Road networks derived from high spatial resolution satellite remote sensing data

There is a strong demand for accurate and up-to-date road network information. Road network knowledge is crucial for the creation and the update of maps, geographic information system (GIS) database, transportation or land planning. For local authorities, cartography of the road network is needed for urban planning, dirty water collection through gutter network (most often located under roads), traffic flow analysis or pollution mapping. Closely related applications are geo-marketing, electricity and telecommunication networks, databases for car navigation ... Currently, road network cartography is essentially done by human interpretations of high resolution aerial images and additional in situ information. This is a long and tedious work that requires to be done again for each update of the road network. High spatial resolution imagery is recently available for civilian applications and reveals the very fine details of the imaged area. Examples of high resolution satellites are SPOT 5, Ikonos, Quickbird, OrbView or EROS. The term ‘high resolution' is relative and refers to satellites with spatial resolutions better than 5 meters in the panchromatic channel (one can even talk about very high resolution when the image resolution is better than 1 meter). The current availability of high spatial resolution images represents an undeniable asset to Earth observation. The urban environment, that is the most difficult context because of its high complexity and information density, could benefit the most from high resolution imagery (Puissant and Weber, 2002). In addition to the increased precision for the road detection and location, high resolution satellite imagery can be used for numerous cases where the access to the studied area is difficult: administrative constraints, authorization to overfly the area, conflicts, wars or natural catastrophes... Moreover, satellite means is significantly cheaper than aerial or in situ data acquisition campaigns. As promising as it is, the use of high resolution images for road extraction induces a change in the road representation, and a significant increase of noise. Moreover, quantitative assessment of the results has to be redesigned when dealing with such images. In this chapter, a new method suitable for high resolution images is proposed. Originally designed for urban area, this method can naturally be applied on easier cases such as rural or semi-rural areas. The chapter is structured as follow: the change in the road representation induces by high resolution imagery is first presented. A short survey on road extraction with the evolution from linear to surface models for road is proposed (section 2). A new method for extracting road networks from high spatial resolution images is then described. It models roads as a surface and is built on cooperation between linear and surface representation of roads. In order to overcome local artifacts, the method makes use of advanced image processing tools, such as active contours and the wavelet transform (section 3). An example of application of the method on a high resolution image from the Quickbird satellite is proposed. The result is quantitatively assessed compared to human interpretation (section 4). This chapter concludes with a discussion on the principal benefits of the method and on future prospects (section 5).