In: Wagner W., Sz^kely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
281 
URBAN VEGETATION DETECTION USING HIGH DENSITY FULL-WAVEFORM 
AIRBORNE LIDAR DATA - COMBINATION OF OBJECT-BASED IMAGE AND POINT 
CLOUD ANALYSIS 
B. Hofle 3, *, M. Hollaus b 
a University of Heidelberg, Department of Geography, 69120 Heidelberg, Germany - hoefle@uni-heidelberg.de 
b Vienna University of Technology, Institute of Photogrammetry and Remote Sensing, 1040 Vienna, Austria - 
mh@ipf. tuwien. ac. at 
KEY WORDS: Airborne LiDAR, Vegetation mapping, Open source GIS, Object-Based Image Analysis 
ABSTRACT: 
In this paper, a new GIS workflow for fully automated urban vegetation and tree parameter extraction from airborne LiDAR data is 
presented. The strengths of both raster- and point cloud-based methods are combined, in order to derive a vegetation map layer as 
well as single tree parameters (e.g. tree height and crown width) in an efficient way. The workflow is implemented in GRASS GIS 
making use of standard GIS functionality and newly developed tools providing access to point cloud analysis. An edge-based 
segmentation delineates potential tree crowns, which are further aggregated to single trees or group of trees by using local 
topological information (e.g. percentage of outline touched by neighbors) and constraints on segment geometry (e.g. shape of 
segments). Furthermore, the classification makes use of segment attributes that have been extracted from the full-waveform point 
cloud (e.g. percentage of first echoes, echo width and signal amplitude). A representative study area in the City of Vienna is used to 
demonstrate the applicability of the developed object-based GIS workflow. Buildings and vegetation objects could be separated with 
high accuracy, where at maximum 14% of classified vegetation segments confuse with buildings (mainly building edges). 
Concluding, the unique high density (50 pts/m 2 ) full-waveform LiDAR data open a new scale in 3D object extraction but demanding 
for novel strategies in object-based raster and point cloud analysis. 
1. INTRODUCTION 
Vegetation monitoring and tree inventory play an important role 
in modem urban spatial data management, as many benefits and 
applications inherit from this detailed up-to-date data source 
such as monitoring of functions (e.g. noise and pollution 
mitigation) and creation of 3D city models (Vosselman, 2003). 
Compared to predominant studies on vegetation detection and 
characterization mainly in purely forested areas (e.g. Hyyppa et 
al., 2001), this study concentrates on urban areas, which have a 
high structural complexity with a multitude of different objects 
(e.g. temporary objects, vegetation on top of buildings, road 
signs, power lines and cables). Previous studies using airborne 
LiDAR for vegetation and single tree detection, respectively, 
used image-based methods, e.g. including orthophotos 
(Hirschmugl et al., 2007; Iovan et al., 2007; Secord and Zakhor, 
2007) and derived raster layers e.g. first-last-pulse difference 
(Liang et al., 2007). Novel approaches include full-waveform 
(FWF) information (e.g. echo width and amplitude) for urban 
object detection in the 3D point cloud directly (Mallet et al., 
2008; Rutzinger et al., 2007 and 2008) and for 3D segmentation 
of single trees (Reitberger et al., 2009). 
Hofle et al. (2008) showed the suitability of aggregated FWF 
attributes attached to segments derived from image analysis for 
tree species discrimination. This paper aims at transferring this 
combined object-based image (Blaschke, 2010) and point cloud 
analysis from forested to complex urban areas, in order to 
derive a GIS-ready vegetation mask. To date, high point density 
(~50 echoes/m 2 ) LiDAR data are increasingly available in 
particular for urban areas. This high point density does not 
allow to apply computationally intensive 3D point cloud 
analysis in an operational manner, which is a prerequisite for 
being used in city map and cadastre generation. Thus, this study 
presents a novel GIS framework for full-waveform LiDAR data 
land cover classification, making use of both image and point 
cloud analysis (cf. Hofle et al., 2009). 
2. STUDY AREA AND DATASETS 
The study area is located in the city centre of Vienna, Austria, 
and comprise the three city parks: Rathauspark, Volksgarten 
and Burggarten (Fig. 1). A great variety of planting and tree 
species can be found within the test site, e.g. alley of trees, 
short-cut trees, hedgerows and shrubs. Deciduous trees are 
predominant, e.g. beech (Fagus sylvatica), Norway maple (Acer 
platanoides), plane (Platanus acerifolia), linden (Tilia cordata, 
platyphyllos) and chestnut (Aesculus hippocastanum) and 
sparsely coniferous species. The area is characterized by large 
building complexes (e.g. city hall, Burgtheater and parliament). 
Furthermore, artificial objects such as fences, cars, power lines, 
park benches as well as a high amount of people are usually 
present in this central part of the city, and therefore also 
included in the airborne LiDAR elevation datasets. 
* Corresponding author.