Classification and change detection of urban areas based on airborne laser scanning data
No Thumbnail Available
URL
Journal Title
Journal ISSN
Volume Title
School of Engineering |
Master's thesis
Authors
Date
2024-12-18
Department
Major/Subject
Geoinformatics
Mcode
Degree programme
Master's Programme in Geoinformatics
Language
en
Pages
71
Series
Abstract
The City of Helsinki procures yearly dense airborne laser scanning point cloud data of its neighbourhoods. This data is important in the upkeep of the three-dimensional city model, and monitoring of the urban landscape. The urban landscape changes quickly even in the span of a year, hence being able to detect larger changes is necessary for city planning. Additionally, the point cloud is part of the open data that the city provides for the public. Therefore, classification of the point cloud at an adequate level of accuracy is important and necessary. The current classification method is semi-automatic requiring significant manual correction, hence finding alternative methods for more automatised approaches would be beneficial. This thesis aims to provide an automatic classification procedure that utilises a random forest classifier in addition to comparing two methods of change detection, geometry-based and object-based. The trained classifier claims to have an overall accuracy of 0.96 and the trees and buildings have F1-scores of 0.98 and 0.96, respectively. The high accuracy values are contrasted by the visual results. The visual results show that the classifier can recall trees and buildings well, however, false positives decrease the reliability of the model. Trees and cars are mistakenly classified as buildings, and cranes, poles and building corners are segmented as trees. Hence, while the process is automatic, further manual correction is required to ensure adequate quality of results. The change detection methods are found to be contrasting. The geometry-based approach is faster since it can process raw point cloud data and can provide a decent overview of the larger changes. It cannot provide information on what has caused the change, unlike the object-based approach which provides labelled information of each change. However, the accuracy of the object-based approach is heavily tied to the accuracy of the classification; a poor classification will result in poor change detection results.Description
Supervisor
Vaaja, MattiThesis advisor
Gröhn, SimoKeywords
point cloud, urban environment, change detection, classification, random forest classifier, machine learning