THE RENDERING OF BUILDING TEXTURE FROM LAND-BASED VIDEO
Zuxun Zhang, Zhizhong Kang
School of Remote Sensing and Information Engineering, Wuhan Univ., Wuhan 430079, P.R. China - (zxzhang ,
zzkang)@supresoft.com.cn
KEY WORDS: Rendering, Mobile Mapping; Image Sequences; Texture; Mosaic; Rectification; Distortion
ABSTRACT:
Mobile mapping is a hot issue in the communities of photogrammetry and computer vision nowadays. Mobile mapping systems
represent a significant advance in integrated, multi-sensor digital mapping technology, providing an innovative path towards rapid
and cost-effective collection of high-quality and up-to-date spatial information. Therefore how to acquire geometric and texture
information from the images for reducing human operations and enhancing efficiency is a challenge in mobile mapping data
processing. In this paper, the rendering of building texture from land-based video is implemented. Firstly, some preprocessing of
images is implemented. The occlusion of tree is removed because it goes against the extraction of lines from buildings. Close-ranged
DC images are greatly affected by lens distortion, so lens distortion is calibrated and the influence is eliminated accordingly. Then,
Land-based Video image sequences are automatically rectified to eliminate nonlinear distortion thanks to the large obliquity of the
posture of DC. The building texture is finally mosaic from rectified image sequence and refined by the process of texture analysis.
1. INTRODUCTION
Real texture data, i.e. images, are required to make building
objects more genuine in 3D city modeling(3DCM ). Textures of
terrains and the top of buildings are acquired from aerial images
presently. The textures of building facades, however, are still
acquired from images taken manually with digital camera.
Since lacking index of images, the processes of searching
images, preprocessing images and corresponding images with
buildings are implemented manually. As a result, the problems
of low efficiency and high labour intensity emerge. Therefore,
i's becoming a hot issue in communities of digital
photogrammetry , GIS and computer vision that how to solve
these problems and relevant research are carried out (Debevec.P
et al., 1996 ; Cipolla.R et al., 1999).
The acquisition of land-based video for buildings has the
characteristics of convenience and rapidness. The resolution of
textures obtained from land-based video is furthermore higher
than those obtained from helicopter-based video. Therefore, it
will be aided greatly by the automatic processing of textures
from land-based video that the rapid reconstruction of texture
for building facades in 3DCM. Accordingly, it’s discussed in
this article that the rendering of building textures from land-
based video.
Firstly two groups lines are extracted from a single image. The
occlusion of tree is removed employed the information of hue
and lines with regular orientation because it goes against the
extraction of lines from buildings. Close-ranged DC images
have nonlinear distortion due to the large obliquity of the
posture of DC and obvious lens distortion. Radial lens
distortion parameter 1 is estimated because the major part of
the lens distortion is eliminated by applying it only. Due to lens
distortion, straight lines in object space appear as slightly
curved lines in the image. Here we break up long curved image
lines into smaller sections extracted later. Therefore, those
sections extracted along the same curved image line are bound
to be collinear. A model of adjustment system is then evolved
according to this conclusion to estimate the radial lens
distortion parameter k1.To eliminate the nonlinear distortion
cause by large obliquity of the posture of DC, the automatic
rectification of images is implemented. The key of automatic
rectification of images is how to acquire camera parameters
from the image information. Since the shape of buildings is
always regular, there are many straight lines in building facades
parallel to X-axis and Y-axis in object space. Namely the
orientation of these lines is known. Therefore, it is presented to
calibrate DC parameters that an algorithm combined with the
constraint of straight lines bundle and the constraint of known
orientation of parallel lines in object space.
The automatically mosaic process of fagade texture is finally
implemented employed the strategy of selecting mosaic point
pair combined with correlation coefficient and geometric
constraint of relative orientation. Tone adjustment is completed
for the color difference between each two images contributing
to illumination.
2. ESTIMATION OF LENS DISTORTION
As we know, close-ranged DC images are greatly affected by
lens distortion and the influence should be eliminated
accordingly. In this article, an algorithm that estimates lens
distortion according to the curvature of straight lines thanks to
lens distortion. Because the major part of the lens distortion is
b. Lines extracted after dimidiation
C
_—_—— Ta :
A C E
c. The curvature of straight lines
Fig.1 The determination of curvature of
straight lines caused by lens distortion
Inter!
elimii
estim
As F
while
There
is exi
het
of co
denot
distor
distor
accor
distor
Since
from
distin
3.1 €
As I
infori
becau
range
in th
illusti
ellips
remai
mean
concl
remo:
stabil
Here
Occh
3.2 (
Lines
Linc
em
