SEMI-AUTOMATIC DETECTION OF LAND USE CHANGE FROM DIGITAL AERIAL PHOTOS
Hiroshi Masaharu, Sakae Matsumoto, Hidekazu Hoshino
Geographical Survey Institute, Japan
Yutaka Ohsawa
Saitama University, Japan
Yoichi Oyama, Nobuhiko Ogino, Hiroyuki Shimizu
Association of Precise Survey and Applied Technology, Japan
Commission IV, Working Group 3
KEYWORDS: Change Detection, Land Use, Revision, Updating, Mapping, DEM/DTM, Matching, Softcopy Photogrammetry
ABSTRACT
It is necessary for map revision work to find out land use changed areas from aerial photos. Several attempts utilizing computer
processing of digital aerial photo images have been made and examined aiming at increasing efficiency of change detection. Among
them, three methods and their results are reported in the paper. One is to compare old and new DEMs derived from automated stereo
matching. It was possible to detect new construction of houses. But further study is required to apply the method to various change
patterns. Another method is to pick up roads by color classification of color aerial photos. By comparing the results of road
extraction, new roads could be visualized. The third method is to use existing digital map data to mask out unnecessary area from
aerial photo image and thus to enhance the efficiency of manual change detection. Some increase of efficiency was observed but with
the slight increase of omissions of detection. Further elaboration is required for practical application of these methods. It is
considered that combination of multiple data sources together with effective combination of multiple methods is necessary for
improvement.
1. INTRODUCTION
Digital cartographic data corresponding to large scale maps
such as 1:2,500 have been prepared in many municipalities in
Japan nowadays. The wider the covering area grows, the more
the needs to update digital maps are recognized. To increase the
efficiency of revision work of digital cartographic data is one of
the most important research target in mapping technology. On
the other hand, digitized aerial photo image has become readily
available owing to the development of digital photogrammetric
instruments. This situation motivates us to develop techniques
of automated digital map revision through computer processing
using newly taken aerial photo images. However image
understanding of aerial photos by computer is still challenging
field of research and complete automation do not seem to be
accomplished in near future. Therefore we set up our research
target as change detection from aerial photo data to help works
of human operators. Several attempts have been made in this
research project. Some of the results are described in detail in
the paper.
The methods so far tried can be classified as follows: (1) to limit
possible candidates of changed areas by comparing the results
of some kind of classification obtained through digital image
processing of aerial photos of two different time; (2) to increase
efficiency by limiting the area to be examined by human
Operator using existing digital cartographic data; (3) To enhance
operator’s visual recognition with various digital image
processing to aerial photos. As for the class (1), we tried and
examined many methods, such as a method to use height
information obtained through automatic stereo matching, a
method to classify possibility of change with correlation
coefficient between aerial photo image of two different times, a
Classification method using color information of color aerial
photos, comparison of power spectrum of land use patterns after
Fourier transform and so on. Regarding the class (2), masks
corresponding to roads and existing houses were made from
digital cartographic data and were overlaid on new aerial photos
so as to help to detect newly constructed houses. Various edge
detection and edge enhancement methods including emboss
filtering have been applied and compared in the class (3).
2. CHANGE DETECTION BY THE COMPARISON OF
NEW AND OLD DEM FROM STEREO MATCHING
Digital photogrammetric instruments are nowadays becoming
popular. They have capability to produce minute DEMs (digital
elevation models) by automatic matching of stereo pair aerial
photos. The differénce of DEMs derived from old and new
photos are used as the indication of change.
2.1 Method
2.1.1 Test site and used materials and system: The test site
was chosen in Fukuyama city, Japan where aerial photos of
1987 (1:12,500 Color) and 1994 (1:25,000 B/W) are
available as well as an old map of 1987 on a scale of 1:2,500.
The test site is shown in Figure 1 (Map) and Figure 2
(Orthophoto). The digital photogrammetric system used here is
named PyramidStereon (Oyama, 1996) which is composed of a
personal computer, stereo viewing system, peripheral devices,
and software.
2.1.2 Method: Aerial photos are scanned in 50 um sampling
interval for old photos and in 25 um for new photos
respectively, both corresponding to 0.625 m on the ground.
DEMs with 2.5 m grid interval were generated from both the
photos by stereo matching. DEM was also generated from
contour lines of the 1:2,500 topographic map. Another DEM
from old aerial photos was made by a human operator using an
analytical stereo-plotter. The height on the roofs or tree tops
547
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
