Extracting Road Network by Excluding Identified Backgrounds from High-Resolution Remotely Sensed Imagery

摘要

The road network is one of the most important elements in navigable maps and location-based services, but the information on road networks is hard to acquire and update. This study aims at developing a new strategy to extract road networks from remotely sensed imagery, taking the bare soil, water, vegetation and buildings as the background while viewing the road network which is being targeted as the foreground. The background vegetation can be identified by its diagnostic spectral characteristics, and water also can be detected by its spectral features but the extraction of water is usually mixed with some information from shadows. Bare soil can mostly be detected and extracted, but the extraction result is sometimes mixed with red signature from red-coloured roofs. Building extraction was the most difficult step in the experiments conducted in this study, although most of the roofs were extracted successfully. The resulting extraction of the road network from the background showed that majority of main roads were extracted successfully, especially in the area with low building density. The results also demonstrated the proposed processing strategy has issues when extracting backgrounds and could be limited when it applies to more complicated areas.