Development of a graph-based approach for building detection

Abstract

This paper describes the development of an approach for automatic building detection from aerial imagery. The development has concentrated on two aspects: to systematize the approach and to develop robust algorithms. The whole process of building detection is divided into four small stages of line extraction, line-relation-graph generation, building hypothesis generation, and building hypothesis verification. Robustness has been achieved by considering only the mathematical and geometric relations between lines in the course of generating building hypotheses. By preventing any assumptions related to intensity values of an image, this approach can be applied to more wider range of imagery. Graph structure was used for building hypothesis generation and it was shown that the use of graph structure could simplify the problem of building detection in a great deal. This paper reports several examples of building detection with the approach developed. The performance of each example is evaluated quantitatively with six quality measures defined here. The results of performance assessment support the robustness of the approach developed.