Automated Extraction of Building Features from LiDAR: Assessment of Software and Industry Capability

Jonah Sullivan(Presenter)

Geoscience Australia, Environmental Geoscience Division

Executive Summary

Light Detection And Ranging (LiDAR) data has become increasingly popular for extracting features (buildings, trees) through improvements in data capture and processing capabilities. GA currently holds LiDAR data covering 2% of Australia’s land area but representing 70% of the built up area where Australians live. To date, most of this data has not been processed to identify buildings. The continuing acquisition of LiDAR data provides an opportunity for GA to capitalise on its diverse application for elevation, hydrology and built environment feature detection.

This research had two main objectives:

  1. to investigate the capability of our current and available in-house tools and the capability of the spatial industry to extract building geometry information from LiDAR data with minimal manual intervention
  2. to understand our current LiDAR and supplementary datasets and the suitability for feature extraction

A study area of inner Launceston, Tasmania was chosen due a recent acquisition with good vertical accuracy (95% of values within ±15cm) and a high point density (≈4 points per square metre) as well as corresponding imagery and reference data to validate the results.

Five software packages were assessed for their ability to identify building features within the unclassified and/or classified (level 3) LiDAR point cloud and extract a geometrically correct polygon representation of individual building’s roof outline and capture height values of individual building floor, eave and maximum building height.

Five service providers carried out the same task of extracting building roof-outlines and their floor, eave and maximum building height. The providers were contracted based on their specified methodology, potential outcomes and price. Having a choice of using classified or unclassified LiDAR data, the service providers had a month to complete the work using their preferred software and techniques.

Feature and area-based assessment methods were developed to assess the spatial output of software packages and service providers against a reference dataset provided by the Launceston City Council. Height attributes were assessed comparatively against each other. There was particular emphasis on the amount of processing time and manual intervention required when determining the most effective methodology.