Akademik Veri Yönetim Sistemi
8th Symposium on Networking the World with Remote Sensing of ISPRS-Technical-Commission, Kyoto, Japonya, 9 - 12 Ağustos 2010, cilt.38, ss.1063-1068
An approach was developed for the automatic extraction of the rectangular and circular shaped buildings from high resolution satellite imagery using Hough transform. First, the candidate building patches are detected from the imagery using the binary Support Vector Machines (SVM) classification technique. In addition to original image bands, the bands NDVI (Normalized Difference Vegetation Index), and nDSM (normalized Digital Surface Model) are also used in the classification. After detecting the building patches, their edges are detected using the Canny edge detection algorithm. The edge image is then converted into vector form using the Hough transform, which is a widely used technique for extracting the lines or curves of the objects. The vector lines and curves that represent the building edges are grouped based on perceptual groupings, and the building boundaries are constructed. The proposed approach was implemented using a program written in MATLAB (R) v. 7.1 programming environment. The experimental tests were carried out in the residential and industrial urban blocks selected in the Batikent district of Ankara, the capital city of Turkey using the pan-sharpened and panchromatic IKONOS images. The results obtained indicate that the proposed building extraction procedure based on SVM and Hough transform can be effectively used to extract the boundaries of the rectangular and circular shaped buildings. For the industrial buildings, we obtained quite satisfactory results with the average Building Detection Percentage (BDP) and the Quality Percentage (QP) values of 93.45% and 79.51%, respectively. For the residential rectangular buildings, the average BDP and QP values were computed to be 95.34% and 79.05%, respectively. For the residential circular buildings, the average BDP and QP values were found to be 78.74% and 66.81%, respectively.