Conference PaperPDF Available

Structural Salience of Landmarks for Route Directions


Abstract and Figures

This paper complements landmark research with an approach to for- malize the structural salience of objects along routes. The aim is to automat- ically integrate salient objects—landmarks—into route directions. To this end, two directions of research are combined: the formalization of salience of objects and the conceptualization of wayfinding actions. We approach structural salience with some taxonomic considerations of point-like objects with respect to their positions along a route and detail the effects of different positions on the con- ceptualization process. The results are used to extend a formal language of route knowledge, the wayfinding choreme theory. This research contributes to a cogni- tive foundation for next generation navigation support and to the aim of formal- izing geosemantics.
Content may be subject to copyright.
Structural Salience of Landmarks for Route Directions
Alexander Klippel1and Stephan Winter2
1Cooperative Research Centre for Spatial Information
Department of Geomatics, The University of Melbourne, Australia
2Department of Geomatics, The University of Melbourne, Australia
Abstract. This paper complements landmark research with an approach to for-
malize the structural salience of objects along routes. The aim is to automat-
ically integrate salient objects—landmarks—into route directions. To this end,
two directions of research are combined: the formalization of salience of objects
and the conceptualization of wayfinding actions. We approach structural salience
with some taxonomic considerations of point-like objects with respect to their
positions along a route and detail the effects of different positions on the con-
ceptualization process. The results are used to extend a formal language of route
knowledge, the wayfinding choreme theory. This research contributes to a cogni-
tive foundation for next generation navigation support and to the aim of formal-
izing geosemantics.
1 Introduction
This paper investigates the structural salience of objects along routes. The structural
qualities considered are induced by embedding the route into a street network. Objects
are called structurally salient if their location is cognitively or linguistically easy to
conceptualize in route directions.
A generic and formal method of assessing the structural salience of objects with
the goal of finding a landmark selection process for route directions is proposed. Such
a measure of structural salience complements a formal model of salience recently de-
veloped (see, e.g., [1,2]), building on an earlier characterization of the nature of land-
marks [3]. This model assumes visual, semantic and structural qualities of objects that
contribute to their salience. Measures for structural qualities have been left out of this
model so far, a gap which will be filled by this paper.
The proposed method in this paper utilizes a conceptual approach to spatial infor-
mation exemplified by route direction elements (e.g., [4,5]), and extends the wayfind-
ing choreme approach, i.e., a formal language for the specification of conceptual route
knowledge, for the inclusion of salient features. This paper classifies structural aspects
of landmarks, especially their location with respect to the re-orientation actions per-
formed at the nodes of the underlying street network, which are frequently called deci-
sion points.
The linguistic complexity of characterizing the relative location of a landmark is dis-
cussed in relation to the conceptual complexity of realizing its wayfinding affordance.
Klippel, A., & Winter, S. (2005). Structural salience of landmarks for route directions. In A. 
G. Cohn & D. M. Mark (Eds.), Spatial Information Theory. International Conference, 
COSIT 2005, Ellicottville, NY, USA, September 14-18, 2005, Proceedings (pp. 347-362). 
LNCS 3693. Berlin: Springer. (pls quote original)
The hypothesis in this paper is that the conceptual complexity influences the selection
of a landmark by direction givers, and hence, the salience of this landmark.
The described approach of the paper brings together two lines of research that
have been unrelated so far: the formalization of salience and the conceptualization of
wayfinding actions. The paper is organized as follows: We start by reviewing work on
landmarks and conceptualization processes relevant for interactions with spatial envi-
ronments. For the conceptualization of landmarks as route elements we start by in-
stantiating a taxonomy that classifies the location of landmarks with respect to routes.
Building on this general classification scheme, we detail the different possibilities of
locating landmarks at decision points. We investigate the conceptual complexity of us-
ing different types of landmarks, and derive a measure of structural salience from their
ease of use. This measure can be combined with the existing measures for visual and
semantic salience [1]. We briefly discuss the extension of the wayfinding choreme the-
ory and the formalization of conceptual route knowledge by relying on the developed
taxonomy, and conclude with a discussion.
2 Landmark research and conceptual approaches to spatial
Since the work by Lynch [6] on elements that structure our urban environmental knowl-
edge, the concept of landmarks has inspired multiple research papers. There are some
simple and straightforward facts that can be manifested from this research:
Everything that stands out from a scene can be a landmark [7].
In certain contexts, for example route following, even road intersections can be
Landmarks are pertinent for finding one’s way.
Landmarks are remembered/learned early on (i.e., landmark knowledge) [8].
Landmarks structure environmental knowledge, for example, as anchor points [9].
Landmarks are used to communicate route knowledge verbally and graphically
Landmarks are integrated in route directions to varying degrees, with greater quan-
tities at origins, destinations and distinguished decision points [11,12].
Landmarks at street intersections (decision points) are more pertinent when a change
in direction is required [13].
Landmarks generally work better than street signs in wayfinding (e.g., [14]).
While we understand how and why, people use landmarks in communication, and
hence in memory and mental processes, the technical basis to automatically construct
wayfinding directions with landmarks is still limited. Suitable formalisms for character-
izing conceptual route knowledge that can be flexibly adapted to canonical and personal
preferences are still missing.
Part of the problem is formalizing the concept of a landmark, such that a service
can identify objects of some landmarkness or salience, i.e., objects that differ from
their background [7]. According to a recent proposal [1,2, 15] the salience of objects
is determined by visual, semantic and structural qualities. These qualities can be char-
acterized to provide an overall measure of salience. The approach proposes measures
for visual qualities (such as the size, form, or texture of an object), and semantic quali-
ties (such as their prominent or labelled use); measures for structural qualities have not
been included so far. Objects that show large visual and semantic salience show good
compliance with cognitively salient objects, i.e., landmarks chosen by people for their
communication of routes.
The measure of salience can be adapted to context [15], and can be weighted by
advance visibility along a route [16], which makes it route specific. The approach has
also been adopted for data mining in topographic data sets [17,18].
An approach orthogonal to the work on the salience of objects is taken by wayfind-
ing choreme theory [19, 4,20]. This approach, however, can provide the missing aspect
of structural salience. It is based on wayfinding events such as re-orientations and turns
at street intersections as primitives; conceptual primitives of turns are derived from this.
The conceptualization of actions and events3and their formal specification is a re-
cently well discussed area of research [21, 22]. Complementary to other computational
formalisms [23] the wayfinding choreme theory stresses the cognitive aspects of route
knowledge by making (cognitive) conceptual primitives the basis of the formalism. Yet,
the focus on conceptualization and the development of a formal language [4] offers
many ties to recent discussions on the formalization of conceptual spaces [24], e.g., for
landmarks [25], and the general approach of integrating cognitive semantics (sometimes
referred to as conceptual semantics) into information systems [26]. The wayfinding
choreme approach seeks conceptual primitives as a foundation for a formal language of
space in which the number of basic elements are restricted and the combinatorial pos-
sibilities are constraint by the represented knowledge, in our case, the linear character
of a route. Additionally, the focus on conceptual aspects of information creates a basis
for multimodality in that the externalization of conceptualizations can be specified in
various output formats (e.g., [12,27]): verbal, graphical or gesture.
The conceptualization of an action at a decision point, however, is not only depen-
dent on the angle of the turn, i.e., the geometric representation of the trajectory of the
movement, but also on the street structure in which the action is embedded [28]. Addi-
tionally, the possibility of relating the turning action to supplementary information—for
example a landmark—has an influence on the conceptualization.
A formal specification of conceptual primitives of landmark locations should there-
fore allow the characterization of different layers of interaction with the environment
and grasp the resulting conceptual primitive adequately. In this paper, we will further
elaborate the conceptual approach and focus on landmarks, specifically their structural
salience induced by the conceptualization of a wayfinding action. Based on this charac-
terization we will extend the rules specified for the higher order route direction elements
(HORDE) [29,30] to allow for different levels of granularity in route directions.
3In this article we do not distinguish between actions and events.
3 Conceptualizing landmarks as route elements
Some of the observed characteristics of landmarks discussed in Section 2 concern their
structural qualities with respect to a given route: the route structure (co-)determines
which salient objects are selected to give route directions. Hence, this paper develops a
classification schema for point-like landmarks depending on their location relative to a
route and the route’s structural embedding in the street network. For some ideas on other
types of landmarks and their relation to route directions see [20]. We further show that
some locations are conceptually easier and conceptually less ambiguous than others,
especially with regard to building complex route elements, i.e., combining conceptual
primitives into HORDE.
3.1 Landmark locations
The following taxonomy of landmarks induced by a route embedded into a street net-
work is illustrated in Figure 1. Landmarks can occur (i) distant from the route (distant
landmarks), (ii) somewhere along route segments (segment landmarks), or (iii) at spe-
cific route nodes (node landmarks). Segment landmarks and node landmarks can be
grouped as either close to or on-route. Route nodes are also called decision points.
Fig.1. A structural taxonomy of landmarks; the abbreviations are detailed in the text.
Distant and on-route landmarks. With respect to a route, these categories of land-
marks have different degreesof freedom regarding their location. A distant landmark is
not determined in its exact location in two dimensions. The conceptualization details an
area, a region in which the landmark is placed, or a line of sight. A segment landmark,
on the other hand, has its location determined only within the one-dimensional interval
between two route nodes; their exact location along the segment—the linear reference
from start or end node—remains under-specified. In contrast, node landmarks are con-
strained in their location by a node of the street network or decision point. With respect
to their function in route directions and the conceptualization of the action that has to
be performed, both their location and their remaining degrees of freedom in location
have to be reflected, for example, in the type of verbal reference.
Distant landmarks fulfill a variety of functions, for example, global orientation, re-
assurance and confirmation [31]. Their actual location (or distance) is irrelevant as long
as the direction or visibility can be taken as reference. Distance generally is not a crite-
rion for exclusion from route directions. Consider for example the route direction:
Follow the street until you see the castle distantly on your right. (1)
Yet, the effect of distant landmarks on the conceptualization of route parts, for ex-
ample, spatial chunking [32], is rather complex. Many parameters that influence the
conceptualization of these distant landmarks are not primarily spatial (at least not pla-
nar spatial). They are therefore not the focus of this paper.
Within the category of landmarks that are close to or on-route (cf. [33]) segment
landmarks [34] are used for reassurance and confirmation. The influence of segment
landmarks on chunking is discussed in Section 3.3. In contrast, node landmarks may be
used as an anchor for action (re-orientation and turning), and hence, their location with
respect to the route is relevant. But node landmarks can also occur at decision points
where no re-orientation is necessary. Within our taxonomy we write segment landmarks
as lm
, and node landmarks as lm
Route directions cannot neglect any necessary re-orientation, but they can neglect
confirmations that may occur either between decision points or at decision points where
no re-orientation is required. That means, node landmarks at decision points with re-
orientation are essential, and other landmarks are less important or optional. This char-
acterization establishes a first indication of structural salience.
Node landmarks. A further common distinction is made between decision points
with direction change, dp
, and decision points without direction change, dp
. This
distinction has to be accounted for in the taxonomy of node landmarks. It has been
shown that landmarks at dp
are more pertinent to route knowledge [11,13]. Within
our taxonomy we coin them lm
, and landmarks at dp
are coined lm
, such that
, lm
At a more detailed level of spatial granularity, it is of interest where with relation
to any decision point, dp
or dp
, a landmark is placed. Not every node landmark
is equally suited to aid wayfinding and to be integrated into route directions and the
conceptualization of route parts, respectively.
Note that this characterization is based on locational properties, i.e., the location of
a landmark with respect to the physical layout of an intersection. It is not a character-
ization based on the visual or semantic salience of a landmark. This characterization
is also a specification of the locational properties from the perspective of mental con-
ceptualization processes, i.e., the conceptualization of an action performed in a spatial
environment. In experimental settings (e.g., in [32]) it was made sure that at dp
marily those landmarks that can be integrated into a route direction conceptually easily
are used. This integration is afforded by the landmarks’ location with respect to the
action at the decision point. More specifically, landmarks at dp
were chosen that are
passed immediately before a turning decision. These node landmarks may be located
on the left or on the right side of the route, independent from the direction of the turn.
A natural language example would be:
Turn right after the post office. (2)
Based on the idea that the action performed (or imagined) at a decision point is the
pertinent factor for the conceptualization process, we introduce further sub-concepts for
landmarks, namely landmarks passed before the action is performed, lmb, landmarks
not directly passed, lmn, and landmarks passed after decision, lma(see Figure 1, and
also Figure 2 for more details). These concepts can be specified for decision points with
a direction change, but also for decision points without a direction change.
Fig.2. Possible locations of node landmarks with re-orientation. The different locations result in
different conceptualizations, and not every location of a landmark functions equally well as an
identifier for the required decision.
At dp
, landmarks passed before decision, lmb
, work equally well for all turning
concepts. That is, they are straightforward to conceptualize as the turning occurs im-
mediately after them and does not conflict with the overall branching structure of the
decision point. Compare the use of a lmb
Make a sharp right turn after the post office. (3)
with a lmn
or lma
Make a sharp right turn at the intersection where the post office is. (4)
which represents here a more precise, but also more complicated direction ‘make a
sharp right turn at the intersection where the post office is at the opposite corner’. Es-
pecially at more complex intersections, where it is difficult to conceptualize the location
of a landmark, a lmb
is the only unambiguously identifiable one.
3.2 A route direction grammar with node landmarks
Having a taxonomy for the location of landmarks with respect to a route, we can in-
tegrate them into the wayfinding choreme route grammar [4]. Generally, two turning
concepts have been differentiated, standard turning concepts hSTCiand modified turn-
ing concepts hMTCi[4]. Both can be extended to incorporate the different landmark
locations. To this end, the node landmark and its location is added to the wayfinding
choreme grammar as an annotation. We exemplarily detail the notation for the wayfind-
ing choreme of a right turn, wcr, added with a landmark passed before the decision,
Likewise for other turning concepts.
3.3 Spatial chunking with node landmarks
This section exemplifies the influence of structural aspects of landmark positions within
yet another aspect of route directions, the change in granularity by applying higher
order route direction elements (HORDE). We discuss here the possibilities of chunking
with node landmarks at decision points without a direction change, lm
preceding a
have two functions: First, they are used to identify a decision point resulting
in a verbalization such as ‘go straight at the intersection where the McDonald’s is’.
Second, they are used in a way analogous to a segment landmark lm
, such as ‘pass
the McDonald’s and turn right at the Shell gas station’. Here it is not specified whether
the lm
is placed at a decision point or between two decision points. The second case
might be an example of spatial chunking.
Two distinctions are pertinent for lm
that determine their function in spatial chunk-
ing. First, whether the landmark is passed before (lmb
) or after (lma
) the action of
straight crossing the intersection. Second, whether a landmark is present at the chunk
terminating dp
ahead. We observe that lm
only appear within small chunks (similar
to segment landmarks lm
) if at least one and possibly both of the following condi-
tions are met: (i) the lm
is of the type lma
; (ii) there is an easy to conceptualize node
landmark at the chunk terminating dp
. Figure 3 illustrates these assumptions. The fol-
lowing cases demonstrate some rules that can be distinguished and integrated into the
wayfinding choreme route grammar:
Consider the first example in Figure 3. lm
is passed before orientation: lmb
and at the chunk terminating dp
there is a landmark. The resulting concept is:
concept is over-specified when only two decision points are present; the first land-
mark should be left out, even if it is the somewhat more salient one.
Consider the second example in Figure 3. lm
is passed before re-orientation: lmb
but no landmark is present at the chunk terminating dp
. This situation has to be
IS TURN RIGHT AT THE NEXT INTERSECTION, or alternatively, in a more complex
Now turn to the third example in Figure 3. lm
is passed after re-orientation: lma
and a landmark is present at the corresponding dp
, for example, a lmb
. The re-
sulting concept is similar to the first case: PASS FIRST LANDMARKAND TURN
RIGHT AT SECOND LANDMARK’. When only two decision points are involved,
the first landmark is left out.
Finally, consider the fourth example in Figure 3. lm
is passed after re-orientation:
, but no landmark is present at the corresponding dp
. Here, lm
is used
similarly to a lm
Fig.3. Landmarks at a decision point without a direction change, and their influence on spatial
chunking in examples 1-4.
So far, distance has not been considered. There are, however,similarities to another
approach to integrate distance in term rewriting as a method to extract conceptually
connected primitives [35]. A more detailed characterization of spatial situations to dif-
ferentiate, for example, the following two concepts is ongoing research: (a) MAKE A
4 Structural salience of landmarks
The previous sections discussed the conceptual approach to characterize the positions
of landmarks with respect to their relevance for route directions. This section extends
the salience model for landmarks to include structural aspects derived from the findings
4.1 The salience model
The salience model [1, 2] provides a measure of salience for all identified objects within
a street network. These measures enable choosing the most salient objects along a spe-
cific route, for example, at decision points with direction change, to enrich route direc-
tions. For static objects the measures can be calculated once, and stored as parameters
of the objects.
The original model of salience was based on three qualities: visual, semantic and
structural [3]. Each quality can be characterized by a normed measure of salience
(with values 0. . . 1), resulting in visual salience sv, semantic salience ss, and structural
salience su. These measures can be combined to a weighted average of joint salience,
so=wvsv+wsss+wusuwith wv+ws+wu= 1 (6)
So far, structural salience has been considered in the model, but it was not developed.
Saliences svand ssare determined by comparing visual and semantic properties of
objects with the properties of other objects in their neighborhood. The more distinct a
property of an object is, in its neighborhood, the higher the object’s salience measure
is. By this way, the figure-ground relation mentioned in Section 2 is quantified. This
Visual and semantic salience is dependent on the properties of objects nearby; it is
a relative property of an object, not an absolute one. For instance, a red facade in
an area where all facades are red will not stand out. But the same facade in a grey
neighborhood stands out.
Joint salience is quantitative, i.e., it can be represented by a real number between 0
and 1. It is not qualitative (e.g., ‘landmark’, ‘no landmark’), as supposed so far in
the previous sections.
It was shown further that weighting individual visual and semantic criteria differently
allows for adapting the salience measure to different wayfinding contexts [15].
This original model of salience was investigated for one class of objects: facades
of buildings. Consider for example Figure 4, which shows eight street facadesof build-
ings at a street intersection, and their visual and semantic salience represented by grey
intensity. According to the original salience model we would choose the object of high-
est salience (in Figure 4, one of the two facades of building c) as a landmark for route
Fig.4. A street intersection with eight street facades; each individual facade has a salience indi-
cated by grey intensity.
4.2 Advance visibility
Visual and semantic salience characterize properties of objects that are independent
from routes and the street network. Route dependent properties of objects can be con-
sidered additionally. Each object can be related to an infinite number of routes, but lo-
cally the number of combinations, i.e., the number of approaching directions, is small.
This means local route-dependent properties can be represented by a small number of
fixed parameters. This number is rarely larger than four; extreme cases, such as the
Arc d’Triomphe in Paris, extend already the concept of an intersection to a large circle
which can be considered as comprising of several intersections.
The salience model can be extended by advance visibility [16]. This measure char-
acterizes the visibility of the object from an approaching direction, and hence, is differ-
ent for each approaching direction. The rationale for this additional component is that
the most salient object at a decision point can form a poor reference in a route direction
if it is not visible in advance but an alternative salient object is. Thus advance visibil-
ity sahas to be balanced with joint salience to characterize total salience st, e.g., by
multiplying the two measures:
Note that sais also a normed value between 0 and 1. Multiplication favors objects that
are at the same time at least to some extent jointly salient and to some extent visible
in advance, compared to salient but not visible, or visible but not salient objects. This
behavior seems to be reasonable.
Consider for example Figure 4 again. The four facades facing towards the street
the wayfinder is approaching are all visible to the wayfinder, but from the four facades
facing the cross-road two are only partially visible (4(a) and (b)), and two are not at all
visible (4(c) and (d)). Hence, the total salience stis largest for the facade of the building
cthat faces towards the street the wayfinder approaches. This facade should be used for
route directions if we consider visual and semantic salience and advance visibility.
At this stage, we have a model that ranks objects by salience and advance visibility,
but remains indifferent to the structural characteristics of the relation between objects
and street network, or to the relation between objects and routes. However, in Section 3
we saw that the relationships between landmark, route and street network influence the
selection process of landmarks. The integration of the structural properties of objects in
the salience model still needs to be done. This means we have to develop
normed salience measures for the identified structural properties of objects (su);
an adaptation of higher order route direction elements (HORDE) for quantitative
measures of landmarkness.
4.3 Structural salience
The discussion of structural properties of landmarks in Section 3 showed that
1. structural properties of objects co-determine their suitability as a landmark;
2. structural properties are, if not quantitative, at least ordered, such that a specific
weight of at least an ordered scale can be attached to each situation;
3. structural properties are determined by the structure of the underlying street net-
work, and locally route dependent, which means they are countable and constant.
The set of weights should reflect the hierarchy of Figure 1, and the distinctions of Fig-
ures 2 and 3. The order reflected in these figures is motivated by the previous dis-
cussions, and partially validated by cognitive, behavioral or linguistic experiments.
Presently, we convert the ordered scale measures of structural salience into ratio scale
by matching an equally partitioned interval from 0 to 1. However, determining more re-
alistic ratio scale weights needs careful human subject testing and is beyond the scope
of this paper.
The third aspect—countable and constant measures—means that the measures can
be stored as properties with each object. They are route- and street network dependent,
and hence their combinatorial complexity is higher than, for example, for advance vis-
ibility. For node landmarks, for example, the complexity is n(n1), with nbeing the
node degree of the street intersection, because the structure requires consideration of
not only the incoming direction, but also the one going off. Note that this includes the
distinction between node landmarks with re-orientation, lm
, and landmarks without,
. This means with street intersection degrees of rarely larger than four the com-
plexity is rarely larger than twelve.
The measure of structural salience can be integrated into the original model of
salience (Eq. 6). It is still one of three components that add up, i.e., an object is salient
if it is visually, semantically or structurally distinct.
If we survey people for measures of structural salience with the one-dimensional
configurational relationships, the results might be mixed up with expectations of ad-
vance visibility. However, in the motivation for structural distinctions we only argued
for cognitive and linguistic simplicity. Hence, advance visibility is different and remains
a component of total salience (Eq. 7).
For an illustration, consider the situation in Figure 5. The situation shows a route at a
decision point with direction change, and some facades with their measures of salience.
According to the discussion in Section 3, landmarks at decision points are more perti-
nent than those along route segments, and at decision points distinctions can again be
made in relation to the action (here: turning right). The structural salience measures for
the given facades reflect this hierarchy. Note that the structural salience measures recur
for buildings (as point-like landmarks), not for facades individually. Advance visibility
is assumed to be equal for all facades facing the street the wayfinder approaches to the
decision point, less for the facade on the cross-road facing the advancing wayfinder,
and zero (not visible in advance) for the facade on the cross-road facing away from the
advancing wayfinder.
Fig.5. A route at a decision point with direction change, and measures of salience for some
facades (svs: visual and semantic salience, su: structural salience, sa: advance visibility).
With Equation 7 we derive the (route- and network dependent) measures of total
salience for the considered facades shown in Figure 6. In the given spatial configuration,
and for the given visual and semantic salience, the facade with st= 0.72 is the most
salient one. This is particularly interesting as it is not the most visually or semantically
salient one. Hence taking into consideration the route- and street network dependent
aspects can change the priorities significantly, a behavior that was sought for.
4.4 Selection process in HORDE
The original salience model did provide a comparison between objects, but did not look
into selection. It still assumed a superordinate selection process that exploits salience
measures to select references to salient objects where needed. In contrast, structural
salience prioritizes visually and semantically salient objects at specific locations along
a route. It establishes a selection process by weighting objects between decision points
against objects at decision points and so on.
Compared to the discussion in Section 3.3 the situation at this stage has changed.
Objects along the route now have more or less salience, and are no longer categorically
considered as ‘landmarks’. The measures of salience along a route form a distribution,
Fig.6. The total salience stfor the facades.
which can further support selection. Let us study the distribution of values with two
Imagine a route through a green suburb of one-family houses. Salience measures
of the objects (facades) along the route differ slightly, but no object stands sig-
nificantly out. The distribution of salience measures has a small variance and no
outliers. Selecting the most salient object along a route segment is possible, but not
really helpful.
Imagine a route along Vienna’s Ringstrasse. There are frequent salient objects (the
parliament, the Burgtheater, the city hall, the university, the stock exchange, and
so on), and the measures of salience vary largely. The wayfinder is attracted, and if
the route description only indicates to ‘walk straight to the Danube’ and does not
mention the attractions, she might feel uncomfortable and wonder if she is on the
right track4.
In other words, in an environment with one or a few outstanding object(s) these objects
can be used as ‘landmarks’ in the categorical sense of Section 3.3. In an environment
with no outstanding objects it is better to refer to other (structural) properties, such as
the number of intersections. The appropriate method to distinguish these two cases is
outlier detection, i.e., basing the decision on the standard deviation of salience in that
environment, not on an absolute threshold value.
An object with a large salience has probably, but not necessarily, structural salience
as well. This means that objects with large salience measures have a high probability of
being at decision points dp
, or in another salient structural relationships to the route.
5 Conclusions and outlook
In this paper we have combined two approaches on formalizing route knowledge rel-
evant to the selection of landmarks and for integrating them into route directions: on
4An option currently investigated relies on recursion to higher levels of abstraction, such as
‘walk straight along the attractive Ringstrasse to the Danube’.
the one hand the salience of landmarks as dominant objects in route knowledge and
route directions, and on the other hand the conceptualization of wayfinding actions in
relation to landmarks, i.e., the integration of landmarks in the formal specification of a
conceptual route language, the wayfinding choreme theory.
Both approaches on their own are well established and the combination of them re-
sults in efficient formalisms addressing several unsolved research questions. Combined,
they allow for the specification of structural salience and will complement the basis for
an automatic, cognitive adequate generation of route directions in wayfinding assistance
The approach of defining the structural salience of landmarks through the applica-
tion of a conceptual approach also offers answers to research questions in the area of
geosemantics; especially, their formalization, standardization and automatization, for
example, for mobile navigation systems. The application of conceptual (cognitive) se-
mantics for geographic information science has recently gained attention through re-
search on ontologies [26].
Other approaches that aim to formally characterize spatial structures have to be
considered in greater detail. Especially the approach of space syntax provides several
concepts that relate to the topics discussed in this paper [36,37].
With the precisiation of location of landmarks at intersection the next step in the
formalization of conceptual knowledge, especially with respect to different modi of ex-
ternalization, has been achieved. The dual approach of a generic concept that in general
specifies the presence of a landmark and the possibility of a more detailed analysis of-
fers a means to model different levels of granularity in route directions. It also offers
a means to contribute the structural salience to models finding salient features by data
mining in text documents [38,39].
6 Acknowledgements
This work has been supported by the Cooperative Research Centre for Spatial Infor-
mation, whose activities are funded by the Australian Commonwealth’s Cooperative
Research Centres Programme. We would like to thank Kai-Florian Richter, Bremen,
who saw an earlier version, and the anonymous reviewers for invaluable comments.
1. Raubal, M., Winter, S.: Enriching wayfinding instructions with local landmarks. In Egen-
hofer, M.J., Mark, D.M., eds.: Geographic Information Science. Volume 2478 of Lecture
Notes in Computer Science. Springer, Berlin (2002) 243–259
2. Nothegger, C., Winter, S., Raubal, M.: Selection of salient features for route directions.
Spatial Cognition and Computation 4(2004)
3. Sorrows, M.E., Hirtle, S.C.: The nature of landmarks for real and electronic spaces. In
Freksa, C., Mark, D.M., eds.: Spatial Information Theory. Volume 1661 of Lecture Notes in
Computer Science. Springer, Berlin (1999) 37–50
4. Klippel, A., Tappe, H., Kulik, L., Lee, P.U.: Wayfinding choremes A language for modeling
conceptual route knowledge. Journal of Visual Languages and Computing (in press)
5. Klippel, A., Dewey, C., Knauff, M., Richter, K.F., Montello, D.R., Freksa, C., Loeliger, E.A.:
Direction concepts in wayfinding assistance systems. In Baus, J., Kray, C., Porzel, R., eds.:
Workshop on Artificial Intelligence in Mobile Systems 2004 (AIMS 2004). Volume Memo
Nr. 84., Nottingham, UK, Sonderforschungsbereich 378 (2004) 1–8
6. Lynch, K.: The Image of the City. MIT Press, Cambridge (1960)
7. Presson, C.C., Montello, D.R.: Points of reference in spatial cognition: Stalking the elusive
landmark. British Journal of Developmental Psychology 6(1988) 378–381
8. Siegel, A.W., White, S.H.: The development of spatial representations of large-scale envi-
ronments. In Reese, H.W., ed.: Advances in child development and behavior. Volume 10.
Academic Press, New York (1975) 9–55
9. Couclelis, H., Golledge, R.G., Tobler, W.: Exploring the anchorpoint hypothesis of spatial
cognition. Journal of Environmental Psychology 7(1987) 99–122
10. Denis, M.: The description of routes: A cognitive approach to the production of spatial
discourse. Current Psychology of Cognition 16 (1997) 409–458
11. Michon, P.E., Denis, M.: When and why are visual landmarks used in giving directions? In
Montello, D.R., ed.: Spatial Information Theory. Volume 2205 of Lecture Notes in Computer
Science. Springer, Berlin (2001) 292–305
12. Tversky, B., Lee, P.: Pictorial and verbal tools for conveying routes. In Freksa, C., Mark,
D., eds.: Spatial Information Theory. Volume 1661 of Lecture Notes in Computer Science.
Springer-Verlag, Berlin (1999) 51–64
13. Lee, P., Klippel, A., Tappe, T.: The effect of motion in graphical user interfaces. In Butz, A.,
uger, A., Olivier, P., eds.: Smart Graphics. Springer, Berlin (2003) 12–21
14. Tom, A., Denis, M.: Referring to landmark or street information in route directions: What
difference does it make? In Kuhn, W., Worboys, M., Timpf, S., eds.: Spatial information
theory. Volume 2825 of Lecture Notes in Computer Science. Springer, Berlin (2003) 384–
15. Winter, S., Raubal, M., Nothegger, C.: Focalizing measures of salience for wayfinding. In
Meng, L., Zipf, A., Reichenbacher, T., eds.: Map-based Mobile Services Theories, Methods
and Implementations. Springer Geosciences, Berlin (2005) 127–142
16. Winter, S.: Route adaptive selection of salient features. In Kuhn, W., Worboys, M.F., Timpf,
S., eds.: Spatial Information Theory. Volume 2825 of Lecture Notes in Computer Science.
Springer, Berlin (2003) 320–334
17. Elias, B.: Extracting landmarks with data mining methods. In Kuhn, W., Worboys, M.F.,
Timpf, S., eds.: Spatial Information Theory. Volume 2825 of Lecture Notes in Computer
Science. Springer, Berlin (2003) 398–412
18. Elias, B., Brenner, C.: Automatic generation and application of landmarks in navigation data
sets. In Fisher, P., ed.: Developments in Spatial Data Handling. Springer, Berlin (2004)
19. Klippel, A., Lee, P.U., Fabrikant, S.I., Montello, D.R., Bateman, J.: The cognitive conceptual
approach as a leitmotif for map design. In Barkowsky, T., Freksa, C., Hegarty, M., Lowe, R.,
eds.: Reasoning with Mental and External Diagrams: Computational Modeling and Spatial
Assistance. Volume TR SS-05-06., Stanford, CA, AAAI Press (2005) 90–94
20. Richter, K.F., Klippel, A.: A model for context-specific route directions. In Freksa, C., Nebel,
B., Knauff, M., Krieg-Br¨
uckner, B., eds.: Spatial Cognition IV. Volume 3343 of Lecture
Notes in Artificial Intelligence. Springer, Berlin (2005) 58–78
21. Casati, R., Varzi, A., eds.: Events. Dartmouth, Aldershot (1996)
22. Zacks, J., Tversky, B.: Event structure in perception and conception. Psychological Bulletin
127 (2001) 3–21
23. Worboys, M.: Event-oriented approaches to geographic information. International Journal
of Geographical Information Science 19 (2005) 1–28
24. G¨
ardenfors, P.: Conceptual Spaces. The MIT Press, Cambridge, MA (2000)
25. Raubal, M.: Formalizing conceptual spaces. In Varzi, A.C., Vieu, L., eds.: Formal Ontology
in Information Systems. Volume 114 of Frontiers in Artificial Intelligence and Applications.,
Amsterdam, NL, IOS Press (2004) 153–164
26. Kuhn, W.: Why information science needs cognitive semantics. In: Paper presented at
Workshop on the Potential of Cognitive Semantics for Ontologies (FOIS 2004), Torino, Italy, (2004)
27. Sowa, J.F.: Knowledge Representation: Logical, Philosophical, and Computational Founda-
tions. Brooks Cole, Pacific Grove (2000)
28. Klippel, A., Montello, D.R.: On the robustness of mental conceptualizations or the scrutiny
of direction concepts. In Egenhofer, M., Miller, H.J., Freksa, C., eds.: GIScience 2004,
University of Maryland, MD, Regents of the University of California (2004) 139–141
29. Klippel, A.: Wayfinding choremes. In Kuhn, W., Worboys, M.F., Timpf, S., eds.: Spatial
Information Theory. Volume 2825 of Lecture Notes in Computer Science. Springer, Berlin
(2003) 320–334
30. Dale, R., Geldof, S., Prost, J.P.: Using natural language generation in automatic route de-
scription. Journal of Research and Practice in Information Technology 37 (2005) 89–105
31. Couclelis, H.: Verbal directions for way-finding: Space, cognition, and language. In Portu-
gali, J., ed.: The Construction of Cognitive Maps. Kluwer, Dordrecht (1996) 133–153
32. Klippel, A., Tappe, H., Habel, C.: Pictorial representations of routes: Chunking route seg-
ments during comprehension. In Freksa, C., Brauer, W., Habel, C., Wender, K.F., eds.: Spatial
Cognition III. Volume 2685. Springer, Berlin (2003) 11–33
33. Lovelace, K.L., Hegarty, M., Montello, D.R.: Elements of good route directions in familiar
and unfamiliar environments. In Freksa, C., Mark, D.M., eds.: Spatial Information Theory.
Volume 1661 of Lecture Notes in Computer Science. Springer, Berlin (1999) 65–82
34. Herrmann, T., Schweizer, K., Janzen, G., Katz, S.: Routen- und ¨
konzeptuelle ¨
Uberlegungen. Kognitionswissenschaft 7(1998) 145–159
35. Kulik, L., Egenhofer, M.: Linearized terrain: Languages for silhouette representation. In
Kuhn, W., Worboys, M.F., Timpf, S., eds.: Spatial Information Theory. Volume 2825 of Lec-
ture Notes in Computer Science. Springer, Berlin (2003) 128–145
36. Davis, L.S., Benedikt, M.L.: Computational models of space: Isovists and isovist fields.
Computer Graphics and Image Processing 11 (1979) 49–72
37. Batty, M.: Exploring isovist fields: space and shape in architectural and urban morphology.
Environment and Planning B 28 (2001) 123–150
38. Tezuka, T., Yokota, Y., Iwaihara, M., Tanaka, K.: Extraction of cognitively-significant place
names and regions from web-based physical proximity co-occurrences. In Zhou, X., Su,
S., Papazoglou, M.P., Orlowska, M.E., Jeffrey, K.G., eds.: Web Information Systems (WISE
2004). Volume 3306 of Lecture Notes in Computer Science. Springer, Berlin (2004) 113–124
39. Winter, S., Tomko, M.: Translating the web semantics of georeferences. In Taniar, D.,
Rahayu, W., eds.: Web Semantics and Ontology. Idea Publishing, Hershey, PA (accepted for
... Bernardini and Peeples (2015) highlight that visually unique landmarks are useful because they are readily distinguished from similar landmarks and background visual noise, for example, a building can be visually distinctive being the only red building in a street of grey ones (Claramunt & Winter, 2007). When considering the structural salience properties, comes from the location or the position along a route (Klippel & Winter, 2005). For example, an urban element can be structurally distinctive because it is located at an intersection and further determined according to the number of streets intersecting at that intersection, the number of streets, a street is connected with or the function of the street for the connectivity within the street network (Claramunt & Winter, 2007). ...
... Contemporary discourse on the visual experience of the city identified that the role of landmarks in wayfinding is an important topic and several scholars investigate the visibility, legibility and visual salience of landmarks (Klippel & Winter, 2005;Omer & Goldblatt, 2007;Kalin & Yilmaz, 2012;Silavi et al., 2017). Some of the methodologies that are used for the visual analysis are fractal dimension (Hagerhall et al. 2004); View shed approach (Sander & Manson, 2007); streetscape analysis (Meetiyagoda & Munasinghe, 2016); serial vision analysis (Kalin & Yilmaz, 2012) and so on. ...
Full-text available
The concepts of imageability and legibility are important aspects of urban design. Many scholars use the terms “imageability” and “legibility” interchangeably, usually examining one concept and applying the implications to the other. This research explores the relationship between these two concepts by answering the research questions: 1. how do people perceive the saliency of landmarks (imageability) and 2. how does the spatial configuration facilitate the visibility level of landmarks (legibility)? The Galle Heritage City in Sri Lanka is considered as the case study. The first part of the empirical study is to assess the level of imageability of urban space users by completing 100 cognitive maps and producing a composite cognitive map that indicates the structural landmarks’ salience or the level of imageability. The second part is the level of legibility of the landmarks by employing the visibility assessment process and the third part compares the two results with a concurrence matrix. The findings highlight that there is a positive relationship between people’s perception (imageability) and level of visibility (legibility). Further, imageability mostly depends on semantic properties than legibility, but legibility predominantly depends on structural properties and visual properties are almost equally important to both concepts.
... Rousell and Zipf (2017) devised algorithms for the extraction, weighing, and selection of landmarks based on their suitability for the generation of landmark-based navigation instructions for pedestrian routes. Raubal and Winter (2002) and Klippel and Winter (2005) proposed measures to formally specify the salience of landmarks for route instructions. ...
... Our results also provide valuable information for the development of mobile pedestrian navigation systems. In Geographical Information Science (GIS), Raubal and Winter (2002) and Klippel and Winter (2005) developed a mathematical model for automatic landmark selections that is a linear combination of visual salience (s v ) , semantic salience (s s ), and spatial (structural) salience (s u ) . Each of these parameters can be weighted by a corresponding weighting factor w v , w s , w u , which leads to the equation s o = w v s v + w s s s + w u s u for the overall salience (s o ) of a landmark in the surrounding. ...
Full-text available
Cognitive studies showed that good landmarks–salient objects in the environment–make it easier for recipients of route instructions to find their way to the destination. Adding landmarks to route instructions also improves mobile navigation systems for pedestrians. But, which landmarks do people consider most helpful when giving route instructions? Four experiments explored this question. In the first experiment, the environment, including the route and landmarks, was presented on a map. The landmarks were located at the four corners of a right-angled intersection. Participants had to select those landmark-based route instructions they considered most helpful. In all other experiments, the environment was presented from an egocentric perspective, either in a video or as a sequence of pictures of intersections. Participants had to select those landmarks they would use in a route instruction. All landmarks had the same visual and semantic salience. The positions of the participants at the intersection were varied. Results show that participants consistently selected landmarks at the side of the road into which they had to turn. Moreover, the participants' position at the intersection affected whether they selected landmarks before or behind the decision point. These results have consequences for human spatial cognition research and for the automatic selection of landmarks in mobile pedestrian navigation systems.
... Therefore, it is reasonable to assume that different passing routes and directions at this road intersection will lead to different complexity of turning decisions. This phenomenon is in line with other empirical studies (Bongiorno et al., 2021;Klippel & Winter, 2005;Perebner, Huang, & Gartner, 2019;Richter, 2009), which show that different walking routes (i.e. different passing routes and directions of crossing intersections) correspond to varying complexity values. ...
The complexity of urban physical environments at road intersections is a primary factor characterizing the difficulty of wayfinding, which is a fundamental spatial activity of human beings in cities. A complex intersection may increase the difficulty of understanding the environment, which may result in incorrect turning decisions and even bring road safety issues. Existing methods measure the complexity of road intersections by solely considering their visual or structural features. More importantly, they only output a single complexity value for each intersection, failing to differentiate the decision-making complexity based on the specific entry/exit branches of a passing branches. This study proposes a computational model to quantify the fine-grained decision-making complexity of road intersections for the navigation data models and navigation systems based on specific passing branches, using the visual, structural, and semantic features from human perspectives. For each pair of two branches (i.e., one entry and one exit) passing through the road intersection, the model will output a specific decision-making complexity score. Furthermore, this study develops a route planning algorithm for generating the minimum complexity route to serve relevant navigation applications. This study contributes to human-centered route planning and communication, as well as enabling potential innovative applications in traffic safety studies and sustainable urban and environmental development.
... Klippel and Winter determined the structural salience of landmarks from the semantic expression of the positional relationship between landmarks and navigation routes. This is a kind of qualitative research method focused on navigation route arrangement [26], where it is difficult to form a difference comparison for the structural salience of landmarks. Winter also proposed an extension of the advance visibility assumption, in which landmarks can be identified early along the roadway with valid landmark structural saliency [27], but does not provide a measure of visibility in complex 3D urban environments and miscellaneous road grids. ...
Full-text available
Mastering the relationship between urban landmarks and urban space morphology in urban planning, landscape planning, and architectural design helps maintain the intelligibility of compact urban districts. The objective of the present study was to numerically determine the structural salience of various landmarks in an urban environment and use it to interpret the intelligibility of the city. Combining the measurement method of 3D visibility and the related principles of space syntax, this study develops a new 3D Node–Landmark Grid Analysis Model (3D NL GAM) for structural salience computation of urban landmarks. In this study, a numerical approach is used to construct a 3D simulation model. Firstly, the visibility of each decision node to landmarks in an urban environment, using a 3D digital model, is measured using the 3D isovist component of Rhinoceros and Grasshopper software. Secondly, links among wayfinding decision nodes and landmarks are established to form a 3D NL GAM. The normalized angular integration of decision nodes and the normalized angular choice of landmarks are computed using the principle of space syntax. Thirdly, the structural salience of landmarks is determined with a function of landmark visibility, spatial properties of landmarks, and wayfinding decision nodes. Finally, a case study was carried out by using a 3D NL GAM to analyze three types of urban areas located in Changsha. The results indicated that large-scale natural landscapes have a higher structural salience among the types of landmarks. The structural salience of architectural landmarks in the combined spatial form of combining tall and low building groups has a clear advantage over the form dominated by high-rise building groups. Raising the height of landmark buildings can modify the structure of the grid analysis model and improve the people aggregation of urban space. The 3D NL GAM can quantify the spatial properties and landmark structural salience of a city and can effectively assist in the evaluation of the intelligibility of built or future urban environments.
... Since the individually perceived optimal route differs among travelers, some existing route planning applications do not only take time efficiency into account, but provide personalized route options based on individual preferences for route choice factors (Funke & Storandt, 2015). Further route planning approaches consider the integration of automatically extracted landmark-based information to provide more natural ways for humans to navigate in the environment (Elias & Brenner, 2005;Klippel & Winter, 2005). ...
This paper presents an approach for promoting routes that reduce exposure of road users to areas that should be temporarily avoided due to traffic related or environmental reasons. At the same time, routes that are already heavily polluted should be avoided, in order to reduce the pollution and distribute it more equally in the environment. In this research, we present a design study for creating route maps that intend to visually communicate favorability of route options to the traveler, in the case of increased air pollution. Our proposed method recommends routes that are calculated as the shortest path while minimizing the current concentrations of particulate matter (PM2.5 and PM10) along the route. Based on a dynamic distribution of traffic flows, our system recommends routes that are not necessarily the shortest or fastest (i.e. individually efficient), but rather the options that avoid areas with particularly high air pollution, while prioritizing other, not so polluted areas; and thus are beneficial to individual and public health (socially favorable). We propose seven different visualization variants for representing line and areal objects in a route map that visualize route options based on pollution levels. A user survey showed that while for most of the variants the symbology has been rated as intuitive, visual attractiveness and suitability for communicating pollution information seems to be limited to less complex visualizations that primarily use variations in color. The focus of the paper is to develop design options to optimally communicate favorable routes by designing different cartographic visualization techniques.
... As a combination of these ideas, this paper will introduce Elastic Step DQN (ES-DQN) -a novel multi-step update that dynamically selects the step size horizon based on consecutive state similarity. The intuition behind the idea of consolidating similar states together is in general inspired by spatial information theory which identified that people have a natural inclination to use landmarks to provide route directions (Michon and Denis, 2001;Klippel and Winter, 2005). When providing directions, individuals tend to summarise the instructions based on broader objectives and use landmarks to anchor and help generate mental images (e.g. ...
Full-text available
Deep Q-Networks algorithm (DQN) was the first reinforcement learning algorithm using deep neural network to successfully surpass human level performance in a number of Atari learning environments. However, divergent and unstable behaviour have been long standing issues in DQNs. The unstable behaviour is often characterised by overestimation in the $Q$-values, commonly referred to as the overestimation bias. To address the overestimation bias and the divergent behaviour, a number of heuristic extensions have been proposed. Notably, multi-step updates have been shown to drastically reduce unstable behaviour while improving agent's training performance. However, agents are often highly sensitive to the selection of the multi-step update horizon ($n$), and our empirical experiments show that a poorly chosen static value for $n$ can in many cases lead to worse performance than single-step DQN. Inspired by the success of $n$-step DQN and the effects that multi-step updates have on overestimation bias, this paper proposes a new algorithm that we call `Elastic Step DQN' (ES-DQN). It dynamically varies the step size horizon in multi-step updates based on the similarity of states visited. Our empirical evaluation shows that ES-DQN out-performs $n$-step with fixed $n$ updates, Double DQN and Average DQN in several OpenAI Gym environments while at the same time alleviating the overestimation bias.
Landmark buildings are salient features for spatial cognition on maps. Distinctive outlines are the major visual characteristics that separate landmark buildings from their surrounding environments. The automatic symbolization of landmark outlines facilitates recognition and map production. As users often recognize landmarks by the outlines of their façades from a street view, this study proposes an automatic method for automatically generating representations of the outlines of landmark buildings in four steps: (1) extract outlines from street-view photographs using GrabCut method, (2) vectorize the extracted building outlines, (3) simplify outline shapes, and (4) symbolize the simplified building outlines in three dimensions (3D). We used the proposed method to generate test data with symbolized outlines for eight buildings in a real-world environment for a wayfinding experiment in which the subjects used the building representations to identify landmark buildings and evaluated their perception of the generated maps. The subjects successfully recognized these buildings based on the symbolized outlines on a map, expressed satisfaction with the manually generated 3D symbols, and reported the same or similar ease of building recognition using 2D or 3D symbolized outlines.
Full-text available
People navigating in unfamiliar environments rely on wayfinding directions, either given by people familiar with the place, or given by maps or wayfinding services. The essential role of landmarks in human route communication is well-known. However, mapping the human ability to select landmarks ad hoc for route directions to a computational model was never tried before. Wayfinding services manage the problem by using predefined points of interest. These points are not automatically identified, and they are not related to any route. In contrast, here a computational model is presented that selects salient features along a route where needed, e.g., at decision points. We propose measures to formally specify the salience of a feature. The observed values of these measures are subject to stochastical tests in order to identify the most salient features from datasets. The proposed model is implemented and checked for computability with a use case from the city of Vienna. It is also crosschecked with a human subject survey for landmarks along a given route. The survey provides evidence that the proposed model selects features that are strongly correlated to human concepts of landmarks. Hence, integrating the selected salient features in wayfinding directions will produce directions with lower cognitive workload and higher success rates, compared to directions based only on geometry, or on geometry and static points of interest.
Full-text available
This paper provides a general framework for the analysis of descriptions of routes, in order to account for the way in which spatial cognition is externalized through discourse. Three cognitive operations are assumed to be involved in the generation of this form of spatial discourse: (a) activation of an internal representation of the environment in which navigation will take place; (b) the planning of a route in the subspace of the mental representation currently activated; and (c) the formulation of the procedure that the user should execute to reach the goal. Two major components of the descriptions of routes are considered, those by which speakers refer to landmarks and those which consist of prescribing actions. Descriptions of two routes in a natural environment were collected from 20 undergraduates. A detailed analysis of the protocols was used to establish a classification of items, in which five classes were defined: prescription of actions without referring to any landmark, prescription of actions with reference to a landmark, reference to landmarks without referring to any associated action, description of landmarks, commentaries. Individual protocols were then used to construct more abstract (skeletal) descriptions, reflecting the essentials of the navigational procedure. Skeletal descriptions confirmed that landmarks and their associated actions were key components of route descriptions.
This paper explores the possibility of organizing map design around conceptual spatial representations (CSRs). CSR refers to a mental representation that is instantiated in interaction with a spatial environment, a spatial representational medium, and/or while solving spatial problems. For this approach we coin the term cognitive conceptual. We detail the basic assumptions in a contrastive manner, i.e., by juxtaposing it against a stylized cartographic approach, and indicate future directions for this research.
Discusses the role of landmarks as points of reference in the psychological development of spatial orientation (finding a way, updating, and route following) and the representation of spatial knowledge. Different definitions of the construct of landmark are presented. (PsycINFO Database Record (c) 2012 APA, all rights reserved)