FIG Peer Review Journal


Supervised Change Detection on Simulated Data Employing Support Vector Machines (4236)

Cristodoulos Psaltis and Charalabos Ioannidis (Greece)
Prof. Charalabos Ioannidis
National Technical University of Athens
9 Iroon Polytechniou St.
Corresponding author Prof. Charalabos Ioannidis (email: cioannid[at], tel.: + 30 2107722686)

[ abstract ] [ paper ] [ handouts ]

Published on the web 2010-01-14
Received 2009-11-19 / Accepted 2010-01-14
This paper is one of selection of papers published for the FIG Congress 2010 in Sydney, Australia and has undergone the FIG Peer Review Process.

FIG Congress 2010
ISBN 978-87-90907-87-7 ISSN 2308-3441


The increasing need for easy and cost effective updating of Geographic Information Systems (GIS), the wide availability of inexpensive high resolution data and the exponential increase in computing power, fuel extensive research in automatic methods for change detection of manmade objects. This paper illustrates a methodology to accomplish such tasks. The main idea of the proposed procedure follows the supervised classification paradigm. The first step is to layer the available data for the same region in different time periods. Then evaluate a number of predefined cues for the whole region and use some manually collected positive and negative samples to train a classifier. Finally this classifier can be used to assert change in the remaining data. The base data chosen are very high resolution orthoimages and digital surface models (DSMs) because they offer both the radiometric and geometric information needed for robust change detection. The classifier selected is the support vector machines (SVM) algorithm because it offers some significant advantages over relevant methods. These advantages include convergence to a global maximum, requirement of a small number of training samples and the availability of good open source implementations. In the paper, emphasis is given in testing the proposed strategy with simulated data, to access its validity and performance aspects. The simulated data were produced automatically with a program developed especially for this purpose. Noise is gradually imported to the testing data to make them more realistic. Noise can be a combination of random radiometric noise for the images, geometric noise for the buildings depicted in the images and finally height noise for the DSMs. Different setups were planned and implemented in all of which the results indicate that the proposed methodology has good performance.
Keywords: Geoinformation/GI; Photogrammetry; machine learning; change detection; Support Vector Machines