OBJECT BASED CHANGE DETECTION USING TEMPORAL LINKAGES P. Hofmann a, *, T. Blaschke b a Austrian Academy of Sciences, Institute of GIScience, peter.hofmann@oeaw.ac.at b University of Salzburg, thomas.blaschke@sbg.ac.at KEY WORDS: Change detection, Object Based Image Analysis, object linkage, temporal relationships, space-temporal modelling ABSTRACT: Change detection plays an important role in GIScience. Using appropriate methods of change detection allows us to observe, detect and analyse spatial processes which took place in the past. Furthermore, it enables us to understand processes in more detail, develop models and predict potential future situations. Remote sensing data as data source for change detection has the advantage of imaging the earth’s surface as is just using electromagnetic radiation. However, using remote sensing data as the basis for change detection has always been difficult since a lot of knowledge from image processing and remote sensing is necessary in order to detect and outline relevant changes. Space-temporal knowledge about the object categories to observe is necessary in order to determine which changes are the result of the natural space temporal behaviour and which are a relevant change. By linking image objects of images taken at different dates via the time axis it is possible in principle to observe and assess their space- temporal behaviour and to decide whether this behaviour is natural or relevant in terms of a change or not. * Corresponding author. 1. INDRODUCTION 1.1 Change detection Change detection based on geo-data is certainly one of the most important and challenging tasks in the GIScience domain. Focusing on multi-temporal remote sensing data, change detection methods are used in order to point out and document changes relevant for diverse application domains. Typical examples of such applications are: mapping processes as like urban sprawl, desertification or dry-out of lakes (Jat et al, 2008; TRIPATHY et al., 1996; Collado et al. 2002; Diouf et al., 2001). That is, when doing change detection with multi- temporal remote sensing data at least two images of a given region taken at different dates (t0 and t1) are compared and differences relevant for the application domain are mapped. For detecting changes over longer periods and with data measured at more than two dates (t0 … tn) the term monitoring is commonly used. In this context, a critical point for change detection using remote sensing data is to detect only the relevant changes. 1.2 Object based change detection A rather simpler approach is to independently segment all images taken at t0, t1 or tn, virtually overlay them and identify corresponding image objects. However, this method presumes a spatial overlap of the temporal corresponding objects in order to establish a respective connection between them. Nevertheless, this way it is possible to observe and document the objects’ courses and to decide whether the observed behaviour is normal or a change. 2. METHODOLOGY 2.1 Image segmentation and object linkage In order to perform an object based image analysis using linked objects it is necessary to generate image objects which are timely independent. That is, each image of t0, t1 and tn needs to be segmented independently. For this purpose we have been using the software eCognition 8.7 by Trimble Germany, which allows to segment images on several scale levels and additionally to link spatially coherent objects using so-called maps. Each map in this particular case represents a single date and image, respectively. The software even allows loading image sequences. It automatically generates for each time frame a respective map. The map concept can also be used to independently segment images of different sensors and link corresponding objects (fig. 1). Proceedings of the 4th GEOBIA, May 7-9, 2012 - Rio de Janeiro - Brazil. p.634 634