See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/313959376

A REVIEW OF POINT CLOUDS SEGMENTATION AND CLASSIFICATION

ALGORITHMS

Article in The International Archives of the Photogrammetry Remote Sensing and Spatial Information Sciences · February 2017
DOI: 10.5194/isprs-archives-XLII-2-W3-339-2017

CITATIONS READS

421 9,526

3 authors:

Eleonora Grilli Fabio Menna

Fondazione Bruno Kessler University of Sassari
27 PUBLICATIONS 1,352 CITATIONS 146 PUBLICATIONS 4,496 CITATIONS

SEE PROFILE SEE PROFILE

Fabio Remondino
Fondazione Bruno Kessler
512 PUBLICATIONS 20,838 CITATIONS

SEE PROFILE

All content following this page was uploaded by Fabio Menna on 06 March 2017.

The user has requested enhancement of the downloaded file.

 The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W3, 2017
3D Virtual Reconstruction and Visualization of Complex Architectures, 1–3 March 2017, Nafplio, Greece

A REVIEW OF
POINT CLOUDS SEGMENTATION AND CLASSIFICATION ALGORITHMS
E. Grilli, F. Menna, F. Remondino

3D Optical Metrology (3DOM) unit, Bruno Kessler Foundation (FBK), Trento, Italy
<grilli><menna><remondino>@fbk.eu

Commission II

KEY WORDS: Point Clouds, Segmentation, Classification, Photogrammetry, Laser Scanning

ABSTRACT:
Today 3D models and point clouds are very popular being currently used in several fields, shared through the internet and even accessed
on mobile phones. Despite their broad availability, there is still a relevant need of methods, preferably automatic, to provide 3D data
with meaningful attributes that characterize and provide significance to the objects represented in 3D. Segmentation is the process of
grouping point clouds into multiple homogeneous regions with similar properties whereas classification is the step that labels these
regions. The main goal of this paper is to analyse the most popular methodologies and algorithms to segment and classify 3D point
clouds. Strong and weak points of the different solutions presented in literature or implemented in commercial software will be listed
and shortly explained. For some algorithms, the results of the segmentation and classification is shown using real examples at different
scale in the Cultural Heritage field. Finally, open issues and research topics will be discussed.

1. INTRODUCTION Laser Scanner and InSAR Data”’ with the aim to segment and
classify points in bare earth and object classes (Sithole &
We are recently witnessing an increasing availability of not- Vosselman, 2003; Sithole & Vosselman, 2004). The study was
interpreted point clouds and 3D models, often shared online using initiated to compare the performance of various automatic filters
point-based rendering solutions (e.g. PoTree) of mesh-based with the purpose of (i) determining the comparative performance
portals (e.g. Sketchfab). If we focus on point clouds, there is a of existing filters, (ii) understanding the influence of point
growing need of innovative methods for the treatment and density on the filter performance and (iii) identifying directions
analysis of these data and for their classification, aimed for future research on point clouds filtering algorithms. A more
ultimately to exploit in-depth the informative value of these recent benchmark is the “Large-Scale Point Cloud Classification
surveys and representations. 3D point clouds are the simplest but Benchmark” (www.semantic3d.net) that provides labelled
at the same time powerful collection of elementary geometrical terrestrial 3D point cloud data on which people can test and
primitives able to represent shape, size, position and orientation validate their algorithms (Fig. 1).
of objects in space. This information may be augmented with
additional contents obtained from other sensors or sources, such
as colours, multispectral or thermal information, etc. For a
successful exploitation of point clouds and to better understand
them, we must first proceed with segmentation and classification
procedures. The former refers to group points in subsets
(normally called segments) characterized by having one or more
characteristics in common (geometric, radiometric, etc.) whereas
classification means the definition and assignment of points to
specific classes (“labels”) according to different criteria.
Due to the complexity and variety of point clouds caused by
irregular sampling, varying density, different types of objects,
etc., point cloud classification and segmentation are very active Figure 1: Example of a segmented and classified point cloud
research topics. There are multiple research studies related to (www.semantic3d.net).
these two topics, many driven by specific needs provided by the
field of application (building modelling, Heritage documentation
and preservation, robotics, etc.). 2. SEGMENTATION
Most of the segmentation algorithms are tailored to work with a
2.5D surface model assumption, coming for example from a A first attempt to group segmentation methods follows the works
LiDAR-based survey. Many algorithms require a fine-tuning of of Sapkota (2008) and Nguyen (2013) and a schematic
different parameters depending upon the nature of data and representation is shown in Figure 2.
applications. Supervised methods are the majority with a training
phase mandatory and fundamental to guide the successive 2.1 Edge-based segmentation
machine learning classification solution. Some of the techniques As described by Rabbani et al. (2006), edge-based segmentation
developed for segmenting point clouds generated from airborne algorithms have two main stages: (i) edge detection to outlines
laser scanning can be applied or easily adapted to terrestrial point the borders of different regions and (2) grouping of points inside
clouds. The results are generally affected by noise and density of the boundaries to deliver the final segments. Edges in a given
the cloud as well as by the quality of the training data. depth map are defined by the points where changes in the local
Different benchmarks were proposed in the research community, surface properties exceed a given threshold. The mostly used
the most comprehensive study being carried out by the previous local surface properties are normals, gradients, principal
ISPRS Working Group III/3 “3D Reconstruction from Airborne curvatures or higher order derivatives. Methods based on edge-

This contribution has been peer-reviewed.

doi:10.5194/isprs-archives-XLII-2-W3-339-2017 339
 The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W3, 2017
3D Virtual Reconstruction and Visualization of Complex Architectures, 1–3 March 2017, Nafplio, Greece

based segmentation techniques are reported by Bhanu et al.

(1986), Sappa and Devy (2001), Wani and Arabnia (2003).
Although such methods allow a fast segmentation, they may
produce not accurate results in case of noise and uneven density
of point clouds, situations that commonly occur in point cloud
data. In 3D space, such methods often detect disconnected edges
making the identification of closed segments difficult without a
filling or interpretation procedure (Castillo et al., 2013).

Figure 3: Point cloud segmented with a region growing algorithm

available in the Point Cloud Library (PCL).

In general, the region growing methods are more robust to noise

than the edge-based ones because of the using of global
information (Liu and Xiong, 2008). However, these methods are
sensitive to (i) the location of initial seed regions and (ii)
inaccurate estimations of the normals and curvatures of points
near region boundaries.

2.3 Segmentation by model fitting

This approach is based on the observation that many man-made
objects can be decomposed into geometric primitives like planes,
Figure 2: Synthetic representation of the segmentation methods. cylinders and spheres (Fig. 3). Therefore, primitive shapes are
fitted onto point cloud data and the points that conform to the
2.2 Region growing segmentation mathematical representation of the primitive shape are labelled
as one segment. As part of the model fitting-based category, two
These methods start from one or more points (seed points) widely employed algorithms are the Hough Transform (HT)
featuring specific characteristics and then grow around (Ballard, 1981) and the Random Sample Consensus (RANSAC)
neighbouring points with similar characteristics, such as surface approach (Fischer and Bolles, 1981). In case the primitives have
orientation, curvature, etc. (Rabbani et al., 2006; Jagannathan and some semantic meaning, then such approach is also performing a
Miller, 2007). Region-based methods can be divided into: classification.
 Bottom-up approaches: they start from some seed points and
grow the segments on the basis of given similarity criteria.
Seeded region approaches are highly dependent on selected
seed points. Inaccurate selection of seed points will affect the
segmentation process and can cause under- or over-
segmentation results.
 Top-down approaches: they start by assigning all points to
one group and then fit a single surface to it. Where and how
to subdivide unseeded-region remain the main difficulty of
these methods.
Region-based algorithms includes two steps: identification of the
seed points based on the curvature of each point and growing
them based on predefined criteria such as proximity of points and
Figure 4: Segmentation of 3D point cloud by geometric primitive
planarity of surfaces. The initial algorithm was introduced by
fitting.
Besl et al. (1988) and then several variations were presented in
the literature. The region growing method proposed by
The HT is used to detect planes (Vosselman et al., 2004),
Vosselman et al. (2004) has introduced the use of colour
cylinders and spheres (Rabbani et al., 2006). The RANSAC
properties beside geometrical criteria. Surface normal and
method is used to extract shapes by randomly drawing minimal
curvatures constraints were widely used to find the smoothly
data points to construct candidate shape primitives. The
connected areas (Klasing et al., 2009; Belton and Lichti, 2006)
candidate shapes are checked against all points in the dataset to
whereas Xiao et al. (2013) proposed to use sub window as the
determine a value for the number of the points that represents the
growth unit. Ackermann and Troisi (2010) used a region growing
best fit.
approach to segment planar pitched roofs in 3D point clouds for
Tarsha-Kurdi et al. (2007) compared RANSAC and 3D Hough
automatic 3D modelling of buildings. Anh-Vu Vo et al. (2015)
transform for automatically detect roof planes from LiDAR-
presented an octree-based region growing approach for a fast
based point clouds. Despite the limitation encountered in both
surface patch segmentation of urban environment 3D point
methods, RANSAC resulted the more efficient considering
clouds.
segmented results and running time. It can process a large amount
A collection of region growing algorithms is available in the
of input data in negligible time. On the other hand, 3D HT
Point Cloud Library (http://pointclouds.org). Figure 2 shows the
resulted slower and more sensitive to the segmentation
results of a segmentation done by a region growing algorithm
parameters values.
implemented in the pcl::RegionGrowing class. The purpose of
A popular strategy in reverse engineering involves locally fitting
the algorithm is to merge/join similar points and deliver a set of
primitives like planes, cylinders, cones using RANSAC based
clusters with points belonging to the same smooth surface.
methods (Schnabel et al. 2009).

This contribution has been peer-reviewed.

doi:10.5194/isprs-archives-XLII-2-W3-339-2017 340
 The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W3, 2017
3D Virtual Reconstruction and Visualization of Complex Architectures, 1–3 March 2017, Nafplio, Greece

Chen et al (2014) has proposed a modified RANSAC grouping is done by minimizing the sum of squares of distances
segmentation algorithm that is less sensitive to noise, maintains between point and the corresponding cluster centroid. The
topological consistency, and avoids over and under-segmentation original K-means algorithm presented by MacQueen et al. (1967)
of building primitives, through the localized sampling to segment was then exploited for point clouds by various researchers
the polyhedral rooftop primitives and then through the (Comaniciu and Meer, 2002; Lavoue ́ et al., 2005; Yamauchi et
application of a region growing based triangulated irregular al., 2005, Zhang et al., 2008; LeCun et al., 2015).
network (TIN) to separate the coplanar primitives.
Several extensions are available within the Point Cloud Library: 2.5.2 Segmentation based on hierarchical clustering
 MLESAC (Maximum Likelihood Estimation SAmple and These methods compute representative measures (features) for
Consensus) each (3D) point, based e.g. on geometrical and radiometric
 MSAC (M-estimator SAmple and Consensus) characteristics: point position, locally estimated surface normals,
 PROSAC (Progressive Sample and Consensus) residuals of best fitting surface procedures, points reflectance,
Model fitting methods are fast and robust with outliers. Their etc. They usually create a hierarchical decomposition of a dataset
efficiency for the 3D detection of geometrically simple by iteratively splitting the dataset into smaller subsets until each
parameterized shape such as cylinders, spheres, cones, torus, subset consists of only one object (Ng and Han, 1994). Xiaohu
planes and cubes has been proven, providing an efficient shape Lu et al. (2016) recently presented a novel hierarchical clustering
descriptor with insight over the geometrical properties of a point algorithm which clusters any dimensional data and can be applied
cloud sample. As it falls short for complex shapes or fully to mobile mapping, aerial and terrestrial point clouds.
automated implementations, the use of the richness of surface
geometry through local descriptors provide a better solution
(Poux et al., 2016). In the architectural field, details cannot 3. CLASSIFICATION
always be modelled into easily recognisable geometrical shapes.
Thus if some entities can be characterized by geometric Once a point cloud has been segmented, each segment (group) of
properties, others are more readily distinguished by their colour points can be labelled with a class thus to give some semantic to
content (Barnea and Filin, 2013). the segment (hence point cloud classification is often called
semantic segmentation or point labelling). Point cloud (or mesh)
2.4 Hybrid segmentation technique classification is gaining interest and becoming a very active field
of research (Weinmann et al., 2013; Guo et al., 2014; Niemeyer
In kind of approach, more than one method is combined, e.g. to et al., 2014; Schmidt et al., 2014; Weinmann et al., 2014; Xu et
detect planar segments, in order to exploit the strength a method al., 2014; Hackel et al., 2016). The class labelling procedure is
and bypass the weakness of other methods (Vieira and Shimada, normally achieved following three different approaches:
2005; Lavoue ́ et al., 2005; Benko and Varady, 2004). The  a supervised approach, where semantic categories are
success of these hybrid methods depends on the success of the learned from a dataset of annotated data and the trained
underlying methods. model is used to provide a semantic classification of the
entire dataset. A large amount of annotated data is normally
2.5 Machine learning segmentation mandatory to train the model.
 an unsupervised approach, where the data is automatically
Some segmentation algorithms are based on machine learning partitioned into segments based on a user-provided
methods, such as hierarchical clustering, K-means or mean shift. parameterization of the algorithm. No annotations are
Machine learning (including deep learning, neural network, etc.) requested but the outcome might not be aligned with the
is a scientific discipline concerned with the design and user’s intention.
development of Artificial Intelligence algorithms that allow  an interactive approach, where the user is actively involved
computers to take decisions based on empirical and training data. in the segmentation/classification loop by guiding the
A learner can take advantage of examples (data) to capture and extraction of segments via feedback signals. This requires a
infer characteristics of interest of their unknown underlying large effort from the user side but it could adapt and improve
probability distribution. Data can be seen as examples that the segmentation result based on the user’s feedback.
illustrate relations between observed variables. In machine As suggested in Weinmann et al. (2015) most of the approaches
learning, unsupervised learning is a class of problems in which for point cloud classification consider the different components
one seeks to determine how the data are organized. It is of the classification process (i.e. neighbourhood selection,
distinguished from supervised learning (and reinforcement feature extraction and classification) independently from each
learning) as they rely on a set of provided training examples other. However, it would be desirable to connect these
(features) to learn how to correctly perform a task. Features play components by sharing the results of crucial tasks across all of
a fairly important role in these problems and their definition is them. Such a connection would not only be relevant for the
one of the bottleneck of machine learning methods. Features with interrelated problems of neighbourhood selection and feature
high quality can simplify learning models to interpret the models extraction, but also for the question of how to involve spatial
more easily and enhance algorithm performance with respect to context in the classification task.
both the speed and the accuracy. Weinmann et al. (2015)
discusses the suitability of features that should privilege quality
over quantity. This shows a need to prioritize and find robust and 4. APPLICATIONS IN THE CULTURAL HERITAGE
relevant features to address the heterogeneity in a point cloud. FIELD
Machine learning methods are quite robust and flexible, however
they rely on the point cloud density and normally feature very Processes like segmentation and classification can be apply at
long computational time. different scales in the field of Cultural Heritage, from an entire
archaeological sites to a small artefact. In the following sections
2.5.1 Segmentation based on K-means clustering some examples are reported and commented.
It is a method based on an algorithm able to classify or to group
set of (3D) points into K groups using attributes/features. The

This contribution has been peer-reviewed.

doi:10.5194/isprs-archives-XLII-2-W3-339-2017 341
 The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W3, 2017
3D Virtual Reconstruction and Visualization of Complex Architectures, 1–3 March 2017, Nafplio, Greece

Figure 5: 3D point cloud of the archaeological site of Paestum (left) and segmentation to separate bare earth and man-made structures (right).

Figure 6: The digitized and textured Neptune temple in Paestum and the classification of its archaeological elements (Fiorillo et al., 2013).

4.1 Segmentation with a Prismatic Buffered TIN method and, employing a probabilistic approach, the points with high
uncertainty can be removed from the wrong class.
The Prismatic Buffered Triangular Irregular Network (PBTIN) Farella (2016) used the same classification algorithm for the
method presented in Menna & Troisi (2007) was applied to separating man-made structures from natural elements (trees,
separate man-made structures from bare earth in the UAV-based bushes, etc.) in point clouds of military fortifications built during
dense point cloud of the entire archaeological site of Paestum the First World War (WWI) on the hills around Trento, Italy (Fig.
(Italy). The method is based on an iterative densification of a TIN 8).
built on some initial seed points belonging to the class bare earth.
As shown in Figure 5 the algorithm was used to separate the
5. CONCLUSIONS
terrain from any man made structure and vegetation standing off
the bare earth. This segmentation method is better suited for The paper presented a review of point clouds segmentation and
structured point clouds while for unstructured point clouds it classification algorithms. The review is probably not fully
suffers of noise, outliers and inhomogeneous point densities. exhaustive but it reports many approaches suitable for the
geospatial and heritage communities. The key difference among
4.2 Segmentation with primitive fitting most segmentation approaches is the method or criterion used to
Digital copies of large and complex monuments could be hardly measure the similarity between a given set of points and hence
accessible and manageable by non-experts and policy makers. for making the grouping decisions. All methods were grouped in
Segmentation and classification procedures could facilitate five categories based on their core approach.
understanding, interpretation, management and access to According to what the user’s needs and data type, the correct
complex information. In Fiorillo et al. (2013) semantic algorithm must be chosen. Many segmentation algorithms, for
knowledge was manually added to a 3D model of a Greek-Roman example, are tailored to work with structured or LiDAR data and
temple to allow a better understanding and management of the thus are simply not applicable to other types of point clouds.
heritage object (Fig. 6). Other methods work with RGB information, but not every point
cloud is always coloured. Many solutions rely on a large set of
4.3 Machine learning segmentation parameters which need to be fine-tuning for every processed
dataset. Therefore it’s impossible to make a generalization and
The Canupo segmentation algorithm (Brodu and Lague, 2012) decide which approach is the best as each dataset is distinct and
implemented in CloudCompare was used to separate vegetation needs to be treated as such. Although advances are being made
and stones on the Seiano’s gate – the imperial entrance to the in the field of segmentation and classification, particularly with
archaeological site of Pausilypon (Naples, Italy) (Fig. 7). The machine learning methods, most of the processing steps still
Canupo plug-in allows to create own classes as well as to use requires human intervention to achieve satisfactory results.
existing classifiers for segmenting point clouds into subsets (e.g.
vegetation, ground, etc.). This supervised method is based on 3D
geometrical properties of the point cloud across multiple scales

This contribution has been peer-reviewed.

doi:10.5194/isprs-archives-XLII-2-W3-339-2017 342
 The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W3, 2017
3D Virtual Reconstruction and Visualization of Complex Architectures, 1–3 March 2017, Nafplio, Greece

Figure 7: Supervised point cloud segmentation of the Seiano’s Gate point cloud (red: vegetation; light blue: stone).

Figure 8: Point cloud classification to distinguish man-made military structures (red) from vegetation areas (grey) (Farella, 2016).

REFERENCES classification of complex natural scenes using a multi-scale

dimensionality criterion: applications in geomorphology.
Ackermann S, Troisi S, 2010. Una procedura di modellazione International Archives of the Photogrammetry, Remote Sensing
automatica degli edifici con dati LIDAR. Bollettino SIFET, Vol. and Spatial Information Sciences, Vol. 68, pp.121-134.
2, pp. 9-25.
Castillo, E., Liang, J., & Zhao, H., 2013. Point cloud
Aitelkadi, K., Tahiri, D., Simonetto, E., Sebari, I., & Polidori, L., segmentation and denoising via constrained nonlinear least
2013. Segmentation of heritage building by means of geometric squares normal estimates. In Innovations for Shape Analysis (pp.
and radiometric components from terrestrial laser scanning. 283-299). Springer Berlin Heidelberg.
ISPRS Annals of Photogrammetry, Remote Sensing and Spatial
Information Sciences, Vol. 1(1), pp. 1-6. Chen, D., Zhang, L., Mathiopoulos, P. T., & Huang, X., 2014. A
methodology for automated segmentation and reconstruction of
Ballard, D. H., 1991. Generalizing the Hough transform to detect urban 3-D buildings from ALS point clouds. IEEE Journal of
arbitrary shapes. Pattern Recognition, Vol. 13(2), pp. 183-194. Selected Topics in Applied Earth Observations and Remote
Sensing, Vol. 7(10), pp. 4199-4217.
Barnea, S., & Filin, S., 2013. Segmentation of terrestrial laser
scanning data using geometry and image information. ISPRS J. Comaniciu, D. and Meer, P., 2002. Mean shift: A robust ap-
of Photogrammetry and Remote Sensing, Vol. 76, pp. 33-48. proach toward feature space analysis. IEEE Transactions on
Pattern Analysis and Machine Intelligence, Vol. 24(5), pp. 603-
Belton, D. and Lichti, D. D., 2006. Classification and
619.
segmentation of terrestrial laser scanner point clouds using local
variance information. International Archives of Farella, E.M, 2016. 3D Mapping of underground environments
Photogrammetry, Remote Sensing and Spatial Information with a hand-held laser scanner. Proc. SIFET annual conference,
Sciences, Vol. 36(5), pp. 44-49. Lecce, Italy.
Benkő, P., & Várady, T., 2004. Segmentation methods for Fiorillo, F., Fernández-Palacios, B. J., Remondino, F., & Barba,
smooth point regions of conventional engineering objects. S., 2013. 3D Surveying and modelling of the Archaeological
Computer-Aided Design, Vol. 36(6), pp. 511-523. Area of Paestum, Italy. Virtual Archaeology Review, Vol. 4(8),
pp. 55-60.
Besl P.J., Jain R.C., 1988. Segmentation through variable order
surface fitting. IEEE Transaction on Pattern Analysis and Fischler, M. A., & Bolles, R. C., 1981. Random sample
Machine Intelligence, Vol. 10. consensus: a paradigm for model fitting with applications to
image analysis and automated cartography. Communications of
Bhanu, B., Lee, S., Ho, C. C. and Henderson, T., 1986. Range the ACM, Vol. 24(6), pp. 381-395.
data processing: representation of surfaces by edges. Proc. 8th
International Conference on Pattern Recognition, pp. 236-238. Guo, B., Huang, X., Zhang, F. and Sohn, G., 2014. Classification
of airborne laser scanning data using JointBoost. ISPRS Journal
Brodu, N., Lague D., 2012. 3D Terrestrial LiDAR data of Photogrammetry and Remote Sensing, Vol. 92, pp. 124-136.

This contribution has been peer-reviewed.

doi:10.5194/isprs-archives-XLII-2-W3-339-2017 343
 The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W3, 2017
3D Virtual Reconstruction and Visualization of Complex Architectures, 1–3 March 2017, Nafplio, Greece

Hackel, T., Wegner, J.D., Schindler, K., 2016: Fast semantic Sithole, G. and Vosselman, G., 2004. Experimental comparison
segmentation of 3d point clouds with strongly varying density. of filter algorithms for bare-Earth extraction from airborne laser
ISPRS Annals of Photogrammetry, Remote Sensing and Spatial scanning point clouds. ISPRS Journal of Photogrammetry and
Information Sciences, Vol. III(3), pp. 177-184. Remote Sensing, Vol. 59(1), pp.85-101.
Jagannathan, A. and Miller, E. L., 2007. Three-dimensional Tarsha-Kurdi, F., Landes, T., & Grussenmeyer, P., 2007. Hough-
surface mesh segmentation using curvedness-based region transform and extended ransac algorithms for automatic
growing approach. IEEE Transactions on Pattern Analysis and detection of 3d building roof planes from lidar data. Proc. ISPRS
Machine Intelligence, Vol. 29(12), pp. 2195-2204. Workshop on Laser Scanning, Vol. 36, pp. 407-412
Klasing, K., Althoff, D., Wollherr, D. and Buss, M., 2009. Com- Vieira, M., & Shimada, K., 2005. Surface mesh segmentation and
parison of surface normal estimation methods for range sensing smooth surface extraction through region growing. Computer
applications. Proc. IEEE International Conference on Robotics aided geometric design, Vol. 22(8), pp. 771-792.
and Automation, pp. 3206–3211.
Vo, A. V., Truong-Hong, L., Laefer, D. F., & Bertolotto, M.,
Lavoue’, G., Dupont, F. and Baskurt, A., 2005. A new CAD mesh 2015. Octree-based region growing for point cloud segmentation.
segmentation method, based on curvature tensor analysis. ISPRS Journal of Photogrammetry and Remote Sensing, Vol.
Computer-Aided Design, Vol. 37(10), pp. 975-987. 104, pp. 88-100.
LeCun, Y., Bengio, Y. & Hinton, G., 2015. Deep learning. Vosselman, M.G., Gorte, B.G.H., Sithole, G. and Rabbani, T.,
Nature, Vol. 521, pp. 436-444 2004. Recognising structure in laser scanning point clouds.
International Archives of Photogrammetry, Remote Sensing and
Liu, Y. and Xiong, Y., 2008. Automatic segmentation of unorga- Spatial Information Sciences, Vol. 36(8/W2).
nized noisy point clouds based on the gaussian map. Computer-
Aided Design, Vol. 40(5), pp. 576-594. Wani, M. A., & Arabnia, H. R. , 2003. Parallel edge-region-based
segmentation algorithm targeted at reconfigurable multiring
Lu, X., Yao, J., Tu, J., Li, K., Li, L., & Liu, Y. , 2016. Pairwise network. The Journal of Supercomputing, Vol. 25(1), pp. 43-62.
Linkage for Point Cloud Segmentation. ISPRS Annals of
Photogrammetry, Remote Sensing and Spatial Information Weinmann, M., 2016. Reconstruction and Analysis of 3D
Sciences, Vol.III(3), pp. 201-208. Scenes. Springer International Publishing, ISBN 978-3-319-
29246-5, 233 pages.
MacQueen, J. et al., 1967. Some methods for classification and
analysis of multivariate observations. Proc. 5th Berkeley Weinmann, M., Schmidt A., Mallet C., Hinz S., Rottensteiner F.,
symposium on mathematical statistics and probability. Jutzi B., 2015. Contextual classification of point cloud data by
exploiting individual 3D neighbourhoods. ISPRS Annals of
Menna, F. and Troisi, S., 2007. PBTIN: Prismatic Buffered TIN Photogrammetry, Remote Sensing and Spatial Information
un algoritmo di filtraggio per dati ALS. Bollettino SIFET, Vol. 1. Sciences, Vol. II(3-W4).
Ng, R. T. and Han, J., 1994. Efficient and effective clustering Weinmann, M., Jutzi, B. and Mallet, C., 2014. Semantic 3D
methods for spatial data mining. Proc. 20th Int. Conference on scene interpretation: a framework combining optimal
Very Large Databases. neighborhood size selection with relevant features. ISPRS Annals
Nguyen, A. and Bac Le, 2013. 3D Point Cloud Segmentation: A of the Photogrammetry, Remote Sensing and Spatial Information
survey. Proc. IEEE 6th International Conference on Robotics, Sciences, Vol. II-3, pp. 181-188.
Automation and Mechatronics. Weinmann, M., Jutzi, B. and Mallet, C., 2013. Feature elevance
Niemeyer, J., Rottensteiner, F. and Soergel, U., 2014. Contextual assessment for the semantic interpretation of 3d point cloud data.
classi- fication of lidar data and building object detection in urban ISPRS Annals of Photogrammetry, Remote Sensing and Spatial
areas. ISPRS Journal of Photogrammetry and Remote Sensing, Information Sciences, Vol. II(5-W2)
Vol. 87, pp. 152-165. Xiao, J., Zhang, J., Adler, B., Zhang, H. and Zhang, J., 2013.
Poux F., Hallot P., Neuville R., R. Billen, 2016. Smart point Three-dimensional point cloud plane segmentation in both struc-
cloud: definition and remaining challenge. Proc. 11th 3D Geoinfo tured and unstructured environments. Robotics and Autonomous
Conference, 20–21 October 2016, Athens, Greece Systems 61, pp. 1641-1652.

Rabbani, T., Van Den Heuvel, F., & Vosselmann, G., 2006. Xu, S., Vosselman, G. and Oude Elberink, S., 2014. Multiple-
Segmentation of point clouds using smoothness constraint. entity based classification of airborne laser scanning data in
International Archives of Photogrammetry, Remote Sensing and urban areas. ISPRS Journal of Photogrammetry and Remote
Spatial Information Sciences, Vol. 36(5), pp. 248-253. Sensing, Vol. 88, pp. 1-15

Sapkota, P. P., 2008. Segmentation of Coloured Point Cloud Yamauchi, H., Lee, S., Lee, Y., Ohtake, Y., Belyaev, A., &
Data. PhD Thesis, International Institute for Geo-Information Seidel, H. P., 2005. Feature sensitive mesh segmentation with
Science and Earth Observation, Enschede, The Netherlands. mean shift. Proc. IEEE Shape Modeling and Applications
International Conference, pp. 236-243.
Sappa, A. D., & Devy, M. , 2001. Fast range image segmentation
by an edge detection strategy. Proc. IEEE 3rd 3-D Digital Zhang, X., Li, G., Xiong, Y., & He, F., 2008. 3D mesh
Imaging and Modeling, pp. 292-299 segmentation using mean-shifted curvature. Proc. Int.
Conference on Geometric Modeling and Processing, pp. 465-
Schnabel, R., Degener, P. and Klein, R., 2009. Completion and 474.
reconstruction with primitive shapes. CGF Eurographics, Vol.
28(2), pp. 503-512.
Sithole, G. and Vosselman, G., 2003. Report: ISPRS comparison
of filters. ISPRS Commission III - WG3.

This contribution has been peer-reviewed.

doi:10.5194/isprs-archives-XLII-2-W3-339-2017 344

View publication stats