e ——---
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
multiple sensors, and related information from associated
databases, to achieve improved accuracy and more specific
inferences that could be achieved by the use of single sensor
alone (Hall and Llinas, 1997). Integration of disparate data
sources is a problem that is being investigated for several years.
Methodologies for integrating different data sources can be
divided into two major categories, viz., sensor dependent and
those that are sensor independent (KrishnaMohan et al., 2000).
[n the former, specific image formation models are
incorporated into the analysis process. Solberg et al. (1994)
attempted to fuse LANDSAT-TM and SAR images using SAR
specific image formation model. This has the advantage of
using most appropriate features for analysing the image data.
Sensor independent approaches are based mainly on integrating
segmentation maps produced by analysing data sets
individually. Our study belongs to the second category.
Sensor independent approaches are based mainly on integrating
segmentation maps produced by analysing data sets
individually. For instance, region boundaries produced by
textural segmentation of SAR images can be merged with
intensity edges produced from visible/infrared images. Here it
is assumed that the segmentation procedures applied to
individual data sets are responsible for accounting for all the
sensor related issues and the integration module is independent
of the origin of various segmentation maps input to it. Image
fusion can be performed at three different processing levels
such as pixel or data level fusion, feature level fusion, and
decision or interpretation level fusion (Varshney, 1997; Pohl
and Genderen, 1998). Le Moigne and Tilton (1995)
integrated edge and region data for refined image segmentation
and applied it to segmenting LANDSAT TM data.
KrishnaMohan et al. (2000) suggested IRS image classification
by fuzzy k-means clustering and fuzzy set theoretic integration
of landuse /landcover data of two dates. :
The studies on change detection based on differencing
radiometrically-normalized images have been performed (Fung,
1990; Heo and Fitzhugh, 1999; Itthi and Wallapa, 2002). For
change-detection studies multitemporal data was used (Smara
et al, 1998; Bruzzone et al, 1999; Saraf, 1999). Remote
sensing data have also been integrated with GIS data for
environmental studies (Fegan, 2000; Rocha and Tenedorio,
2001; Xie et al., 2002.). Urban activities were estimated by
using thermal and large-scale vector maps (Michaelsen and
Stilla, 2001). Chu and Agarwal (1993) developed a system to
integrate edge and region maps to produce accurate
segmentation from multisensor data consisting of laser range,
velocity, intensity and thermal infrared images.
3. UTILIZED DATA
High-resolution remote sensing data over some Japanese cities
such as Tokyo (TABI, AISA) and Yokohama areas (ADS40)
were acquired incorporating GPS/IMU for geometric positional
accuracy preservation. ADS40 has the ground spatial distance
(GSD) of 20 cm for the multispectral and 10 cm for the
panchromatic range. Data of AISA (VNIR-Eagle) and TABI
sensors have 1.5m GSDs for this study (table.1). The images
used in this work currently mainly cover the DSM of two dates
generated from pixel matching of stereo pairs of ADS 40
digital airphotos, thermal data from TABI sensor and hyper-
spectral data from AISA sensor. The main applications of this
work will be in the area of image databases for property
evaluation and other government controlled applications. The
LIDAR data acquired was tested for change detection studies
and the results were not presented here due to lack of space.
902
4. PRESENT AND FURTHER APPROACH
From a data fusion perspective, an automated integration of
Remote Sensing and GIS was suggested. Data from several
sources (image and non-image) was used and fused in
combination with the goal of producing output information of
higher quality than obtainable from a single source. We aimed to
automate the data fusion and GIS analysis for application of
change detection study for property tax evaluation and others,
Our study encompasses:
e Capability of two date DSM data obtained from a pixel to
pixel matching was examined for broad as well as subtle
change-detection application.
e Integration of thermal images from TABI sensor and
hyperspectral images from AISA to best differentiate
man-made structures from vegetated features was attempted.
e NDVI extracted from AISA and thermal image fusion results
evaluated for the extraction of subtle man-made structures.
e Finally, GIS data such as road network to remove mobile
objects and building plan will be used to extract very subtle
changes in man-made structures.
e Importance of colour information other than geometric
characteristics of roofs was elucidated.
e To justify the present fusion analysis for property tax
application, the importance of extracted data from DSM
change-detection was shown.
e As a future study broad-change-detection results will be
fused with NDVI and thermal images.
Multi-temporal, multi-resolution and multi-sensor data were
fused to make thematic maps. We aimed to develop a novel
technology to integrate multi-date, multi-resolution, and
multi-sensor data for the purpose of subtle change detection.
We used DSM data obtained from LIDAR and ADS40 stereo
images, NDVI derived from hyperspectral sensor, road network
map along with other derived information.
Figure 1 shows the flow of our present and future methodology
involved. The figure 1 explains the idea about our fusion for
the automated mapping and GIS resources purposes. First
DSM data of two dates obtained from pixel-matching was used
to identify broad as well as subtle changes in man-made and
vegetated objects. Surface temperature from TABI images and
NDVI from AISA images were integrated. It is also planned to
use Road network area maps, roof parameters and colour
information of roof to detect subtle changes in features.
Ultimately we are interested in combining image and textual
database concepts into a single system in which both can be
used to reinforce the confidence in results of changes and
subsequent GIS queries. We initially considered the questions
of combining several feature detectors to improve their
performance and reliability on a wider variety of images.
Moving on from the idea of actually combining individual
detectors we are now experimenting with the use of ADS40
image based DSM, TABI sensor data Hyperspectral and
infrared colour images through some efficient image processing
algorithm and secondly through a GIS.
Inte
Fig
c
aut
To
par
Ro
dat
che
car
wit
lan
5.1
Sin
lan
ma
ma
bro
DS
sub
aut
For
im:
b):
cha
sho
der
cor
Inf
exi:
pla
/20
