Classification of the Approaches

  Compared to the ultimate goal, the extraction of objects for GIS databases or for site models, the results of all presented approaches are far from being useful in practice at least for images of different characteristics and complex contents. Nevertheless, for restricted domains the most recent approaches are getting closer to being useful in practice. This may be the reason why quantitative performance evaluation has received more interest recently. Ruskoné and Airault 1997 [25] and Wiedemann et al. [38] show for simple scenes a completeness and correctness of up to 80% and 95%, respectively, compared to manually measured ground-truth.

The most sophisticated approaches are the most recent ones. They are [24,25] and [3,29,2,17]. De Gunst and Vosselman's approach [6] is the one using GIS information to the greatest extent.