Structural analysis of multiple building for mobile robot intelligence

This paper describes an approach to detect and recognize multiple buildings in the real urban environment. Visual and geometrical features of line segments are used to classify a building with the other objects such as sky, tree, and roads. The line segments and their two neighbored regions are extracted from detected edges of image. A model of line segment (MLS) is constructed by color information of neighbored regions. This model is used to roughly rule out the line segments of non-face pattern. The rest is clustered into dominant vanishing points (DVPs) including one vertical group and five horizontal ones in maximally. The intersections create the criteria to identify the facet of building. The proposed method is used for two applications. The first performance is for detecting buildings of 100 image database which contains non-building images of outdoor scene. The second experiment is recognition of a set of 80 test images from 500 stored images in database which consists of multiple buildings. All images are taken in Ulsan metropolitan city in South Korea under differences of viewpoint, daytime, camera system and weather.