University of Twente Student Theses


3D tree modelling using mobile laser scanning data

Pratihast, Arun Kumar (2010) 3D tree modelling using mobile laser scanning data.

[img] PDF
Abstract:Vegetation is an important component in an urban environment. In recent years, technological advancement has offered opportunity for virtual city modelling. Incorporation of environmental components in the model enables better planning and decision making. In this context, integration of 3D tree models adds a more complete and realistic view in virtual city modelling. Mobile laser scanning (MLS), is an active technique to capture highly dense 3D point cloud of larger urban areas in a rapid and cost effective way(Norbert et al., 2008). The acquired point cloud is highly suitable for the extraction of 3D cadastre information, urban objects inventory and 3D city modelling. Urban environments are a mixture of heterogeneous objects like poles, traffic signs, buildings and trees. Thus, the tree delineation technique developed for forestry applications using airborne laser scanning (ALS) is not directly applicable. Attempts made so far using highly dense point cloud to model the tree are however, comparatively unrealistic, labour-intensive and time consuming. In this research, a fully automated workflow for single tree modelling using MLS point clouds is proposed. The workflow starts with pre-processing detection of tree point cloud from a dense mixture of urban objects. After this, the structure of tree point cloud is simplified by applying a 3D alpha shape algorithm. In the remaining point cloud, connected groups of trees are separated and detected by analysing the structure of the canopy and the appearance of tree stems if they are visible. After labelling laser echoes belonging to a single tree, the tree model parameters are derived. The minimum required model parameters, which are derived from the separated alpha shape point clouds, are tree height, base height, stem diameter, crown length, width and crown shape. The obtained parameters are used to generate the approximate model of the tree. 3D file format of the model is developed and finally exported to appropriate 3D environment. The quality of the tree models is tested as a function of data reduction and shape simplification by applying different alpha values. Furthermore the realistic appearance of the models is checked against acquired photographs. Performance of the workflow is evaluated in terms of tree detection and data reduction rate. The overall quality of tree detection is achieved more than 80%. The result shows that the developed modular structure of workflow reduces more than 80% points during the pre-processing and more than 90% points during the 3D alpha shape generation without loosing the important information. This result concludes that the presented workflow is applicable for large data set of varying point density. Keywords: 3D tree model, MLS, point cloud, 3D alpha shape
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page