DETECTING CHANGE IN ROAD NETWORKS USING CONTINUOUS RELAXATION LABELING S.Gautama, A.Borghgraef TELIN, Ghent University, St.Pieternieuwstraat 41, B-9000 Gent, Belgium – sidharta.gautama@ugent.be KEY WORDS: change detection, GIS, matching, registration, artificial intelligence ABSTRACT: In this paper we examine a system based on computer vision for automated detection of change and anomalies in GIS road networks using very high resolution satellite images. The system consists out of a low-level feature detection process, which extracts road junctions, and a high-level matching process, which uses graph matching to find correspondences between the detected image information and the road vector data. The matching process is based on continuous relaxation labelling. It is driven by spatial relations between the objects and takes into account different errors that can occur. The result is an object-to-object mapping between image and vector dataset. The mapping result can be used to calculate a rubbersheeting transformation which is able to compensate for local distortions. A measure of change is defined based on the number of null assignments. We show how combined with a condition to characterize acceptable errors this measure is useful and reliable to characterize inconsistencies between image and vector data. 1. INTRODUCTION A major challenge in the production and use of geographic information is assessment and control of the quality of the database. The rapid growing number of sources of geospatial data pose severe problems for integrating data. One of the major challenges for content providers to face is the problem of upgrading their current databases to a higher accuracy and ensuring the quality of the information. Current techniques cannot support this in a cost-effective way due to the necessary manpower. Automated detection of change and anomalies in the existing databases using very-high- resolution (VHR) satellite images can form an essential tool to support quality control and maintenance of spatial information. The main problem to address is the difference in data representation. To be able to compare geospatial vector data with images, the information in the images needs to be described in terms of object features. Automatic detection of man-made objects however is a difficult problem. Shadow, occlusion and variety in appearance all give rise to a fragmented and imprecise description of the image content, especially if consistent detection is required over large datasets. Making a reliable statement about the quality of the GIS data requires knowledge about the performance of the different object detection techniques. Our focus is the development of such reliable and predictable techniques. Rather than producing results on a selected number of images, we wish to be able to characterize the performance of the detection. This performance characterization is an essential step towards reliable tools within an operational environment. The proposed system for change detection is based on feature based spatial registration, where detected features in the image are registered to corresponding features in the vector data. The system consists out of two stages: 1) a low-level feature detection process, which extracts roads and junctions using an improved ridge detector, and 2) a high-level matching process, which uses graph matching to find correspondences between the detected image information and the road vector data. The graph matching process, based on continuous relaxation labeling, is driven by the spatial relations between the features and takes into account different errors that can occur (e.g. spatial inaccuracy, data inconsistencies between image and vector data). The matched features can be used to calculate a rubbersheeting transformation between image and vector data, using triangulation. Such a transformation is able to compensate for the local distortions that can occur between the datasets. Additionally can the object-to-object mapping be used to define measures of change between datasets.