Characterization of Strategies

  Fischler et al. [8] fuse the result of several line extractors while McKeown and Denlinger [19] combine edges and profiles (cf. Table 3). Baumgartner et al.'s approach [3] is based on two or more scales. A grouping of medium complexity in 2D is used by [36,40,6], and with even more complexity by [3]. Control is mostly data-driven. In [6] the GIS-data is used model-driven, which, together with the data-driven extraction, results in a mixed control. McKeown and Denlinger [19] is based on a mixed strategy for the analysis of disturbances while [3,17] use hypothesize and verify. For search, the F^*-algorithm is used in [8], while [25,29] are based on heuristic approaches to find optimum paths. Internal evaluation is simple for most approaches or not used at all.

In summary, there is a change from simpler to more complex strategies. The same is true for the internal evaluation, where recently more and more complex modeling is used. For roads, specific methods for search are used.

 

Table 3: Characterization of the Strategies for Road Extraction

Approach	Fusion	Grouping	Control/Search	Evaluation
Roads as Lines in Low-Resolution Imagery: Fischler et al. 1981[8]	several line extractors	--	data-driven; F*	simple
Road-Grouping and Network Creation: Vasudevan et al. 1988 [36]	--	2D; medium	data-driven; --	no
Road Detection and Tracking: Zlotnick and Carnine 1993 [40]	--	2D; medium	data-driven; --	no
Road Detection and Tracking: McKeown and Denlinger 1988 [19]	edges + profiles	--	mixed (disturbances); --	simple
Road Network Construction by Interpreting the Local Context: Ruskoné 1996 [24,26,25]	--	--	data-driven; optimal paths	simple
Multi-Resolution, Semantic Objects, and Context for Road Extraction: Baumgartner et al. 1997 [3,29,17]	2 or more scales	2D, complex	hypothesize and verify; optimal paths	simple
Utilization of Maps and GIS: de Gunst 1996 [6]	--	2D; simple	mixed (GIS); --	medium