Probability based road network detection in satellite images

Abstract

Road network detection is the process of detecting and extracting the road network from very high resolution satellite and aerial images. It is essential for many applications like map generation and updating. To do this road network detection, resolution of satellite and aerial images plays an important role. If experts try to label the road pixels manually, it will take more time and will lead to errors. Hence an automatic method is proposed here. Major operations of the proposed system are road network detection, estimation of road center pixel and road shape extraction. First, edge pixels are detected. Then, they are refined. Based on probability, road center pixels are estimated using edge pixels as observations. Next, road shape is extracted from the estimated center pixels using graph theory. The proposed method is tested on satellite (Quick bird and Ikonos) images. Obtained results indicate that the proposed method works well with 94% of accuracy when compared with the one existing in the literature. This work can be envisaged as a potential contribution to the science of automatic road network extraction from high resolution imagery.