Automatic Road Extraction of Urban Area from High Spatial Resolution Remotely Sensed Imagery

For the significance of road information to the city management, urban roads are subjects of great concern to be extracted from remotely sensed images. With the availability of high spatial resolution images from new generation commercial sensors, how to extract roads quickly, accurately and automatically has been a cutting-edge problem in remote sensing related fields. Present main approaches of automatic road extraction cannot fully exploit the spectral information of roads in the imagery and get the required accuracy. Considering the road knowledge, we develop a new approach to extract roads accurately and automatically, in which spectral and geometric features of roads are both considered and represented. The approach contains three steps: rough classification, which enhances the full exploitation of spectral contents and ensures the continuity of roads for the following steps; road connection algorithm, which extracts road skeletons roughly; and result grooming, which includes connecting, smoothing, linking and produces the final result. We take Beijing City as a study case and use QUICKBIRD image to implement the approach. As results turn out, the approach achieved a satisfactory accuracy of 96.7% on main roads while 74.3% on secondary roads and proves to be of high practical value. * Tel.: +86-10-58801865; fax: +86-10-58805274. E-mail: [email protected].