Please use this identifier to cite or link to this item:
Title: Outlier detection and data filtering in LiDAR data with multiple attributes
Authors: Pang, Gang
Degree: M.Phil.
Issue Date: 2011
Abstract: Outlier detection and data filtering are two research topics in the area of LiDAR data processing, and have attracted lots of research attentions in recent years. The former one is considered as an essential preprocessing step for overall LiDAR data filtering and modeling, while, the latter one is necessarily required in the step of digital elevation model (DEM) generation. However, both the two issues always face great challenges in the automatic data processing. For outlier detection, it has proven to be surprisingly difficult to automatically remove low outliers in form of clusters. While, for data filtering, it has also suffered from great difficulties, especially in urban areas. Literature reviews demonstrate that most existing algorithms for both issues are mainly focusing on the analysis of single attribute: either the height or spatial neighborhood relationship information for outlier detection, and only geometrical information for data filtering. However, since the charactertics of outliers in LiDAR data are both single points and also clusters with elevations, either much higher or lower than the surrounding points, to effectively remove both types of outliers, it is necessary to analysis both the height and spatial neighborhood relationship information together. In parallel, since the LiDAR system simultaneously provides not only geometric information which mainly refers to height data but also radiometric information which mainly refers to intensity data, both of the two data describe the same features geometrically, to separate terrain points and off-terrain points, it is suggested that the comprehensive utilization of both two data may be advantageous over using either data individually. Thus, to fit the requirements of multiple attributes data processing for both issues, the minimum covariance determinant (MCD) based multiple attributes model is proposed in this study which extends traditional data processing methods from single attribute to multiple attributes, from one dimension to multiple dimensions. Specially, for outlier detection, we define the connectivity based outlier factor (COF) which indicates the spatial neighborhood relationship of a point to its neighbors as an attribute; then the COF attribute and the height attribute are extracted to organize a 2-D space, in the formed 2-D space, the proposed MCD-based multiple attributes model is conducted to identify outliers. Two typical experimental data are used of evaluating the performance of the proposed method. Comparative results by using the COF, Height, the proposed "COF + Height" and other eight representative algorithms are generated and analyzed. The result shows that the newly developed method can highly detect most outliers in both forms: individual and cluster. For data filtering, similar as removing outliers, both the height and intensity attribute are extracted to organize a 2-D space; in the formed 2-D space, the proposed model is conducted to separate terrain points and off-terrain points. Typical experimental data are utilized for checking the performance of the proposed method. Both quantitative and qualitative assessments of the results are carried out. By comparing with eight representative methods at the ISPRS filter test, it shows that our method is fair by minimizing the Type II error. In which, Type II error in our method ranks at about top 3 of every sample region with others, and simultaneity, Type I error and Total error ranks at a middle level.
Subjects: Remote sensing.
Digital mapping.
Hong Kong Polytechnic University -- Dissertations
Pages: xi, 150 p. : ill. ; 30 cm.
Appears in Collections:Thesis

Show full item record

Page views

Last Week
Last month
Citations as of May 28, 2023

Google ScholarTM


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.