Hi @toshibamaher,

The tutorial that you first posted your question to creates flat roofs from the extrusion to the mean building height. Unfortunately generating roofs exactly to your data is a rather more complex task that may be better handled with another application, as mentioned by wkharto.

There's been a lot of discussion from the Community on different ways of achieving desired roof types. Though it might not be directly applicable to your data, perhaps it'll give you some ideas to try, or provide a good starting point for anyone else looking to do something similar. Here are some of the discussions that I've noticed on creating roofs:

• Mark's discussion on creating roofs has a lot of good ideas and even a few attached workspaces
• A Q&A on creating roofs
• A Q&A on creating arched roofs
• A Q&A on defining roof shapes (hip roofs)

And some articles that cover generating roofs (not exactly to data):

• Town of High Level
• City of Gavle in 3D

These articles cover generating roofs, creating approximate roof ridges. Hope these can provide some ideas for anyone seeking to do something similar.

- Jovita

@Maher as aready mentioned before, you could have a look at:

• CityGRID from UVM Systems (http://www.uvmsystems.com/index.php/en/)
• Building Reconstruction from Virtual City Systems (https://www.virtualcitysystems.de/en/ )

to start your research.

Hi @Maher,

I've been working in the field of 3D City Modeling for about 13 years with massive use of FME, but what you're looking for can't be done with a bunch of workspaces. There is a reason why you can find expert software on the market (e.g. UVM Systems: CityGRID; Virtual City Systems: Building Reconstruction), as this kind of data is highly complex and require sophisticated algorithms.

In principle it comes down to the two questions (on the geometric part of your topic): 1. Reliable plane detection and 2. geometric correct plane intersection. On both fields a lot of (scientific) research has been done in past years always with the goal of highest possible automation. Conclusion was, that under scientific-laboratory like conditions you can get 50%-70% automatically derived (but not by FME), but to get 100% you'll always have to invest interactive work.

Besides the geometric work every city model has to have a semantic structure that has to be imposed on your data too. It's not sufficient to extract some planes and visualize them together, as you for sure know. You'll have to define relationships beween them, construct parent-child relationships, assign IDs, attach geometric and thematic attributes etc.

I'd advise you to investigate existing city modelling software and explore possibilities to support them by help of FME.