XXIX-B3, 2012 
09. Automatic Road 
n Classifier Fusion, 
pp. 1-6 
-based simplification 
Graphics 24 (2), pp. 
2005. Mesh saliency, 
lings of SIGGRAPH 
ialitative comparison 
oftware ^ packages. 
mendedPubl/Vitaly/ 
zical characterisation 
esis, University of 
04. The Automatic 
, University of New 
Australia, 
load?doi=10.1.1.133 
). Automatic Road 
Classifier Fusion , 
Joint Event, 
arnumber-5137739 
L., 2010.Quanfu B. 
AR Data Based on 
| Congress on Image 
2133. 
CLASSIFICATION BY USING MULTISPECTRAL POINT CLOUD DATA 
Chen-Ting Liao ^ *, Hao-Hsiung Huang * 
* Department of Land Economics, National Chengchi University, 
64, Sec. 2 Zhinan Rd., 11605 Taipei, Taiwan — 99257006@nccu.edu.tw, hhh@nccu.edu.tw 
Commission III, WG III/2 
KEY WORDS: Classification, Image Matching, Close Range Photogrammetry, Infrared, Point Cloud 
ABSTRACT: 
Remote sensing images are generally recorded in two-dimensional format containing multispectral information. Also, the semantic 
information is clearly visualized, which ground features can be better recognized and classified via supervised or unsupervised 
classification methods easily. Nevertheless, the shortcomings of multispectral images are highly depending on light conditions, and 
classification results lack of three-dimensional semantic information. On the other hand, LiDAR has become a main technology for 
acquiring high accuracy point cloud data. The advantages of LiDAR are high data acquisition rate, independent of light conditions 
and can directly produce three-dimensional coordinates. However, comparing with multispectral images, the disadvantage is 
multispectral information shortage, which remains a challenge in ground feature classification through massive point cloud data. 
Consequently, by combining the advantages of both LiDAR and multispectral images, point cloud data with three-dimensional 
coordinates and multispectral information can produce a integrate solution for point cloud classification. Therefore, this research 
acquires visible light and near infrared images, via close range photogrammetry, by matching images automatically through free 
online service for multispectral point cloud generation. Then, one can use three-dimensional affine coordinate transformation to 
compare the data increment. At last, the given threshold of height and color information is set as threshold in classification. 
1. INTRODUCTION 
Passive remote sensing systems record electromagnetic energy 
reflected or emitted from the surface as two- dimensional 
multispectral images. The general used bands are blue 
(0.45~0.52um), green (0.52~0.60um), red (0.63~0.69um), Near 
Infrared (0.70~1.3um), Middle Infrared (1.3~3um) and thermal 
Infrared (3~14um). Due to ground features have their own 
characteristic in different spectrum, while classifying through 
Multispectral Images, generally, higher divergence between 
bands, may lead to higher classification accuracy. Therefore, 
one can interpret ground features effectively by collecting 
multispectral images, for example, healthy vegetation reflects 
massive near infrared light, and water body absorbs near 
infrared light, so one can use near infrared light with other 
bands for recognizing vegetation and water body. 
LIDAR is an active remote sensing system, which can acquire 
ground feature point cloud data through laser scanning 
technique; this allows remote sensing data development toward 
three-dimensional space. Point cloud data includes three 
dimensional coordinates, intensity and other abundance spatial 
information, which contains much more potential to interpret 
ground features than two-dimensional image does. In general, 
LIDAR scans ground features by single band laser light, for 
instance, green lasers at 0.532um has water penetration ability, 
and vegetation has high sensitive to near infrared laser light 
region in 1.04um to 1.06um (Jensen, 2007). Generally, point 
cloud data is acquired only through single band laser light, and 
lack of multispectral information. 
Consequently, this research uses close-range photogrammetry 
method to collect visible light and near infrared images, and 
" Corresponding author. 
135 
chooses free online service — Photosynth, which is provided by 
Microsoft, as automatically image matching technique for point 
cloud generation. After exporting the point cloud data, one can 
use three-dimensional affine coordinate transformation to merge 
multispectral point cloud and visible light point cloud data, as a 
check for the accuracy and precision for multispectral point 
cloud data. Comparing with point cloud data generated by using 
visible images, increment of multispectral point cloud data 
acquired by adding near infrared images were then evaluated. 
Thereafter, the multispectral point clouds for ground feature 
were classified. The results of classification have been analysed, 
for understanding whether the point clouds generated by 
multispectral information have good potential in classification. 
2. BACKGROUND 
Ground features have diffuse reflectance properties respectively. 
Understanding of the spectral reflectance of ground features can 
assist in distinguishing and recognizing diffuse ground features. 
Generally, collecting visible light image can only acquire 
spectral reflectance from 0.4um to 0.7um by collecting other 
band, e.g. near infrared light, the spectrum beyond visible light 
can be obtained, which can interpret ground feature effectively. 
By matching multispectral images through free online service, 
such as Photosynth, one can get point cloud data from image 
collected in close range; via three-dimensional coordinate 
transformation, combining it with visible light point cloud data, 
therefore one can compare and analyze the benefit of 
multispectral images on increasing point cloud data and 
classification assisting ability. 
The following sections will introduce the advantage in ground 
feature interpretation by adding near infrared, brief introduction