Computational Aspects: How Landmarks Can Be Observed, Stored, and Analysed - Landmarks: GIScience for Intelligent Services - page 163

5.5 Criticisms of Existing Approaches

As has been shown in this chapter's discussion so far, existing approaches to
landmark identification and landmark integration are not really integrated. You may
consider this a minor issue because it seems that an integration is a fairly straight-
forward engineering task. But such a lack of integration certainly has prevented
widespread application of the presented approaches beyond basic research. There
are other more serious challenges for the widespread use of landmarks in location-
based services, though. These will be further discussed in this section.

5.5.1 Data Challenge

Commercial systems use efficient algorithms to calculate shortest or fastest paths
based on metric distances and using references to street names, which are easily
extractable from a geo-referenced network representation of the street layout. In
Sect. 5.3 , we discussed some algorithms that account for landmarks in path search.
In order to make this efficient, these landmarks need to be embedded into the
existing network structure, i.e., graphs need to be annotated with objects that may
serve as landmark candidates. Some systems utilize points of interest, such as
gas stations or hotels. These are a potential source for landmark candidates (as
implemented in the WhereIs routing service), but mostly POIs serve as selectable
destinations or commercial announcements. Accordingly, there is a bias towards
specific categories of objects. Further, their distribution and density will vary
greatly—there are more POIs in a city center compared to a suburb or some rural
farm land. This uneven distribution has consequences for the quality of landmark-
based services.
Integrating landmarks into the network structure used for path search requires
suitable data structures. The Urban Knowledge Data Structure discussed in Chap. 4
is a candidate for such a structure. However, even with such a data structure at hand,
there is still the need to fill it with actual data. For most of the presented approaches,
this is a highly data intensive process. As most approaches are based on individuals,
individual objects have to be described in great detail, which is especially true
for the calculation of façade salience as discussed in Sect. 5.2 . The required
information is hard to collect automatically and, thus, labor intensive, which makes
this an expensive endeavor. It is therefore unlikely that such a collection will ever
materialize on a commercial scale and lead to a city-, nation-, or even world-wide
database of landmark candidates due to the immense collection efforts and costs.
This data challenge has been further discussed by Sadeghian and Katardzic [ 43 ] ,
and Richter [ 38 ] .