PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
February 2014
133
Filtering Airborne Lidar Data by
Modified White Top-Hat Transform with
Directional Edge Constraints
Yong Li, Bin Yong, Huayi Wu, Ru An, Hanwei Xu, Jia Xu, and Qisheng He
Abstract
A novel algorithm that employs modified white top-hat
(
MWTH
) transform with directional edge constraints is pro-
posed in this study to automatically extract ground points
from airborne light detection and ranging (lidar) data.
MWTH
transform can effectively distinguish above-ground objects
that are smaller than the window size and higher than the
height difference threshold. Directional edge constraints
significantly decrease omission errors from protruding ground
features. Incorporating
MWTH
transform and directional
edge constraints enables the simultaneous consideration of
the size, height, and edge characteristics of lidar data for
judging above-ground objects. Experimental results verify
that the proposed algorithm exhibits promising performance
and high accuracy in various complicated landscapes, even
in areas with dramatic changes in elevation. The proposed
algorithm has minimal omission and commission error
oscillation for different test sites, thereby demonstrating its
stability and reliability in a wide range of applications.
Introduction
Airborne light detection and ranging (lidar) technology
has become a powerful and popular tool for rapid spatial
data acquisition with acceptable spatial accuracy and large
density (Filin and Pfeifer, 2006; Meng
et al.
, 2009b; Shan
and Sampath, 2005). Lidar can obtain three-dimensional (
3D
)
coordinates of the surface of the Earth in a more convenient
manner compared with traditional photogrammetric and field
surveying methods. Lidar is also unaffected by external light
conditions and requires few ground control points. These in-
comparable merits of lidar have attracted considerable atten-
tion from specialists and scholars in diverse fields. Although
lidar systems have been widely utilized in various practical
applications such as topographic surveying and environmen-
tal planning (Hill
et al.
, 2000; Stoker
et al.
, 2006; White and
Wang, 2003), effectively processing raw data and accurately
extracting useful information still remain a major challenge in
complex situations, especially for areas with steep slopes or
rough surfaces (Chen, 2007).
Raw lidar data sets contain both ground and non-ground
points such as buildings, vegetation, vehicles, and electri-
cal wires. The first important step in digital terrain model
generation and object extraction is to separate obtained point
clouds into ground and non-ground points. This process is
called filtering (Vosselman, 2000; Zhang
et al.
, 2003). Manual
classification and final quality control account for approxi-
mately 60 percent to 80 percent of total lidar data processing
time because no efficient algorithms are available for filtering
(Flood, 2001; Sithole and Vosselman, 2003). Considering the
presence of complex and changeable landscapes in a surveyed
field, lidar data filtering is difficult to automate in computers,
especially for large areas with varying terrain characteristics
(Bartels and Wei, 2010; Silvan-Cardenas and Wang, 2006;
Sithole and Vosselman, 2004; Zhang and Whitman, 2005).
Various approaches have been developed to filter lidar
point clouds in recent decades (Bartels and Wei, 2010; Sil-
van-Cardenas and Wang, 2006; Sithole and Vosselman, 2004;
Zhang and Whitman, 2005). These methods are mostly based
on the assumption that most terrain surfaces have gradual ele-
vation changes, whereas above-ground objects possess abrupt
elevation changes compared with nearby ground (Bretar and
Chehata, 2010). Moreover, the sizes of objects are within
a limited range. Larger above-ground objects usually have
more evident height differences, so a larger height thresh-
old is necessary to filter the larger objects. The slope-based
approach inspects slopes or height differences among nearby
points. A predefined threshold is utilized for filtering based
on the assumption that gradients between ground and non-
ground points are distinctively different (Shan and Sampath,
2005; Sithole, 2001; Vosselman, 2000; Wang and Tseng, 2010;
Wang and Shan, 2009). The morphological approach involves
a series of morphological operations, such as openings and
closings, to separate objects and backgrounds (Chen
et al.
,
2007; Li and Wu, 2009; Petzold
et al.
, 1999; Wu
et al.
, 2010;
Zhang
et al.
, 2003). The surface interpolation approach and
triangular irregular network densification approach iteratively
approximate the ground under strong angle and distance con-
straints (Axelsson, 1999 and 2000; Kraus and Pfeifer, 1998;
Lee and Younan, 2003; Pfeifer
et al.
, 2001; Sohn and Dow-
man, 2002). The directional scanning approach involves the
calculation of slopes and elevation differences along a one-di-
mensional (
1D
) scan line in a specified direction and identifies
ground points based on information along the scan line (Meng
Yong Li and Bin Yong are with the State Key Laboratory of
Hydrology-Water Resources and Hydraulic Engineering,
Hohai University, Nanjing 210098, China
;
).
Huayi Wu is with the State Key Laboratory of Information
Engineering in Surveying, Mapping, and Remote Sensing,
Wuhan University, Wuhan 430079, China.
Ru An, Hanwei Xu, Jia Xu, and Qisheng He are with the
School of Earth Sciences and Engineering, Hohai University,
Nanjing 210098, China.
Photogrammetric Engineering & Remote Sensing
Vol. 80, No. 2, February 2014, pp. 133–141.
0099-1112/14/8002–133
© 2013 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.80.2.133
