IEEE Geoscience and Remote Sensing Magazine - December 2020 - 42

one public, influential data set is based on image-derived
points, and its range is just a very small area around a single
building [132]. More efforts are therefore needed.
3)	 RGB-D sensors are limited by their close range, so they are
usually applied in indoors. In PCS studies, plane segmentation is the main task for RGB-D data. In PCSS studies, since
several benchmark data sets are derived from RGB-D sensors,
many deep-learning-based methods are tested on them.
4)	 Few relevant PCS or PCSS studies have been done involving
InSAR point clouds, but these have shown potential in urban monitoring, especially in regard to building structure segmentation.
BENCHMARK DATA SETS
Public standard benchmark data sets have demonstrated significant effectiveness for algorithm development, evaluation, and

comparison. Most of them are labeled for PCSS rather than
PCS. Since 2009, several benchmark data sets have been available for PCSS. However, early data sets have many shortcomings. For example, Neither the Oakland 3D Point Cloud MLS
data set [96], the Sydney Urban Objects MLS data set [133],
the Paris-Rue-Madame MLS data set [134], the IQmulus and
TerraMobilita Contest MLS data set [35], nor the ETHZ CVL
RueMonge 2014 multiview stereo data set [132] can sufficiently provide both different object representations and labeled
points. The Karlsruhe Institute of Technology and Toyota Technological Institute of Chicago (KITTI) data set [135] and New
York University Depth Dataset V2 (NYUv2) data set [136]
have more objects and points than the aforementioned data
sets, but they do not provide a labeled point cloud directly.
These must be generated from 3D bounding boxes in KITTI or
depth images in NYUv2.

TABLE 2. AN OVERVIEW OF PCS AND PCSS APPLICATIONS SORTED ACCORDING TO DATA ACQUISITIONS.
URBAN
IMAGE
DERIVED

Building façades: [65] (2018/RG), [66]
(2005/RG); PCSS: [67] (2018/DL), [68]
(2018/DL), [69] (2017/DL), [70] (2019/DL)

ALS

Building plane PCS: [72] (2015/R), [73]
(2014/R), [74] (2007/R, HT), [75] (2002/
HT), [76] (2006/C), [77] (2010/C), [78]
(2012/C), [79] (2014/C); urban scene: [80]
(2007/C), [81] (2009/C); PCSS: [82] (2007/
ML), [83] (2009/ML), [84] (2009/ML), [85]
(2010/ML), [86] (2012/ML), [87] (2014/
ML), [88] (2017/HT, R, ML), [89] (2011/ML),
[90] (2014/ML), [4] (2013/HT, ML)

MLS

Buildings: [93] (2015/RG); urban objects:
[94] (2012/RG); PCSS: [89] (2011/ML),
[95] (2015/ML), [5] (2015/ML), [8] (2012/
ML), [90] (2014/ML), [96] (2009/ML), [97]
(2017/ML), [98] (2017/DL), [99] (2018/DL),
[100] (2019/O, DL)

TLS

Building/building structure PCS: [103]
(2007/R), [93] (2015/RG), [104] (2018/
RG, C), [105] (2008/C); buildings and
trees: [106] (2009/RG); urban scene:
[107] (2016/O, C), [108] (2017/O, C), [109]
(2018/O, C); PCSS: [6] (2015/ML), [110]
(2009/O, ML), [111] (2016/ML), [67] (2018/
DL), [98] (2017/DL), [2] (2018/O, DL), [112]
(2019/DL) [70] (2019/DL)

FOREST

INDUSTRY

Tree structure
PCS: [91]
(2004/C); forest
structure: [92]
(2010/C)

Plane PCS: [101] (2013/R), [102] (2017/R)

Tree PCSS: [113]
(2005/ML)

Plane PCS: [114] (2011/HT)

RGB-D

InSAR
NOT
-MENTIONED
DATA

INDOOR
Plane PCS: [71] (2015/HT)

Plane PCS: [115] (2014/HT), [104] (2018/RG, C);
PCSS: [116] (2012/ML), [117] (2013/ML), [118]
(2018/DL), [119] (2018/DL), [98] (2017/DL), [1]
(2017/DL), [120] (2017/DL), [3] (2018/DL), [2]
(2018/DL), [99] (2018/DL), [121] (2018/DL),
[70] (2019/DL), [112] (2019/DL), [122] (2019/
DL), [123] (2019/DL), [124] (2019/DL), [125]
(2019/DL), [126] (2019/DL), [100] (2019/O, DL);
instance segmentation: [127] (2018/DL), [128]
(2019/DL), [123] (2019/DL), [124] (2019/DL)
Building/building structure: [47] (2015/C),
[45] (2012/C), [46] (2014/C)

Tree PCS: [48]
(2015/C)
[129](2005/HT),
[130] (2015/R),
[131] (2018/R)

RG: region growing; R: RANSAC; C: clustering based; O: oversegmentation; ML: machine learning; DL: deep learning.

42

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

DECEMBER 2020
IEEE Geoscience and Remote Sensing Magazine - December 2020

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - December 2020
