Digitization, Broadband and Surveying (LDBV), our team developed and delivered an AI-supported system for automated building change detection at statewide scale. The project addresses a key challenge in maintaining up-to-date cadastral building information: regularly identifying buildings that are not yet documented in the Digital Cadastral Map (DFK), as well as structural changes in already recorded buildings. To this end, we designed a machine learning workflow that learns building-related patterns from Bavaria’s official aerial remote sensing products, including true orthophotos and image-based surface height models, and automatically derives change candidates by comparing AI-based results with the cadastral reference. The system was developed in close collaboration with LDBV experts and tailored to operational constraints such as radiometric variability in aerial imagery, consistent processing across large areas, and the need for interpretable outputs that support human verification.

The resulting solution has been integrated into LDBV’s internal production environment and is used in a semi-annual monitoring workflow. It supports the detection of undocumented buildings as well as relevant building height changes, such as building upstocking (additional storeys). By automatically providing change candidates for targeted inspection, the system reduces manual effort and improves the efficiency of large-area cadastral update processes.

Publications:

Qingyu Li, Hannes Taubenböck, Yilei Shi, Stefan Auer, Robert Roschlaub, Clemens Glock, Anna Kruspe, Xiao Xiang Zhu (2022). Identification of undocumented buildings in cadastral data using remote sensing: Construction period, morphology, and landscape. International Journal of Applied Earth Observation and Geoinformation, 112, pp. 102909. 

Robert Roschlaub, Clemens Glock, Karin Möst, Frank Hümmer, Qingyu Li, Stefan Auer, Anna Kruspe, Xiao Xiang Zhu (2022). Implementierung einer KI-Infrastruktur zur automatisierten Erkennung von landesweiten Gebäudeveränderungen aus Luftbildern. ZFV-Zeitschrift für Geodasie, Geoinformation und Landmanagement

Qingyu Li, Yilei Shi, Stefan Auer, Robert Roschlaub, Karin Möst, Michael Schmitt, Clemens Glock, Xiao Xiang Zhu (2020). Detection of Undocumented Building Constructions from Official Geodata Using a Convolutional Neural Network. Remote Sensing, 12(21).