search

GANys / cityjson-pointcloud

Extension to support LAS 3D point clouds for CityObjects geometries. Works with CityJSON 1.0.1.
3 stars 0 forks source link

some issues #2

Open hugoledoux opened 2 years ago

hugoledoux commented 2 years ago

I checked your Extension and spotted the following issues:

  1. you give +geometry as an extra attribute but in the example file it's one child property of the CityObject, it should be a child of "attributes". Why does it validate then? Because any new property is allowed... Notice that if you remove the Extension your file is still valid according to the schemas...
  2. you use inheritance to add new attributes to Building by adding one to _AbstractBuilding, but JSON doesn't have the concept of inheritance so this cannot work... you'd need to use the anyOf or oneOf mechanism for it to work.

One solution to integrate point cloud would be for the CityJSON schema to have always MultiPoint as possible geometry, for all the CityObjects. This would be simple to do and I could maybe modify for a v1.1.1 release...

As for the external references to LAS files, then it's either in the "attributes" or you define in an Extension new CO for all the types (eg "+BuildingPC"). But I'm also wondering, with this system, the idea is that one LAZ file would contain only the point of say one building? Thus somewhere a server is hosting 100s of cropped files?

GANys commented 2 years ago

Thanks a lot for your corrections:

  1. Indeed, it fell through the cracks. Still, the best improvement could come from the core and allow MultiPoint for every geometry.
  2. Beforehand, I used the extension thanks to the Measur3D architecture. There, inhenritance is supported and help me reducing complexity.

Integrating MultiPoint in every geometry is certainly a solution. But, this goes against the basic CityGMl model. I am not confortable enough to propose such a solution. Moreover, it is up to the core team to propose such an advance. I am however happy if my thoughts can help move things forward. It remains to be seen if this does not pose a problem on other conceptual levels.

Finally, about the "pointFile", there is a problem, yes. From my point of view, using an architecture such as the Smart Point Cloud, would be a decent solution. I am influenced by the proximity of Dr. Poux, I confess. But it is still under research.

The TUMünchen team agrees obvisouly. None of the 3 scenarios seems to provide a good solution:

Beil, C., Kutzner, T., Schwab, B., Willenborg, B., Gawronski, A., and Kolbe, T. H.: INTEGRATION OF 3D POINT CLOUDS WITH SEMANTIC 3D CITY MODELS – PROVIDING SEMANTIC INFORMATION BEYOND CLASSIFICATION, ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., VIII-4/W2-2021, 105–112, https://doi.org/10.5194/isprs-annals-VIII-4-W2-2021-105-2021, 2021.

There is still much room for improvement and manipulation.

GANys commented 2 years ago

I proposed a bunch of modifications on the last commit. I followed most of your ideas. It is much complex now but should work in any condition.