Object-based Classification of Integrated Multispectral and Lidar Data for Change Detection and Quality Control in Urban Areas

Digital spatial data are underlying strong temporal changes. The typical approach of updating these changes is to check the data manually by superimposing them on up-to-date orthoimages from aerial or satellite camera systems. The update cycles of large data sets are in the range of several years because the manual inspection of the data is very cost and time consuming. However, spatial analyses for planning purposes are only meaningful if they are calculated with up-to-date data. Automatic data acquisition, update and quality control procedures are needed in order to provide up-to-date geo-databases. In this paper an approach is presented that increases the quality of the interpretation process on the one hand by using already existing data from Geographical Information Systems (GIS) as prior information and on the other hand by combining image data from different sources. The approach is based on the evaluation of automatically derived training data sets from existing GIS data. Therefore the approach is fully automatic and no human interaction is necessary. The result is not only a classification of the objects but also a distance vector that describes the quality of the classification. This distance vector can be used for an automatic evaluation of the automatic image interpretation as well as for automatic quality control of already existing GIS databases.