Structuralization of LiDAR Point Cloud
Shing-Hung Lin* Jen-Jer Jaw**
Department of Civil Engineering, National Taiwan University
1 Roosevelt Rd. Sec. 4, Taipei, 10617, Taiwan, R.O.C.
**Tel: 886-2-23678645, **Fax: 886-2-23631558
*Email: poa@vip.url.com.tw,
**jejaw@ce.ntu.edu.tw
ABSTRACT Inherited from the data acquisition nature, LIDAR point cloud are of random
distribution rendered with geometric 3-D coordinates as well as radiometric intensities. To infer
from LIDAR point cloud for explicit information in object space, further processes must be
carried out. The authors investigated in this study by exploiting the concept of structuralization
on LIDAR point cloud where the following subjects were targeted: (1) tabling the LIDAR point
cloud attributed with geometric, radiometric information and topologic relationship of the points
within near neighborhood; (2) classifying the point cloud into planar and isolated points by
iterative validation through geometric and radiometric inference; (3) outlining the plane
boundary; (4) deriving and localizing 3-D line and corner features by intersecting neighboring
planes. The resultant features, including planes, 3-D lines, and corners from the proposed
processing procedures are advantageous for the purposes of object reconstruction and other
photogrammetric applications.
1 INTRODUCTION
LIDAR system emerging as an effective tool for 3-D data acquisition in photogrammetric
community can be originated from the research of Lindenberger (1993) who investigated and
implemented a laser-profiling system for topographic mapping. The technology soon advances
to LIDAR systems which allow rapid 3-D point cloud acquisition with along and across the
flying track within the past decade. With the direct geo-referencing and canopy penetrating
capabilities, airborne and ground-based LIDAR systems are favored in versatile applications,
including coast and power line mapping, forest inventory, digital elevation models (DEM) and
digital surface models (DSM), 3-D city modeling, and being an effective part of hazards
mitigation systems, etc. Due to the discrete nature of LIDAR point cloud, certain operations
need to be processed before any object features can be identified and thus utilized. Processing
LIDAR point cloud by the concept of structuralization becomes a hot research topic for
extracting higher level features, such as 3-D lines, planes, and objects. An excellent review of
structuralization can be referred to Sarkar and Boyer (1993) who later gave an insight into 3-D
data structuralization. (Boyer and Sarkar, 1999). Lee (2002) employed “Perceptual
Organization” for discriminating objects from ground structures. A recent work of segmenting
and classifying LIDAR data by using octree structure can be referred to Wang and Tseng (2004).
In this study, the authors aims at extracting 3-D features from LIDAR point cloud by following
the procedures of patch growing, plane hypothesizing, plane validation, planes grouping, and
3-D line intersection. The processing chain iterates itself until no further information can be
verified. An attribute table offering geometric inference and verification is the key component to
the success of data processing.
