Building Change Detection from Digital Surface Models and Multi-spectral Images

A new method for building change detection from Digital Surface Models (DSM) and multi-spectral images is presented. The DSM can be generated from Airborne Laserscanner (ALS) data or by image matching techniques. From the multi-spectral image, the Normalised Difference Vegetation Index (NDVI) is computed and used in the change detection process. The workflow of the method consists of two stages. In the first stage, the DSM, the NDVI, and surface roughness parameters derived from the DSM are used in a classification technique based on the Dempster-Shafer theory for data fusion. In the case of ALS data, the height differences between DSMs created from first and last pulse data can also be considered. This technique is used to detect buildings. In the second stage of processing, these building detection results are compared to an existing building data base, and changes between the existing data base and the new data set are determined. This paper focuses on the second processing stage, in which the actual change detection is carried out. The method is designed to classify buildings and building parts as being confirmed, changed, new, or demolished. The change detection method considers the facts that the original data and the building detection results can have a different topology and that small differences between the data from the two epochs might be caused by different levels of generalisation or errors caused by a misalignment or insufficient resolution of the sensor data. Examples for the performance are given using DSMs generated both from ALS data and by image matching, highlighting the different properties of these data for building change detection.