Three-Dimensional Grouping and Information Fusion for Site Modeling from Aerial Images

This paper demonstrates the utility of data fusion when applied to the problem of site model reconstruction. We combined the results from hierarchical image matching, feature based building detec- tion, robust plane fitting, and heuristic assemble algorithms to form an accurate, robust site model reconstruction system. In the future, the techniques described here will be extended to more complex buildings, including gabled and curved roofs, by fitting the elevation data to a wider variety of geometric models. An additional goal is to use the partially closed chains as focus of attention mechanisms and to explore approaches for recovering surface structure between arbitrary sequences of corners and lines. Overall, we expect to continue the investigation of plausible strategies for grouping generic elements into complex structures and for simultaneously fusing information from multiple sources into coherent models. The results shown here are encouraging. The accuracy of the final reconstructions can be observed from the visually consistent renderings. Through the careful combination of primitive elements and special purpose strategies, we have the beginnings of an automatic, accurate, and functional system