BUILDING BOUNDARY EXTRACTION
FROM HIGH RESOLUTION IMAGERY AND LIDAR DATA
Liang Cheng*, Jianya Gong, Xiaoling Chen, Peng Han
State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing,
Wuhan University, Wuhan, China - lcheng.geo@gmail.com,
-(jgong, cxl)@lmars.whu.edu.cn, -zb34xiaowu@163.com
KEY WORDS: Building boundary, High resolution image, Lidar data, Data integration
ABSTRACT:
Building boundary data are necessary for the real estate industry, 3D city models and many other applications. In this study, a novel
approach integrated high resolution imagery and Lidar data is proposed for automatically obtaining building boundaries with precise
geometric position and details. The high resolution images were used to directly extract the building boundaries with precise
geometric position, our approach is focused on improving the correctness and completeness of the extracted boundaries by
integrating Lidar data. The approach consists of four steps: Lidar data processing, building image generation, line segment extraction,
and boundary segment selection. Firstly, the segmented building points need to be determined from raw Lidar data. Then, a building
image is generated by processing an original image using a bounding rectangle and a buffer, which are derived from the segmented
building points. Based on the building image and rough principal direction constraints, an algorithm is proposed to estimate the
principal orientations of a building, which ensures the accuracy and robustness of the subsequent line segments extraction. Finally,
an algorithm based on Lidar point density analysis and Kmeans clustering is proposed to identify accurate boundary segments from
the extracted line segments dynamically. The experiment results demonstrated that the proposed approach determined building
boundaries well.
1. INTRODUCTION
Building boundary data are necessary for the real estate
industry, city planning, homeland security, flood management,
and many other applications. The extraction of building
boundary is also a crucial and difficult step toward generating
city models. The automatic extraction of a building boundary
from an image has been a research issue for decades; many
related studies have been reported. McGlone and Shufelt (1994)
extracted building boundaries by calculating vanishing points
using the Gaussian Sphere technique to detect horizontal and
vertical lines. Mohan and Nevada (1989) described an approach
based on perceptual grouping for detecting and describing 3D
objects in complex images and applied the method to detect and
describe complex buildings in aerial images. Xu, et.al., (2002)
employed a Gabor filter to eliminate noisy edges, and then used
a Normalized Central Contour Sequence Moment to select
regular contours. A detailed review of techniques for the
automatic extraction of buildings from aerial images was made
by Mayer (1999). However, a difficult problem still exists in
automated extraction of building boundaries because it is
almost impossible to automatically distinguish building
boundaries from other line segments in a high accuracy only
based on aerial imagery. Moreover, a robust solution for
boundary extraction is needed because too much complex
information is contained in an image, especially for a very high
resolution image.
Many studies also focused on boundary extraction by using
Lidar point clouds. Weidner (1996) used the difference between
DSM and DTM to determine the building outlines. Vosselman
& Sander (2001) and Haala & Brenner (1999) used plan maps
to support boundary extraction from Lidar points. Many related
papers have been published (Morgan and Habib, 2002;
Rottensteiner and Briese, 2002; Overby, et.al., 2004;
Vosselman and Kessels, 2005; Brenner, 2005). In general, it is
hard to obtain a detailed and geometric precise boundary only
using Lidar point clouds considering its low spatial resolution.
To eliminate noise effects and get building boundaries with
precise geometric position, some researchers used the minimum
description length (MDL) method to regularize the ragged
building boundaries (Weidner and Forstner, 1995; Suveg and
Vosselman, 2004). Zhang et.al.(2006) used Douglas-Peucker
algorithm to remove noise in a footprint, then adjusted the
building footprints based on estimated dominant directions.
Sampath and Shan (2007) performed building boundary tracing
by modifying a convex hull formation algorithm, then
implemented boundary regularization by a hierarchical least
squares solution with perpendicularity constraints. However,
regularization quality is also dependent on the point density of
Lidar data; and limitation of Lidar data resolution and errors in
filtering processes may cause obvious offset and artefacts in the
final regularized building boundary (Sampath and Shan, 2007).
Although it is difficult to obtain building boundaries with
precise geometric position using Lidar data, Lidar data is able
to directly provide measured three-dimensional points. On the
other hand, although very high resolution images can provide
building boundaries with precise geometric position, the
accuracy of automatic boundary extraction is still in a low level.
It seems to be valuable to extract building boundaries by
integrating very high resolution imagery and Lidar data.
However, how to integrate the two data sources for building
boundary extraction is still a problem; few approaches with
technical details has been published (Rottensteiner, 2005).
* Corresponding author.
693