ne XXXIX-B3, 2012
‚Ss
000. Simplified formulae
variance components in
esy, 74(6), pp. 447-457.
ne, 2009. Non-stationary
least-squares collocation,
508.
An improved model for
on of SAR and SPOT
grammetry and Remote
^, and L. M. Bruce, 2008.
jsaicking for airborne
es, Photogrammetric
2), pp. 169—181.
1981. Random sample
tting with applications to
graphy, Communications
ige features from scale-
— Journal of Computer
ns and Least Squares,
n, Maryland, 497 p.
2005. A performance
; Transactions on Pattern
27(10), pp. 1615-1630.
lunder detection and data
nts, Journal of Surveying
tion of building outlines
optical features, ISPRS
note Sensing, 58(1-2), pp.
and H. Kutterer, 2009.
nd cluster analysis as
geodetic applications,
891.
it, 2000. Elements of
: in GIS, McGraw-Hill,
C. Liu, 2009. Conjugate
nator for the gray-level
titute of Engineers, 32(5),
QUALITY ANALYSIS OF VEHICLE-BASED SEQUENCE IMAGES RELATIVE
ORIENTATION BASED ON COMPUTER VISION
YAN Li? XU Zhenliang *° *
aSchool of Geodesy and Geomatics, Wuhan University, 129 Luoyu Road, Wuhan 430079, China
Email: iyan@sgg. whu.edu.cn;xuzhentiang@whu.edu.en.
b School of Geomatics, Liaoning Technical University, 47 Zhonghua Road, Fuxin 123000, China
Commission III, WG III/5
KEY WORDS: Relative Orientation, Direct Geo-referencing, Computer vision, Sequence images, Mobile mapping, SIFT,
RANSAC
ABSTRACT:
Analyzed the quality and its influence factors of relative orientation of vehicle-based sequence images by comparing with the result
obtained from Position & Orientation System using Direct Geo-referencing. Studies have shown that, under normal circumstances,
the image Relative Orientation based on Computer vision is more robust, and under special conditions, it is more practical for
analyzing and increasing the quality of vision measurement. Besides these, the method can be borrowed by many surveying and
mapping related fields, taking indoor robot mobile environment awareness and unmanned automobile for instance.
1. INTRODUCTIONS
Recent years, the using of POS (Position & Orientation
System) in providing direct geo-referencing for imaging
equipment makes it a breakthrough in converting MMS
(Mobile Mapping System) to realistic productivity; it has been
widely used in such space information service areas as digital
city and intelligent transportation. Besides, it broadens the
content of information mapping and surveying, improving the
work efficiency and quality magnificently. But, in such areas
as traditional photogrammetry, low altitude remote sensing
based on unmanned airborne, close range photogrammetry on
the ground, considering of the cost, safety and complexity, it is
more practical to acquire the relative attitude of images by
traditional vision methods, that is resection; Besides, in indoor
robot environment sensing and unmanned driving field, it
commonly uses vision to achieve autonomous navigation. So,
RO (Relative Orientation) based on vision is still of great
value in many areas.
In photogrammetry and CV (Computer Vision), RO between
sequential images is achieved by the coplanar conditions of
homologous points between two images and corresponding
object points, in recent years, with the improvement of
efficiency of matching algorithm, RO has realized automation
in numerous of industrial photogrammetry software, (LI Jian
etal 2010) , but this model mainly suites for conditions when
the heading changes little, this model usually uses
linearization and LS (Least Square) iterating to get the
orientation parameters, but this model does not suits for MMS
because of the great heading, the iterating is easily plunging
into the local optimization and leads to failure if the initial
value is not chose properly. Besides, the precision of
traditional photogrammetry is just theoretic, objective
precision of RO cannot be acquired, so one cannot analyze the
factors that affect the precision thoroughly; the occurrence of
high precision DG (Direct Geo-referencing) makes it possible
to check the quality of RO based on traditional method. Based
on this, this paper proposes a direct way to analysis the RO
absolute error of sequential vehicle-based images based on DG
and CV.
2. THE VISION RO PROCEDURES OF SEQUENTIAL
IMAGES
The RO procedures of vehicle-based sequential images include
the selection of stereo images, feature points extraction and
matching, the estimation and evaluation of RO Euler angle;
the basic processes are described as follows:
(1) The selection of sequential images. The MMS (mobile
mapping system) is mainly aimed at acquiring city
environment data, so when selecting sequential images, we
can choose some building images with some degree of
overlapping and with abundant feature points.
(2) Feature points extraction and matching of sequential
images. The cameras on mobile mapping system usually
belong to small frame non-metric camera, and the sequential
images of MMS usually have large heading angle, the
matching algorithms based on gray correlation in traditional
photogrammetry can hardly insure the correctness of the RO
(Schenk and Toth,1993). Fortunately, the high degree
overlapping of MMS sequential images provides basic
fundamentals for image matching and RO. Besides, SIFT
(Lowe,2004), which is famous in matching algorithms in CV,
makes it possible to solve this question. The features of SIFT
are invariant to image scale and rotation, and are shown to
provide robust matching across a substantial range of affine
distortion, change in 3D viewpoint, addition of noise, and
change in illumination. The features are highly distinctive, in
the sense that a single feature can be correctly matched with
*YAN Li, professor, Ph.D supervisor. His research field includes digital photogrammetry,computer vision and Mobile Mapping
Technology(MMT).
