Automatic extraction of road maps: Determining the sealed surface

This paper proposes an unsupervised method to obtain road maps from highly resolved (better than 1 m) panchromatic images. As a starting point it is assumed that an incomplete skeletal representation of the road map, i.e. the basic road network, is available. For example this can easily be gained through a straightforward thresholding. In the first step of the road map creation the network is completed using a maximum likelihood extrapolation approach. In it series of evaluations it was shown that this increases the network completeness, on average, by slightly over a tenth of the actual road network. In the second step the network is converted to the road map, i.e. the representation of areas that are part of the actual roads. Again, a maximum likelihood approach was applied with its parameters described through the local neighbourhood. In total completeness and correctness of more than 90% and 95%, respectively, were achieved.