Computational Aspects: How Landmarks Can Be Observed, Stored, and Analysed - Landmarks: GIScience for Intelligent Services - page 162

A first broad distinction is made between approaches to landmark identification
and to landmark integration. Some approaches focus on properties of the landmark
candidates—how do they differ from other geographic objects in their surrounding?
This corresponds to a static view on landmarks, similar to what Klippel describes as
structure in wayfinding [ 24 ] . Other approaches account for the location of landmark
candidates along a route to assess their suitability, which corresponds to a dynamic
view or function in Klippel's terms. Further, approaches differ in the (assumed)
source of data they use. Some approaches use 'classical' spatial databases of the
kind attached to a typical GIS. Some use some web harvesting technique to source
data from the web, either from general websites or from web catalogs, such as the
yellow pages. Others do not specify where their data comes from, which has been
marked as abstract in Table 5.7 . Approaches also may differ in the geometry that
landmark candidates can have. A majority assumes them to be point-like, however,
some also consider more complex geometries (e.g., polygons). Finally, approaches
may aim to identify individual objects ( instances ) or categories of objects ( types ).
If we look at Table 5.7 the first thing to observe is that approaches to landmark
identification predominantly use a static view on landmarks while landmark integra-
tion adopts a dynamic view. This makes sense since landmark identification needs
to find all geographic objects that may serve as a landmark in principle. These
approaches need to assess the salience of objects, i.e., check whether an object
sufficiently stands out from its local surroundings. As long as objects do not change,
this has to be done only once and can be run as a preprocessing step in an actual
system.
Landmark integration looks for landmarks that are actually useful in a given
context. The selected landmarks need to be visible, sensibly describe the given situ-
ation, and support conceptualizing what to do in the situation. These are functional
characteristics, as they depend on the specific situation, such as the current route
to follow. As discussed, landmark integration does not necessarily choose the most
salient landmark, but the most relevant one. Establishing the context, i.e., specifying
to a sufficient degree the parameters influencing the selection process, is a major
challenge here (see also the final discussion in Chap. 7 ) .
Consequently, landmark identification and landmark integration are often seen as
independent steps (e.g., [ 12 ] ). This view becomes apparent from Table 5.7 as well.
Each approach for landmark identification uses a concrete data source. However,
most approaches to landmark integration do not specify the data source used. These
approaches typically assume a set of landmark candidates to be given, i.e., some
kind of landmark identification has to have happened previously.
It can further be observed that all approaches except [ 9 ] use individuals rather
than types when assessing an object's suitability as landmark. And almost all
approaches follow Lynch's tradition of assuming landmarks to be point-like
objects [ 29 ] . While Duckham et al. [ 9 ] acknowledged that other geometries for
landmarks may be useful, only Richter [ 36 , 40 ] fully implemented an approach to
determine the role of landmarks with different geometries.