Automatic high-rise building extraction from aerial images

Abstract

The diversity and complexity of high-rise building shapes make it very difficult to extract them automatically with traditional modeling and matching algorithms. In this paper, starting from detection of vertical lines as the key evidences, a strategy for automatic extraction of high-rise buildings in monocular high-resolution aerial images was developed under the evidential reasoning methodology. In all phases of this strategy, much useful information in object space and image space were combined, and multiple advanced techniques of computer vision and signal processing were incorporated to reason out locations, heights and roof boundaries of high-rise buildings step by step. Sequentially, 3D high-rise building reconstruction was achieved.