Semi-automatic approach toward mapping of flat-roofed buildings within a non-stereoscopic environment

Abstract

In this paper, algorithms for semi-automatic flat roof extraction are described which focus on extraction from panchromatic medium scale aerial images with a known model solution. The algorithms developed are based on a 2D approach to solving the 3D reality. In order to initialise the extraction, the operator needs to point, in the left image space, within a typical radiometric area of the desired roof to be extracted. This process defines the algorithm as semi-automatic and from here on the process is fully automatic. The next stages include: extraction of a bounding contour in the left image space, calculation of the approximated height and transformation of the manual pointer from the left image to the right image space, extraction of a bounding contour in the right image space, simplification of the contours, an iterative process for precise height extraction and final mapping of the building. These algorithms can be employed in two ways, either as part of a fully automatic process when the pointers are extracted automatically, or standing alone so as to enable a new approach for semi-automatic extraction within a non-stereoscopic environment, without using 3D spectacles. In order to examine the efficiency of the algorithms, a semi-automatic application for flat roof extraction was developed. Based on panchromatic images at a scale of 1:42 000 and scanned at 14 μm resolution, the results obtained are satisfactory with accuracies of 0·97 and 0·46 m for the planimetric and altimetric measurements, respectively. This paper presents the algorithms, the experiments and the results. © 2007 The Authors. Journal Compilation 2007 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd.