Building component detection by using context information and multiple cues for outdoor robot

This paper describes an approach to detect the components, specially the windows, of building for understanding and exploring environment of outdoor robot. Firstly, building surface and wall region are extracted then the wall region is binarized. The noises such as the thin connection, small regions are rejected. The context information and multiple cues such as the coordinates, aspect ratio and aligned coexistence are used to detect the windows. The proposed approach has been experimented with 150 surfaces comprising 1607 windows. We obtained 93.34% detection rate. Furthermore, the method also extracts the candidate regions for detecting the doors of building.