The analysis of possible use of multi sensor system data for automatic building extraction

The monitoring of construction activities and determination of current situation is important for control and decision making process of urban management. In cities, the need for accurate, up to date and fast data for building is steadily increased with urban transformation projects. Especially, in cities located within seismic belts, automated building extraction without any requirement for classical measurement methods and photogrammetric production processes has become an important subject. In this study, automatic building extraction possibilities are investigated with multi sensor system which are consist of LiDAR (Light Detection and Ranging), GPS/IMU and digital camera positioned on the same platform using object-oriented rule-based classification method. For this purpose, current methods used for automatic building extraction are investigated and international and national applications are examined. The problems in automatic building extraction are mentioned using pixel based and object based classification methods. The common problem in automatic building extraction is misclassification of building and other objects (vegetation, ground etc.). In this study, the approach is developed based on the determination of the differences between building and other classes than improvement of the classification using these differences. The accuracy of the classification is improved with creating additional class including vegetation, building border, shadow and ground during the classification process. Digital surface model (DSM), orthophoto and intensity image were produced with the help of multi sensing system data (3D point cloud, intensity and aerial photographs). The additional data including Hough, Canny and band ratio images were produced with the help of the data comes from multi sensor system and utilized for improvement of the targeted building class. The rule set were created with developed approach and tested at two study area in USA and Sivas Turkey. The accuracy assessment of the automatically extracted building class in study area (1) and (2) was performed using created rule set with developed approach.The obtained accuracy for the automatic building extraction was 96.73% for completeness and 95.02% for accuracy for study area (1) and 80% for completeness and 85% for accuracy for study area (2). The faced problems during automatic building extraction process were stated and the suggestions for the solving faced problems were presented.Key words: Automatic building extraction, LiDAR, multi sensor system, rule-based classification, segmentation, fuzzy logic.