Road Extraction from Airborne Lidar Data and Integrated Remote Sensing Data

Download files
Access & Terms of Use
open access
Embargoed until 2017-06-30
Copyright: Liu, Li
Altmetric
Abstract
To extract different types of roads such as arterial roads and local roads and to update the existing road database, a voxel-based skewness and kurtosis balancing algorithm is proposed in this thesis. Vector data is applied to divide lidar points into various tiles. For each tile, a voxel-based skewness and kurtosis balancing algorithm is utilised to extract roads. The initial extraction results are refined by width constraint, spatial interpolation and curve fitting which remove the false positives and join up misclosures. This method is based on the assumption of a normal distribution of the lidar points, and takes little time in parameter-tuning. The test results show an acceptable accuracy and completeness. To deal with areas where vector data is not available, an edge-clustering algorithm is proposed to extract elongated road segments from airborne lidar data and aerial images. Multiple criteria such as height, elevation difference, intensity, band-ratio and length constraint are applied to extract road segments. After the initial extraction, a series of refinements are conducted, including k-Nearest-Neighbouring clustering, searching intersection nodes, spatial interpolation and curve fitting which connect the misclosures and remove the open areas (e.g. parking lots). The results indicate that the proposed approach is practical to extract roads from airborne lidar data and aerial images at an acceptable accuracy. Edge-clustering algorithms often fail to extract complete roads in occluded areas because of the minimum length constraint, which is also validated by the statistics of the undetected road length. To tackle this problem, a colour-space based algorithm is proposed. The main idea of this algorithm is to explore the contextual road information in different colour spaces. Red/Green/Blue channels are transformed into Hue/Saturation/Intensity and Brightness/Blue Difference/Red Difference colour spaces. Various constraints are applied to detect road points. After the extraction, refinements consisting of k-Nearest-Neighbouring clustering, spatial interpolation and curve fitting are applied to connect the misclosures and remove the open areas. The statistics of the tests show better results than the edge-clustering method in terms of the detected road length.
Persistent link to this record
Link to Publisher Version
Link to Open Access Version
Additional Link
Author(s)
Liu, Li
Supervisor(s)
Lim, Samsung
Trinder, John
Creator(s)
Editor(s)
Translator(s)
Curator(s)
Designer(s)
Arranger(s)
Composer(s)
Recordist(s)
Conference Proceedings Editor(s)
Other Contributor(s)
Corporate/Industry Contributor(s)
Publication Year
2015
Resource Type
Thesis
Degree Type
Masters Thesis
UNSW Faculty
Files
download public version.pdf 2.94 MB Adobe Portable Document Format
Related dataset(s)