To understand the utility of this development, it is helpful to first understand what a digital twin, or 3D city model, actually is. A 3D city model is a detailed digital representation of the buildings and objects in an urban area. These virtual replicas provide a comprehensive view of a city's layout and infrastructure, assisting urban planners and decision-makers in understanding the environment and aligning their decisions with long-term goals. For these models to be highly effective at a Level of Detail 2 (LoD2) standard, they must include accurate representations of roof structures. Historically, generating these models from aerial imagery has presented challenges, as orthophotos often suffer from low contrast and shadows, which can lead to missing data.
An artificial intelligence framework developed by Faezeh Soleimani Vostikolaei and Dr. Shabnam Jabari at the University of New Brunswick offers a viable method to ease this bottleneck. The system relies on a bimodal deep learning network. Instead of depending on a single data source, the algorithm merges RGB optical imagery, such as high-resolution orthophotos, with LiDAR height data from Digital Surface Models.
By integrating this height data, the UNB system compensates for the optical limitations caused by shadows and poor lighting. The AI framework learned to classify common building architectures—including flat, gable, hip, and pyramid roofs—achieving an overall accuracy of 97.58%. When generating the final 3D building models, the automated system demonstrated a geometric error of 1.03 meters, adhering to the requirements of the Open Geospatial Consortium's CityGML 3.0 standard.
Economic and Industrial Use
By reducing the manual overhead involved in digitizing urban structures, this technology presents practical opportunities for scalable industrial and commercial applications. The economic utility of these automated models extends across several key sectors.
In the realms of 3D cadastre and urban planning, accurate digital models are highly beneficial for registering property ownership and spatial boundaries. They also provide necessary data for studying volumetric density and estimating populations for future urban development.
Furthermore, precise roof geometries allow energy companies to map solar envelopes effectively. This capability helps identify the specific areas on buildings that are suitable for solar panel installation on a broader, city-wide scale.
The models also show promise in disaster and facility management. They enable predictive modeling that can aid in evacuation planning, identify vulnerable areas, and assist in mitigating flood damage. Additionally, highly detailed building layouts support the management of modern utilities, while the geometric data can help telecommunications companies optimize radio infrastructure and radio-wave propagation.
By automating the transition from raw aerial data to structured 3D assets, this framework provides an algorithmic foundation that could help cities manage their infrastructure more efficiently in the future.