Identifying Key Features in a Building Using a Single Uncalibrated Camera

This paper describes the application of image processing to the automation of external building inspection. A robot is used to locate defects which need to have their positions accurately recorded. The first stage in this process is to locate the robot on the building and a method is proposed where a single image is all that is required to find the robot. A number of edge detectors have been examined for their edge quality when applied to images of buildings. It has been found that the choice of camera is important as different types handle sharp transitions differently. The commonly employed Hough Transform is shown to be rather unsuitable for finding lines for this type of image and an alternative, which links pixels, is used to provide improved feature location. Results are given showing how significant figures are highlighted. A comparison is made between a simple model, a building with few distinct features and a complex building consisting of many windows, balconies and decorative panels.