CityGML is emerging as a model for defining the structures for organising urban information. It aims to identify the meaning and function of objects in a city. Thomas Kolbe became initially involved in CityGML from the beginning. Vector1 Media editor Jeff Thurston met up with him at the recently held Open Geospatial Consortium (OGC) Interoperability Day in Potsdam, Germany for this interview on the topic of CityGML.
V1 Magazine: How did you get involved in CityGML?
Kolbe: It would probably not be wrong to say that I was one of the initiators back in 2002. We had assembled a group of interested people, the Special Interest Group 3D, back then and many people were beginning to realise that there was no way of exchanging 3D city models without incurring a loss of information. The members of the SIG 3D wanted to do more than visualise information alone, they wanted to exchange the underlying information. They were also interested in more than individual buildings, they were interested in looking at whole cities.
It was apparent that the envisaged model quality could not be exchanged easily, so we started working on a data model and an exchange format. But at that time it was not foreseeable that that we would be where we are today. Around 2004 the Ordnance Survey, UK became involved and we started to realise that a German standard alone would not be enough. We changed direction toward a more international approach. Since some of the people were already active in the Open Geospatial Consortium (OGC) then it was possible to begin exploring this work through the OGC.
V1 Magazine: What caused you to appreciate that an international direction would be beneficial?
Kolbe: We were drawn to the usefulness of taking this approach. By then we had a demo that some OGC members had seen, like Bentley and Oracle. They appreciated where this work could lead to and approved our efforts as they expanded within OGC. Then other key players saw the usefulness and began to rally around the concept and that put the work into an international perspective quickly.
This gave us greater confidence in our work. It is rewarding today to watch so many people explore possibilities for using CityGML. It is interesting to see them investigating and realizing the differences between CityGML and KML, for example.
{sidebar id=145 align=right}
V1 Magazine: What is the difference between CityGML and KML?
Kolbe: CityGML is intended as a means to share the information contents of virtual 3D city models. KML is more about visualising the information. CityGML is not meant as an end user format within a browser per se, but instead, it is designed to provide information about the objects themselves. A CityGML dataset may have different visualizations where objects may be selected or highlighted according to different thematic criteria resulting in multiple KML files.
V1 Magazine: Then CityGML is about encoding information?
Kolbe: Yes, but in the second place. In the first place, CityGML defines structures for the organization of urban information which can be used in a broad range of applications. CityGML is not just interested in the shapes and graphical appearances of objects. Instead, it seeks to represent the meaning and functions of objects. It is more about the semantic quality of the information. For example, this might include information about building facades and their specific surface material and noise absorption. That information could then be shared with different applications like noise dispersion simulation.
V1 Magazine: That would suggest that it could be useful in building information modeling (BIM) perhaps?
Kolbe: Yes. In 2005 there was the OGC testbed (OWS-4) and we had a meeting with the people from the International Alliance for Interoperability. At the time we discussed the relationship of CityGML to Industry Foundation Classes (IFC) and it was clear that IFC would be a wonderful data source for CityGML. Also some of parts of the building model in CityGML are inspired by IFC.
V1 Magazine: Why would CityGML and IFC work well together?
Kolbe: It’s important to keep in mind that both models were not only modelling shapes but the underlying purpose of the objects. However, they are both working at different scales. While CityGML is representing objects at the scale of a city, the IFC model is modeling at the building level with a great amount of detail and richness.
For planning and construction, IFC would be more approprate. However, when those objects need to be integrated across the city with others, then it would be better if they were transformed to CityGML and then can be used together with other geographic data. This would then allow for combined spatial and thematic queries on buildings across cities within a GIS or a spatial database. The feasibility of this approach was demonstrated within the OGC testbed OWS-4 in a homeland security scenario.
V1 Magazine: The situation you describe could be likened to CAD for use in designing the buildings and GIS in use for the wider geography.
Kolbe: Yes, that’s a good analogy. The tools change dependent upon the scale and scope. CityGML and IFC are not just different at scale, but they also follow 2 different modelling paradigms. Industrial and engineering professionals tend to have a generative approach. For example, they construct a number of objects and then bring them together to form a building shape. The modelling employs a constructive geometric model which depends upon combinations of parametric geometries.
Alternatively, CityGML is modelled from observation and tailored to what can be seen. In CityGML one might have different information for the inside and outside surfaces of a wall. But IFC models would model the wall as one object. Using CityGML we have to account for the fact that from a single wall actually two ore more different wall surfaces are observable. As you can imagine, IFC allows for much quicker manual construction, but CityGML is much closer to surveying or photogrammetric registration methods.
{sidebar id=146 align=left}
V1 Magazine: What do you see as the interesting direction both CityGML and IFC will take in the future?
Kolbe: One of the questions being asked is, how do we go about deriving IFC models from already existing buildings? It will be interesting in the future to take a CityGML model and try to reconstruct an IFC model from it. CityGML can be easier reconstructed from observations than IFC. I’m interested in using photogrammetry or laser scanning as a means to automatically derive CityGML models. So this could be a promising processing path to come to IFC models for the built-up world with CityGML as intermediate step. Generally, it is much easier to move from an IFC model to a CityGML model but hard to go in the reverse.
BIM is booming in North America and many people are concentrating on IFC. Semantic 3D city models have been much more investigated in Europe so far. I expect we will see BIM in Europe within the near future but also semantic city models coming to the US.
V1 Magazine: What other questions do you find interesting going forward?
Kolbe: I think about those usages of 3D city models where you have a 3D database, but do not employ a graphics card to work with the model. Taking a homeland security example, we might want to have an application where an important politician moves along a particular route. In the application we are interested to find all the windows and buildings which have good view on that route and where possibly a sniper can hide. Instead of virtually visiting all building models with a 3D viewer along the route, we might rather want to query the city model to create a report of all corresponding windows, rooms, and buildings in order to check these. Thus we would exploit the semantic information of a city model along the route, and especially the details that come with a highly detailed CityGML or IFC model so we can locate and identify the windows.
Guided visualisation is another example where CityGML is useful because we might to take a virtual tour. But soon we may get stuck with our 3D viewer in some corner of the 3D model or become over loaded with information and miss the points from the things we want to see. A CityGML model would be searchable, and provide those points or types of information to the visitor, automatically guiding them along the way so they do not miss the types of things they are interested in. We could even ask the model where the best locations are for making a video of the things we like.
V1 Magazine: What sorts of things are you interested in today in your work?
Kolbe: My main interests lie in the semantic modeling of city models and their automatic extraction using photogrammetry or laser scanning. I am interested in developing flexible models that can extract objects from imagery or point clouds and classify them. I like the work which involves the deconstruction of models and the reconstruction of them. Here, the whole field of object recognition still faces the problem that while more flexible and generic models allow to represent more complex real world objects, they are often less specific and thus may not reject wrong hypothesis more often. Another field of interest is the ad-hoc integration of 3D city models, where different model parts from different providers or data sources should be merged automatically.
