Extracting OSM building heights from LiDAR

[Martin Isenburg]

Hello,

Most OpenStreetMap building footprints do not have building heights.
But those are crucial data for everything from disaster response to
measuring economic growth. The blog article by Yuriy Czoli below
demonstrates how to create a building-height footprint using open
LiDAR data and several open source tools:

http://developmentseed.org/blog-2014/blog/2014/08/07/processing-lidar-point-cloud/
http://gist.github.com/YKCzoli/3605e014b8ed09a571e5

Can you replicate this workflow with LAStools using a simpler pipeline
that avoids copying the data into a spatial data base? The strategy
would be to first height-normalize the LAZ tiles using lasground
and/or lasheight and then run the all new lascanopy with a '-lop
buildings.shp' list of building footprints as input and compute the
maximum, the average, and maybe the 90th or 95th height percentile
with options '-max -avg -p 90 95' and output the results to a CSV file
with '-ocsv' from where it can be added as a new attribute to the
OSM buildings. Is anyone interested to try to realize such a
pipeline?

Regards,

Martin @rapidasso

--
http://rapidlasso.com - fast tools to heighten your buildings

[Jonah Sullivan]

I used the method given in the attached document to assign heights to building footprints, with minor variation.

The goal was to get three attributes per structure: ground floor height (above geoid), gutter height (above ground), and maximum height (above ground).

The assumption I used is that the buildings are not rectangular prisms or cubes, but extruded pyramids (residential buildings generally have slanted roofs).

I converted all points within each building footprint into XYZ files and then calculated the minimum (ground floor height) and maximum - minimum (maximum height). For the gutter height I made a 1-metre interior buffer and then averaged the height of those returns within the buffer.

I think I could have used LAZ files instead of text files but the text files are easy to read.

If I were to re-do the analysis I would use LAScanopy.