Robust Model-Based Detection of Gable Roofs in Very-High-Resolution Aerial Images

This paper describes an improved version of our system for robust detection of buildings with a gable roof in varying rural areas from very-high-resolution aerial images. The algorithm follows a custom-made design, extracting key features close to modeling, such as roof ridges and gutters, in order to allow a large freedom in roof appearances. It starts by detecting straight line-segments as roof-ridge hypotheses, and for each of them, the likely roof-gutter positions are estimated. Supervised classification is employed to select the optimal gutter pair and to reject unlikely detections. Afterwards, overlapping detections are merged. Experiments on a large dataset containing 220 images, covering different rural regions with significant variation in both building appearance and surroundings, show that the system is able to detect over 87% of the present buildings, thereby handling common distortions, such as occlusions by trees.