Repeated Structure Detection for 3D
Reconstruction of Building Façade
fro
Data
Yanming Chen, Xiaoqiang Liu, Mengru Yao, Shulin Deng, Feixue Li, Liang Cheng, and Manchun Li
Abstract
This study proposes a new method for repeated structure
detection and the three-dimensional (3D) reconstruction of
building façades from mobile lidar data. Firstly, the building
façade is divided and unrolled to simplify the complex façade
structures, improving the automation of structure detection
and building reconstruction. Subsequently, the unrolled fa-
çade is decomposed into tiles by analyzing the repeated struc-
tures. Tiles with strong similarities are matched and merged
to restore the imperfect façade points. Based on the restored
points and repeated structures, a 3D building façade can be
reconstructed with a complete structure and fine detail. An
analysis is conducted to compare the constructed 3D model
with the lidar points of actual façade. The results of this
analysis demonstrate that the proposed method can effec-
tively deal with missing areas caused by occlusion, viewpoint
limitation, and uneven point density, as well as realizing the
highly complete 3D reconstruction of a building façade.
Introduction
The three-dimensional (3D) reconstruction of a build-
ing model is a hot topic of research in many fields such as
architecture, engineering, construction, change detection
and urban planning(Chen
et al.
, 2014; Huang
et al.
, 2017;
Qin
et al.
, 2015; Xu
et al.
, 2015). It is an important task to
detect and reconstruct buildings from optical images or lidar
data, and therefore has attracted considerable attention over
the past few decades (Huang and Zhang, 2012; Huang
et al.
,
2011; Wang
et al.
, 2016). In recent years, great progress has
been made to reconstruct 3D models of buildings and façades
quickly and accurately (Li
et al.
, 2017; Wang
et al.
, 2016).
The modeling of different features on a building façade has
attracted increased attention because it features in many ap-
plications (Cheng
et al.
, 2018; Wang
et al.
, 2018).
Mobile lidar has become an efficient means of data acquisi-
tion for the modeling of building façades (Leberl
et al.
, 2010).
It is an emerging mobile mapping system that can rapidly
capture road surface features from high-speed vehicles with
an overlooking or upward-looking viewpoint (Cheng
et al.
,
2014). Despite the complications of urban landscapes, this
system offers the advantages of highly precise positioning,
quick acquisition, and abundant façade detail (Yu
et al.
,
2015). Thus, the mobile lidar is extremely advantageous to
the 3D reconstruction of large-scale urban landscapes (Musi-
alski
et al.
, 2013). However, due to the existence of extremely
complex structures, the limitations of data collection, and the
occlusion and disturbance of objects on the street (e.g., trees
and vehicles), the reconstruction of a high-quality building
façade model with a high level of automation presents a chal-
lenging but very worthwhile research topic(Wan and Sharf,
2012).
Consequently, exploratory research into the reconstruction
of building façade models has been undertaken by several re-
searchers and relevant reviews (Haala and Kada, 2010; Musi-
alski
et al.
, 2013; Tang
et al.
, 2010). Specifically, Becker (2009)
constructed the details of windows, doors, and protrusions by
utilizing the transmission characteristics of lidar points, and a
formal grammar was generated for the reconstruction of build-
ing façades. Similarly, Pu and Vosselman (2009) presented a
knowledge-based approach to the automatic reconstruction
of building façade models. However, these methods can be
easily affected by the data quality, especially by occlusions
and noise existing in the points. In addition, other details like
balconies also make reconstruction more difficult.
Urban buildings usually have characteristics stemming
from the time at which they were built, as well as the cul-
ture and tastes of the region. Therefore, similar structures
are likely to be seen in many building façades (Mitra
et al.
,
2013). The detection of similar structures is usually based on
the hypothesis that structure cells present a repetitive lattice
distribution (Minwoo
et al.
, 2009). Many researchers have ex-
tracted building façade structures from images (Ceylan
et al.
,
2012; Xiao
et al.
, 2008). Müller
et al.
(2007) proposed shape
grammar rules that had been derived from façade images to
subdivide the texture of a façade into repeated elements such
as floors, windows, and doors. Li
et al.
(2011) presented a
state-of-the-art method by fusing 2D photographs and 3D lidar
points to decompose depth-layer of façades and produce 3D
consolidated models. Teeravech
et al.
(2014) discovered re-
petitive patterns by using a
RANSAC
-styled sine wave fitting
algorithm, and then decomposed the façade images of build-
ings into floors and tiles automatically.
For building façade point clouds, researchers have also tak-
en advantage of structural regularities to solve the problems
caused by occlusion and noise. Pauly
et al.
(2008) introduced
a framework for identifying repeated elements as regular
lattice structures in point- or mesh-based models. However,
this framework cannot handle “warped” sequential repeti-
tive structures. Bokeloh
et al.
(2009) identified similar and
repetitive structures through the line feature matching algo-
rithm. This algorithm works only when all the structures are
identical, and the reconstruction result is highly dependent
School of Geography and Ocean Science, Nanjing University,
Nanjing 210093, China; Jiangsu Provincial Key Laboratory
of Geographic Information Science and Technology, Nanjing
University, Nanjing 210093, China; Collaborative Innovation
Center for the South Sea Studies, Nanjing University, Nanjing
210093, China; Department of Geographic Information
Science, Nanjing University, Nanjing 210093, China
(
).
Photogrammetric Engineering & Remote Sensing
Vol. 85, No. 2, February 2019, pp. 93–108.
0099-1112/18/93–108
© 2019 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.85.2.93
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
February 2019
93
