Building Footprint Extraction from Very-High-Resolution Satellite Image Using Object-Based Image Analysis (OBIA) Technique

Abstract

Building footprints are an important input in several urban applications such as master plan preparation, development of 3D city building models, rooftop solar energy potential estimation, tax compliance evaluation, or study of population distribution in cities. The very high spatial resolution (VHR) image is invariably required for the extraction of building footprints. However, the conventional pixel-based approaches have limited success in building footprint extraction owing to inherent heterogeneity of the urban environment. In this research, we have therefore applied Object-Based Image Analysis (OBIA) for building footprints extraction from Cartosat-2 series data. The image was segmented into several objects on the basis of spectral and spatial homogeneity of pixels. The objects were thereafter classified using nearest-neighbor classification approach. Finally, these classified objects were further subjected to segmentation into smaller objects and classified using decision-rules. The combination of supervised nearest-neighbor classification with decision-rules resulted in an accuracy of over 82.5% in the extraction of building footprints. The results of the study will be used to develop a 3D city model of Ahmedabad city and assess the changes in the built-up volume in the city.