Questions & Answers - Assessing Global OpenStreetMap building completeness to generate large-scale 3D city models

Disclaimer: This is an archived version of the once interactive session pad with content from conference attendees. Please note that some information was lost during the transformation (i.e. the edit history, user colours, the chat and some formatting).

Back to session page

DOI: https://doi.org/10.5281/zenodo.3922285

Questions

  1. Wow! Further reading on tools?

    found https://github.com/osmbuildings/  - YaguraStation

    1. (I can’t believe how much awesome stuff is going on, so happy :) - indiebio)
  2. How can we easily estimate building height while mapping? - Anonymous +2
  3. Can you discuss more about the feature importance for your random forest model in your study in the Netherlands? - ardieorden
    1. ANS: number of stories
    2. ANS: floor area
    3. ANS: complexity of footprint
    4. ANS: there’s lots of other predictors, but the concern is how available those are globally
  4. Are cities in the Netherlands very similar for this to work? I could imagine it could be more complicated in countries with old and new cities?
  5. The examples shown all seem to assume a flat ground level.  Have you also tested estimation for cities on hills? +2
    1. Answ: Flat terrain was assumed, as slope is not important in the netherlands. But to transfer, they will look into it. :)
  6. Can you use building shadows from satellite imagery to predict heights as you will have some buildings as control? +2
  7. Thanks very much for this very interesting talk! Have you also thought about elevation estimation from earth observation data, such as: SAR data from Sentinel 1 ? Since Radar data might also provide informaitons about building heights, while might be less accureate than 3D point cloud .
  8. Hello! Thanks and congratulations. Have you tried to apply image-recognition ML models on orthophotos to refine your predictions? I’m specifically thinking of CNN, but any other method also (also see question #6 and #9).
  9. what about estimating building hight with terrain photographs such as Mapillary repository? has it been done? +1
  10. Exeter in the UK has around 35% 3D building coverage of the majority of buildings - it would be a good training set. Combined with other data we could see what works best. See this area for the Met Office as an example: https://demo.f4map.com/#lat=50.7271961&lon=-3.4753034&zoom=19
  11. I  believe most of Netherlands have buildings between 1 and 3 floors. Isn’t the absolute error of 1 floor a bit big for that? – Ilya +2
  12. Did you consider including national regulations on mandated distance spaces into the prediction model?
  13. Isn’t the geometry a bit “fuzzy”? some people map very thoroughly, others map basic footprints.+1 nice clarification :)
  14. Can you explain once again how Landsat Raster is being used to predict completeness? ANS: He said they are not using raster data except for monitoring completeness of data  Yes, 2D footprint completeness. He definitely mentioned that!

Comments

  • Looking forward to see if this project will lead to accurate data additions – Natfoot
  • I saw the OSM user “Kolossos” at a conference. He presented a budget lidar for crowdsourcing 3-D data at his booth. But I think it’s only suitable for small buildings not for skyscrapers.   (I would like to see this talk – Natfoot +1)
  • We may be able to use Mapillary to get building height. – Natfoot
  • Just my opinion, but it’d be really nice to see how ortophotos improve that… Guys thanks for paying so much attention, learning a lot.