Page 646 - NGTU_paper_withoutVideo

Modern Geomatics Technologies and Applications

A robust and efficient building segmentation from the LiDAR point clouds

1
2
Fariba Dolati Tamay , Hossien Arefi , Behnaz Bigdeli , Danesh shokri 1
1

1 Department of Photogrammetry and Remote Sensing, School of Surveying and Geospatial Engineering,
College of Engineering, University of Tehran, Tehran, Iran (fariba.dolati, hossein.arefi)@ut.ac.ir,
danesh.shokri.94@gmail.com
2 Department of Civil Engineering, School of Civil Engineering, Shahrood University of Technology
(bigdeli@shahroodut.ac.ir)

Abstract: 3D modelling is an important task in urban planning and notably smart cities. For this purpose, many remote
sensing technologies have been proposed for acquiring 3D data such as Mobile Laser Scanning (MLS) and Aerial Laser
Scanning (ALS) platforms. These two systems provide densely point clouds including not only the positioning data but
also giving information about 3D attributes such as elevation. This study uses both ALS and MLS point clouds in order
to produce 3D model of urban buildings. The proposed algorithm consists of three main steps; (i) pre-processing, (ii)
building detection, and (iii) 3D modelling of the extracted buildings. Regarding the pre-processing, a considerable amount
of points related to the ground and noisy points are eliminated due to accelerating in the time of computation. In the next
step, 2D location of each building, is extracted using the interactive extraction method. Also, the model and type of roofs
are acquired from this step too. Finally, a Random Sample Consensus (RANSAC)- based approach separates the walls of
each building which would lead to the modelling of them by fitting a 3D cube. The algorithm was evaluated in an urban
region with area of about 6 hectares. It played an acceptable role in the both detecting and modelling of buildings with the
98% (precision and recall) and 0.05 (RMSE).
1. Introduction
Three-dimensional model of urban buildings has found significant applications such as in urban planning, cartography,
virtual tourism, and surveillance [1]. There are still several challenges to generate 3D building models from point clouds with
high quality, in which different methods are already proposed by computer vision, and photogrammetry communities [2, 3].
Nowadays, many computations are based on 3D building models. For example, in order to locate solar panels and deployment
of power lines, buildings geometric model plays an important role and needs be considered [4-6].
A well-known remote sensing technology to produce point clouds is Light Detection and Ranging (LiDAR). Point clouds
are utilized to generate 3D models. This technology collects millions of points rapidly as x, y, z positional coordinates [7]. Three-
dimensional structure of surface objects is captured by LiDAR. Due to the nature of system, points are collected in an unorganized
fashion and a significant process is needed to process point clouds data. An important step in urban modeling is to first identify
and extract buildings from other objects such bare earth, and trees [8]. Due to different types of buildings and their density and
composed in urban cities, 3D building modelling is difficult [9].
In order to generate 3D city models in some methods, it is necessary to produce 3D building models.[10]. The level of
detail (LoD) is a common concept for 3D building models in CityGML (City Geography Markup Language) [10]. Four LoDs of
CityGML is shown in Figure 1.

Figure 1. Example of the LoDs for a residential building [11]

1