ul 2004 
50 
8 
142 
‚90% 
2:1% 
3.0% 
  
  
n. 
letection 
esources 
lexity of 
such as 
ime and 
should 
and the 
Romote 
35. 
timating 
8, No.5, 
, 1998. 
ssment 
ronment 
atistical. 
AUTOMATIC FEATURE-LEVEL 
CHANGE DETECTION (FLCD) FOR ROAD NETWORKS 
Haigang Sui, Deren Li, Jiaya Gong 
State Key Lab of Information Engineering in Surveying, Mapping and Remote Sensing (LIESMARS) of Wuhan 
University, 129 Luoyu Road, Wuhan, Hubei Provinice, P.R.China,430079 
haigang_sui(@263.net 
Commission II, WG IUIV 
KEY WORDS: Feature Level Change Detection, Remote Sensing, GIS, Feature Extraction, Knowledge base, Expert System 
ABSTRACT: 
Automatic change detection and data updating is a very important issue for keeping the temporal accuracy and currency of spatial 
data sets. Road networks are one of the most important parts of geographic database. Firstly two kinds of new algorithms for 
detecting feature changes that is buffer detection (BD) algorithm and double-buffer detection (DBD) algorithm are illustrated in 
detail. The corresponding buffer detection distance formulas are deduced theoretically. Then the change detection techniques 
between new map and old map are proposed. For change detection between new/old maps with same map scale the so-called buffer 
detection algorithm is employed and for new/old maps with different map scale a change detection expert system integrated with 
GIS environment is presented. Corresponding experiments results for detection algorithms are given in the paper. The main 
difficulty of automatic change detection for road network between new image and old map lies in two aspects: one is the tracing of 
unchanged road and another is the extraction of new road. For detection and tracing of unchanged old road, automatic detecting 
algorithms based on GIS information are proposed. For the extraction of new road, some new ideas and strategies including hybrid 
feature grouping techniques, automatic road recognition based on knowledge base, knowledge inference for road recognition, road 
re-grouping etc. are discussed. At last conclusions and future work are given. 
1. INTRODUCTION 
With the fast development of city and expansion of urban, road 
networks are prone to change and therefore they become main 
parts to be detected and updated. Road networks are one of the 
most important geo-spatial objects. Updating road network is a 
key work for updating geo-spatial information especially in 
developing countries. Automatic change detection of road 
networks is the first and crucial step. However, for a long time, 
manual operation based on visual interpretation is the primary 
method for updating road and other geographic data. Although 
extraction of road networks and pixel-level change detection 
(PLCD) are not fresh topics, systematic research for automatic 
feature-level change detection (FLCD) and object-level change 
detection (OLCD) for road networks is little reported in the past 
literatures (Peled,1998; Macleod,et.al, 1999; Sui,2002). 
For PLCD, pre-classification and post-classification are most 
important and widely applied methods. Aiming at the 
shortcomings of these methods, some researchers present the 
integration of many change detection methods. They includes 
Markov Random Field (Bruzzone,2000), neural network 
(Dai,1997), mathematic morphology(Maupin,1997), fuzzy logic 
(Dreshler,1993), based on knowledge (Wang,1993), based on 
GIS (Peled,1998) and so on. To solve a series of problems 
including accuracy, speed and automatic processing caused by 
traditional method that is first registration then change detection, 
Li (2002) presented a new idea and algorithm for registration 
simultaneous with change detection. As to application of 
change detection, some software companies including ERDAS 
have integrated some PLCD algorithms into basic function 
modules. However, manual processing is still the major method 
in these commercial software. Compare to PLCD, although 
relative research for FLCD and OLCD has been reported in the 
literatures, their contents mainly focus on extracting objects 
from image and a little care is given for change detection 
algorithms (Chalifoux.1998; Fan,1999: Darvishzadeh,2000). On 
the other hand, the rescarch for change detection between old 
459 
map and new map is very limited although it is more useful in 
real application. 
Indeed, although the research history for change detection is 
longer than 40 years mature and practical automatic change 
detection theory framework and technical system are not 
realized yet. For current change detection theory, stable theory 
basis and suitable evaluation standard and algorithms are lack. 
For current change detection algorithms, relative information 
detween new/old images 1s not fully utilized for PLCD and the 
discussion and research for FLCD is not enough. Aiming at 
these this paper focuses on the research for automatic FLCD. 
The contents are emphasized on two aspects: one is change 
detection for road networks between new map and old maps; 
another is change detection for road networks between new 
image and old map. 
The rest of the paper is organized as the follows. In the next 
section, the algorithms for detecting feature changes including 
buffer detection algorithm and double buffer detection 
algorithm are presented and corresponding buffer detection 
distances are deduced theoretically. In section 3 some new ideas 
and algorithms for automatic change detection between new 
and old map are given. And in section 4 corresponding 
strategies and ideas for automatic change detection for road 
networks between new image and old map are proposed. The 
last section summarizes and concludes with a discussion of 
future work. 
2. THE ALGORITHMS FOR DETECTING FEATURE 
CHANGES 
2.1 Buffer detection algorithm 
2.1.1 The principle of buffer detection algorithm 
For feature level change detection (FLCD), how to compare the 
difference of two features is a key problem. Feature comparison 
problem is often considered as shape matching. And in shape