Chicago03: Building height distribution structure

3 min readDec 13, 2020

Extracte 1 m high-precision building height data of Chicago from. Las radar data. It is necessary to pay attention to how to deal with a large amount of raster data in order to avoid memory overflow and the size of individual files. Here, several methods are added for geographic information data.

If a single raster file is too large, such as 100g, and cannot be read into memory at one time, it needs to be read in blocks, such as using rasterio.windows The method of the method;

Pay attention to the determination of the storage type of raster data, and select the smallest data storage format, such as byte, int8, uint8, etc;

To increase the computing speed, we can try to use the parallel computing provided by numba library and GPU computing method;

Determine the accuracy allowed to solve the problem, and change the calculation method. For example, the clustering of large amount of data in separate grid may require 100 g memory, so reduce the amount of data or convert the data format to execute.

There are many codes involved in this experiment, and individual file codes can be merged. The code functions are as follows:

Extracte DSM and classification data from. Las point cloud data: pdal_ las_ lidar.py.
Extracte DTM gene from. Las point cloud data: generateDTMBUidingHeight.py
Merge raster data: rasterMosaic_ rasterio.py
Grid interpolation, complete missing data: interpolate2D_ 3D.py
Segmentation of building height and production of individual building spatial distribution data : rasterBuildingHeightZSplit.py
Reclassification method for large amount of data (such as 100g) single raster file（ raster.windows Block reading) : rasterBuildingHeightZSSplit_ reclassify.py
Adjustment of grid clustering and point data clustering: pointsClustering.py / rasterClustering.py

Result

City data + DTM data + building height

Horizontal distribution of buildings divided by continuous longitudinal distance