DETECTING CHANGES TO TOPOGRAPHIC FEATURES USING HIGH RESOLUTION
IMAGERY
D. A. Holland, C. Sanchez-Hemandez, C. Gladstone 3
‘Research, Ordnance Survey, Romsey Road, Southampton, UK -
(David.Holland, Carolina.Hemandez, Catherine.Gladstone)@ordnancesurvey.co.uk
Commission VI, WG VI/4
KEY WORDS: Change Detection, Digital Mapping Camera, Cartography
ABSTRACT:
Detecting changes to topographic features is one of the major tasks of a national mapping agency. At Ordnance Survey, Britain’s
national mapping agency, the process of change detection has traditionally been a largely manual, labour-intensive task. One of the
goals of the Research department is to develop automatic and semi-automatic change detection processes which could be developed
into production systems. This paper describes the research undertaken, concentrating on the detection of changes to the built
environment, specifically new and demolished buildings. The research has progressed from a set of potential methodologies;
through research trials of different software packages and different change detection methods; to the adoption of one of the methods
in a full production trial.
1. INTRODUCTION
1.1 Background
of the image, hence missing out one comer of the image and
therefore miss any changes in that comer.
The detection of change is one of the most important aspects of
the work of a mapping agency. In the case of Ordnance Survey,
Britain’s national mapping agency, this is still largely a manual
process, relying on the observations of photogrammetrists and
field surveyors and the notifications provided by external
bodies. In the field, a network of approximately 300 surveyors
is engaged in the process of updating the national topographic
database, used in the production of the flagship
OS MasterMap® large scale topographic data product. Part of
the field surveyor’s task is to note any changes to the natural or
the built environment, in order to mark these changes for
subsequent data capture. In addition to this, local authority
planning departments provide information on planning
applications which may affect buildings and other urban
features. Major house builders inform Ordnance Survey of new
developments, while a further source of intelligence on changes
in the landscape is provided by a commercial survey
organisation, which supplies information on changes which will
affect Ordnance Survey’s products.
The final source of changes is a group of image interpreters in
the Photogrammetric Survey department, whose job is to
scrutinise orthorectified digital aerial photography and mark up
any changes. These changes are subsequently captured in
stereo using digital photogrammetric workstations. In spite of
all the other sources of information, this last phase uncovers
many changes which have, up to that point, gone un-noticed.
Manually scanning through the images is a laborious task, and
one which requires concentration and a disciplined
methodology. In an investigation carried out by the Ordnance
Survey Research last year it was found that when scanning an
image from top left to bottom right, constantly zooming in and
out, it can be difficult to keep track of which areas of the image
have already been looked at. In some cases, a tendency was
found for the image interpreters to gradually migrate to one side
1.2 Automating change detection
In order to help the data collection process, automation of the
workflow could be achieved at various points. Fully automatic
feature capture has been a long-term goal of several ISPRS
working groups, but it is unlikely to be completely realised in
the near future. Semi-automation of the capture of topographic
objects is a feasible option - especially in areas where many
features are repeated (such as housing estates where every
house is almost identical to its neighbour). In Great Britain,
however, this is very seldom the case. The continuous nature of
topographic data revision usually means that changes are very
local in nature, often involving only one or two features within
an area of many un-changed features.
Automation of the change detection process is more feasible,
especially if the system is only required to detect that a change
has occurred, without having to identify exactly what that
change is. In recent years there has been continued
improvement to photogrammetric and image processing
software and the hardware on which they run. This has enabled
systems to become more automated, either by aiding the
interactive data capture process or by performing a pre
processing step on images before the human operator becomes
involved. The nature of the workflow at Ordnance Survey
lends itself to the latter process, in which an automated process
would identify potential changes, then an operator would
confirm or reject each change. It was envisaged that such a
process would save a considerable amount of time in the data
collection workflow.
There has been much research in the last two decades on
techniques to automatically detect changes between images
taken on different dates. Such techniques can be of value, but
often they highlight many changes which are of no interest -
such as minor changes to vegetation, or changes due to the