International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
The paper is organized into seven parts. F irstly, a framework
for an operational road database updating system is presented in
Section 2. Road map updating modes are briefly discussed in
Section 3 followed by a detailed description about road change
detection and updating based on map conflation in Section 4.
Section 5 addresses some key issues in modelling road network
changes. Some preliminary results along this line will be
illustrated in Section 6. Finally, some conclusions will be given
in Section 7.
2. AFRAMEWORK FOR ROAD MAP UPDATING
Intuitively, an operational road map updating system should
include the following three main functions:
1) Generating a new version of road features or the whole
road network either by ground surveying or by road
extraction from imagery;
2) Detecting road changes, i.e. identifying the roads that
remain unchanged, have disappeared, or emerged recently;
3) Updating the road database. This includes updating the
geometric data of the roads; transferring attributes from
the old version to the new version database; and
organizing both versions of the road databases in a spatio-
temporal model.
A lot of work had been done in each of these three areas
separately, but very few researchers have treated the three parts
in a united way.
In this paper a wavelet-based road junction and centerline
extraction processing is initially performed. Map conflation
techniques are then applied for road change detection and
updating. Finally, the change information of the road network is
organized in an efficient way to facilitate spatio-temporal
queries and spatio-temporal analysis. The proposed framework
for an operational road database updating system is illustrated
in Figure 1.
The wavelet-based road junction and centerline extraction has
been detailed in the paper [Zhang & Couloigner, 2004] and will
not be repeated here.
3. ROAD MAP UPDATING MODE
Both road change detection/updating and spatio-temporal GIS
have been under research for more than ten years now. There
are a lot of new ideas and new approaches promoted in both
areas. However, most of the research is carried out separately
and very few people are working on both problems
simultaneously. The author would argue that a spatio-temporal
perspective will be very helpful to develop an operational
system for road change detection and map updating. On the
other hand, a change detection and updating perspective will
also shed some light on the research of temporal GIS.
2!
3
0
3v
- Im]
EE ae er)
= 3
3 + 4
°c.
= New road
= database
=
fro |
= [ | Updating the road database |
SS
S A
9 Conflation of the two different versions of road
= 4
E
E A
S di Feature Matching for road change detection d
\
Road extraction Existing
based on wavelet GIS
A Databases
UONIENXA PEOY
eoreTerenced images
Figure 1 The proposed framework for an operational road
database updating system
Road maps could be updated by ground surveying, either by
using a traditional method (total station, GPS) or by using a
more automatic method (e.g, mobile mapping system).
Usually, a survey team will be informed that some roads have
been changed, go to the site and record the new positions of the
roads. From a spatio-temporal point of view, this method is
most suitable because only the changed roads should be taken
into account and the time stamps could be easily put either at
the tuple level or at the attribute level. In addition, the change is
closely linked to the events which had caused the road to
change. The minus of this method is that it needs many
surveyors to focus on this task in order to record the change
timely. Therefore, it is a costly and labour intensive way to
update a road network database.
The second method is to use a more recent map to update the
old road map. By feature matching, the unchanged and changed
roads can be determined during the mapping time interval. This
is an ad-hoc technology to maintain a road network database.
The revision time may be one year or more than five years
depending on the application purpose and other situations. It is
obvious that this method is close to a snapshot approach to
model the changes. Change has a very coarse temporal
resolution. It may also be very difficult to determine when the
road changes occurred because we have little information about
the events which caused these changes. The transaction
time/database time can be indicated at a table or tuple level
because all the changes have the same transaction time. The
valid time is difficult to determine unless all the changes have
been recorded immediately after their occurrence.
The third method is to extract the road network and detect the
changes based on new remotely-sensed imagery. This
technology has been widely researched for many years.
Although there are few successful fully automated techniques,
there are many partially automated feature extraction
techniques available to detect road network changes. The
limitation is identical than for the second way: same transaction
time for all the road changes and difficulties in identifying the
valid time of the changes.
Intern
The m
based
section
4) M
After a
feature
nodes :
dataset:
more af
followi
4.1 No
Node n
two ve
map co
Both d
used t
distance
discrep:
based o
presenti
structur
arcs an«
4.2 Po
In this
original
