Periodic Reporting for period 1 - 3DBAG (detailed 3D Building models Automatically Generated for very large areas)
Berichtszeitraum: 2022-09-01 bis 2024-02-29
Urban digital twins, or virtual 3D city models, have been used for some time now within specific domains. However, until recently it took a lot of effort to generate suitable 3D building models for a given application due to the required manual editing. Therefore, such detailed 3D models are rarely available for large areas (i.e. entire regions or nations) and typically rely on expensive licences and complex data formats. This makes these models hard to access and difficult to use. The limited availability and accessibility of detailed 3D building models hinders the efficient operation of urban areas and limits the economic and societal potential of 3D models. The solution lies in the unique method we developed in the preceding ERC project, which makes it possible to generate high-quality, detailed and easy-to-use 3D building models fully automatically at a national scale and use these 3D data without any technical or legal restrictions.
Such a digital twin of buildings provides a valuable information source for professionals in sectors such as planning, design, construction, energy and real estate. In addition, it offers a simulation basis for environmental impact studies on noise pollution, floods, wind flows, pollutant dispersion, urban heat islands, etc. Without such 3D data, it is at best laborious but often impossible to provide the insights necessary to efficiently maintain the built environment.
This PoC on ‘detailed 3D Building models Automatically Generated for very large areas’ (3DBAG) builds on the ERC project Urban Modelling in higher Dimensions (UMnD) that aimed at developing a fundamental solution for providing 3D data at application-specific Levels of Detail (LoDs). The project achieved several milestones that led to the origin of this idea. First, we have developed a 3D reconstruction method that automatically generates models at multiple LoDs from a combination of 2D data and height points. Second, to serve different urban applications from the same 3D base data, we have identified the 3D data needs for different applications and the reconstruction algorithms have been aligned to those needs. Third, we have developed a file standard (CityJSON) to accommodate such 3D data in a developer-friendly way to enable easy-use of the data by non-experts. This standard has been successfully adopted by the international standardisation Open Geospatial Consortium and is indeed used in practice to introduce our 3D data in new domains.
Before this PoC, the solution was available in a prototype and was used to generate 3D models for all of the 10 million buildings in The Netherlands. This dataset is available as open data since 2021 and is downloaded frequently and used in many applications. This PoC aimed at expanding this prototype to sustain this dataset as a long-term, reliable, up-to-date and user-friendly service that will remain openly available for everyone.
The impact of this PoC project for both professionals and the large public is high. The results enable a breakthrough in the use of 3D data, making it possible for a larger segment of society to benefit from the information 3D city models contain.
Such a digital twin of buildings provides a valuable information source for professionals in sectors such as planning, design, construction, energy and real estate. In addition, it offers a simulation basis for environmental impact studies on noise pollution, floods, wind flows, pollutant dispersion, urban heat islands, etc. Without such 3D data, it is at best laborious but often impossible to provide the insights necessary to efficiently maintain the built environment.
This PoC on ‘detailed 3D Building models Automatically Generated for very large areas’ (3DBAG) builds on the ERC project Urban Modelling in higher Dimensions (UMnD) that aimed at developing a fundamental solution for providing 3D data at application-specific Levels of Detail (LoDs). The project achieved several milestones that led to the origin of this idea. First, we have developed a 3D reconstruction method that automatically generates models at multiple LoDs from a combination of 2D data and height points. Second, to serve different urban applications from the same 3D base data, we have identified the 3D data needs for different applications and the reconstruction algorithms have been aligned to those needs. Third, we have developed a file standard (CityJSON) to accommodate such 3D data in a developer-friendly way to enable easy-use of the data by non-experts. This standard has been successfully adopted by the international standardisation Open Geospatial Consortium and is indeed used in practice to introduce our 3D data in new domains.
Before this PoC, the solution was available in a prototype and was used to generate 3D models for all of the 10 million buildings in The Netherlands. This dataset is available as open data since 2021 and is downloaded frequently and used in many applications. This PoC aimed at expanding this prototype to sustain this dataset as a long-term, reliable, up-to-date and user-friendly service that will remain openly available for everyone.
The impact of this PoC project for both professionals and the large public is high. The results enable a breakthrough in the use of 3D data, making it possible for a larger segment of society to benefit from the information 3D city models contain.
In this PoC, we addressed both economic and technical issues to further develop the current prototype into a service that can be deployed as a viable product.
Technically, we have achieved the following:
The 3DBAG reconstruction software has been significantly improved to generate 3D models for urban applications with an even higher quality and more robustness.
We have also added new functionalities as respons to users’ needs such as the volume-attribute, shared wall/exterior wall ratio, azimuth/orientation per roof surface, number of floors per building. These users’ needs have been collected via a continuous survey (available via the 3DBAG website), a users' workshop and via presentations at several meetings attended by (potential) users of 3DBAG data.
During the PoC we generated and published 3 new releases of 3DBAG data covering the whole of the Netherlands, with the improved software containing new functionalities in each new release. Consequently, the 3DBAG has been transformed into a solid 3D data source, significantly improved since the first release in 2021.
The new releases also include the use of a new version of the height data source, i.e. AHN4 instead of AHN3. The implementation of this new height source was preceded with a research on the impact of using AHN4 instead AHN3, resulting in a method that optimally combines LiDAR data sources at different timestamps.
The pipeline to apply the reconstruction to all buildings of the Netherlands (including pre- and post processing) has been redesigned, by which each new release can run within 2 weeks, without significant manual intervention.
A quality dashboard has been implemented to monitor and control quality issues for each release as well as a quality indicator per building. This is important both for further developments (since it points at issues that need to be resolved) and for users (since it provides insight into the quality of the reconstructed data of a specific building, which can also be caused by errors in the source (input) data).
The usability has been further increased by a new API that gives access to individual 3D features without the need to download complete tiles. In addition, in a related Horizon Europe funded funded project open source plug-ins have been developed for easy import of the 3DBAG data into design software.
Specific domain needs have been investigated and addressed to generate and enrich the 3D data to better fit with their needs, examples are data needs for energy consumption calculation and wind flow simulations resulting in the afore mentioned attributes.
Finally, methods have been investigated to further improve the quality of 3DBAG:
- A method has been developed to extract roof lines from true orthophotos and to combine it with Dense Image Matching point clouds during reconstruction. Such data sources are cheaper to acquire than LiDAR and have consequently a higher update frequency
- A method has been developed to fill no_data areas within buildings using AI. For this purpose a Building Point Completion benchmark has been generated.
- A method has been developed to repair geometrical and topological errors such as holes in 3D building models.
- A method has been developed to detect glass roofs, since the quality of LiDAR data on glass surfaces is poor.
Technically, we have achieved the following:
The 3DBAG reconstruction software has been significantly improved to generate 3D models for urban applications with an even higher quality and more robustness.
We have also added new functionalities as respons to users’ needs such as the volume-attribute, shared wall/exterior wall ratio, azimuth/orientation per roof surface, number of floors per building. These users’ needs have been collected via a continuous survey (available via the 3DBAG website), a users' workshop and via presentations at several meetings attended by (potential) users of 3DBAG data.
During the PoC we generated and published 3 new releases of 3DBAG data covering the whole of the Netherlands, with the improved software containing new functionalities in each new release. Consequently, the 3DBAG has been transformed into a solid 3D data source, significantly improved since the first release in 2021.
The new releases also include the use of a new version of the height data source, i.e. AHN4 instead of AHN3. The implementation of this new height source was preceded with a research on the impact of using AHN4 instead AHN3, resulting in a method that optimally combines LiDAR data sources at different timestamps.
The pipeline to apply the reconstruction to all buildings of the Netherlands (including pre- and post processing) has been redesigned, by which each new release can run within 2 weeks, without significant manual intervention.
A quality dashboard has been implemented to monitor and control quality issues for each release as well as a quality indicator per building. This is important both for further developments (since it points at issues that need to be resolved) and for users (since it provides insight into the quality of the reconstructed data of a specific building, which can also be caused by errors in the source (input) data).
The usability has been further increased by a new API that gives access to individual 3D features without the need to download complete tiles. In addition, in a related Horizon Europe funded funded project open source plug-ins have been developed for easy import of the 3DBAG data into design software.
Specific domain needs have been investigated and addressed to generate and enrich the 3D data to better fit with their needs, examples are data needs for energy consumption calculation and wind flow simulations resulting in the afore mentioned attributes.
Finally, methods have been investigated to further improve the quality of 3DBAG:
- A method has been developed to extract roof lines from true orthophotos and to combine it with Dense Image Matching point clouds during reconstruction. Such data sources are cheaper to acquire than LiDAR and have consequently a higher update frequency
- A method has been developed to fill no_data areas within buildings using AI. For this purpose a Building Point Completion benchmark has been generated.
- A method has been developed to repair geometrical and topological errors such as holes in 3D building models.
- A method has been developed to detect glass roofs, since the quality of LiDAR data on glass surfaces is poor.
There are several results beyond the state of the art and the potential impact of the results are high:
The project has provided a stable and reliable open source data service, which makes it possible to generate high-quality, detailed and easy-to-use 3D building models fully automatically at a national scale and use these 3D data without any technical or legal restrictions.
The 3D building data has been enriched for specific users’ needs providing ready-to-use information for all kinds of urban applications.
The underlying reconstruction software can be freely and openly used by others to generate detailed building models for specific areas or specific source data.
Several methods have been developed to correct errors in the source data to improve the quality of the 3D reconstructed data. In addition the use of other sources than LiDAR point clouds has been investigated. This enables using other data sources, which can be acquired more frequently and more easily.
The project has provided a stable and reliable open source data service, which makes it possible to generate high-quality, detailed and easy-to-use 3D building models fully automatically at a national scale and use these 3D data without any technical or legal restrictions.
The 3D building data has been enriched for specific users’ needs providing ready-to-use information for all kinds of urban applications.
The underlying reconstruction software can be freely and openly used by others to generate detailed building models for specific areas or specific source data.
Several methods have been developed to correct errors in the source data to improve the quality of the 3D reconstructed data. In addition the use of other sources than LiDAR point clouds has been investigated. This enables using other data sources, which can be acquired more frequently and more easily.