Building roof boundary extraction is an important issue for the application of urban environment studies. Most traditional approaches of building roof boundary extraction use images which work on the basis of the spectral features. However, there are some limitations when the roofs contain complex colour and complicated shape. Thus, in this study, we utilize Object-Based Image Analysis (OBIA) to extract roof boundary from true-orthoimage and Digital Surface Model (DSM) that are generated by high spatial resolution aerial images. Both true-orthoimage and DSM can reach decimetre accuracy and contain detailed building information. In the experiment, we choose a part of Tainan city as our study area which covers various types of buildings and tall trees. The building roofs are quite different as well which can be used to evaluate the potential of the proposed method. The major procedure of OBIA is to segment the image into objects based on spectral and geometrical information, then classify the image objects into different kinds of land cover according to several object features and the developed rule sets. In the case study, the vegetation is classified by vegetation indices at first. After that, the ground and buildings could be separated by height information from DSM. In order to extract roof boundary, the classes of land cover should be removed except for buildings. Then, the roof boundary could be extracted from building class based on the information from the true-orthoimage and DSM.
展开▼
