Urban road network extraction from very high resolution RGB aerial images and DSM data

In this paper, we address the problem of automatic road network extraction in urban areas from very high resolution RGB aerial imagery and the Digital Surface Model (DSM) data. We use an extremely high-resolution image in which the road signature, such as cars, road lines, zebra crossings and the like, can be seen in detail. In this work, we first establish the location of the zebra crossing based on circle mask template matching in an aerial image. The location of these zebra crossings represents the starting point of the road and we can obtain the elevation from the corresponding DSM data. In the DSM data, the elevation of the road and the building is differ significantly; therefore, we expand the starting point based on a local thresholding and improve it using a seeded region growing, to create the initial road region quickly. A road line filter based on morphological opening operations are then carried out to produce the road line. The experimental result shows that the proposed method is run quick enough with good accuracy.