See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/251220786

An Egocentric Urban Viewshed: A Method for Landmark Visibility Mapping for

Pedestrian Location Based Services

Article · January 2008

DOI: 10.1007/978-3-540-70970-1_4

CITATIONS READS

21 587

3 authors, including:

Phil Bartie Simon Kingham

Heriot-Watt University University of Canterbury
48 PUBLICATIONS   1,150 CITATIONS    185 PUBLICATIONS   7,048 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

WellConnected New Zealand View project

Cumulative impacts of air pollution exposure on adult physical and mental health View project

All content following this page was uploaded by Simon Kingham on 18 June 2015.

The user has requested enhancement of the downloaded file.

An Egocentric Urban Viewshed:
A Method for Landmark Visibility Mapping for
Pedestrian Location Based Services

Phil Bartie1, Steven Mills1 and Simon Kingham2

1
Geospatial Research Centre (NZ), Christchurch, 8140, NZ
2
Department of Geography, University of Canterbury, Christchurch, NZ

Abstract

A variety of information can be provided to pedestrians using location

based services in support of tasks such as wayfinding. Typically current
location aware systems use proximity to filter databases for contextual in-
formation. We show that a filter based on the visibility of features is a use-
ful additional capability made possible through the use of digital surface
models. A number of visibility metrics are suggested for adoption by a lo-
cation based service, to provide quantitative visibility information so that
items of interest may be ranked according to a meaningful priority. Real
world experiences validate the usefulness of these metrics, and a number
of improvements are suggested.

Keywords: location-based services; digital surface models; viewsheds; visibility

metrics; usefulness test
1 Visibility Modelling and Pedestrian Navigation

There is a growing interest in the development of location based services

(LBS) in support of pedestrian activities, both rural and urban (Jiang and
Yao 2006). The research presented in this paper is in anticipation of de-
vices able to support natural pedestrian wayfinding and navigation in an
urban context. It is argued that the requirements for pedestrian navigation
are quite different from that of vehicle navigation. While junctions form a
key navigation component for motorists, pedestrians more often use land-
marks as cues (Millonig and Schechtner 2007). The urban environment is
defined by these landmarks, their organisation, and interrelationships
(Fisher-Gewirtzman and Wagner 2003) and there is a strong linkage be-
tween what a pedestrian can see and how they comprehend a city. We
therefore argue that integrating the capability to use landmark visibility in-
formation in a navigational device for urban pedestrian use, requires ‘ego-
centric visibility modelling’.
The term ‘viewshed’ has existed in landscape architecture since the
1960s ( Tandy 1967; Lynch 1976), and has been adopted by many disci-
plines. A viewshed depicts areas which can be seen from a designated ob-
servation point, generated by calculating lines-of-sight (LOS) from that
point to all other locations within the study area. Most Geographic Infor-
mation Systems (GIS) offer the functionality to carry out such calculations
(De Smith et al. 2007). Visibility is one of the most commonly used GIS
analysis tools (Davidson et al. 1993) with an extensive catalogue of re-
search work, including siting radio masts (De Floriani et al. 1994), locating
the most scenic or most hidden routes (Stucky 1998), landscape planning
(Fisher 1996), as a weapon surrogate in military exercises (Baer et al.
2005), and in examining spatial openness in built environments (Fisher-
Gewirtzman and Wagner 2003).
Visibility studies within GIS require access to a digital terrain dataset
for the area of interest. These are usually raster grid datasets, considered to
be 2.5 dimensional, recording a single elevation value (z) for any location
(x,y). In general terms these are known as Digital Elevation Models
(DEM), but may be more specifically referred to as Digital Terrain Models
(DTM) if they reflect the elevation values of the bare earth, or Digital Sur-
face Models (DSM) if they capture building and vegetation elevations.
There have been numerous previous studies on urban visibility, these
have tended to use the 2 dimensional boundary of buildings to calculate
isovists ( Tandy 1967; Benedikt 1979; Turner et al. 2001). To more closely
model urban visibility it is necessary to source a DSM at high resolution,
such that building and vegetation profiles are captured accurately. Light
Detection and Ranging (LiDAR) remote data capture techniques have been
shown to be suitable for this in urban studies (Palmer and Shan 2002; Rot-
tensteiner and Briese 2002), and are considered superior for LOS calcula-
tions than using community contributed 3D models of inconsistent and
questionable accuracy, such as with the Google 3D Warehouse.
This paper discusses visibility modelling techniques, and presents a
number of additional metrics that will allow future location based services
to report landmark information in a more intuitive manner, facilitating the
exploration of the city. The paper draws attention to a number of relevant
areas of visibility research, including cumulative visibility. It then explains
the line-of-sight algorithm used in this research, and a supporting database
architecture to model Features of Interest (FOI). Finally the method is
demonstrated in a real world situation.

2 Visibility Analysis

If every terrain cell in a line-of-sight path is considered between an ob-

server and target it is referred to as the ‘golden case’ (Rana and Morley
2002). Although providing the most accurate results from a terrain model
this method is computationally expensive, and therefore much of the pre-
vious research has focussed on techniques to reduce the number of calcula-
tions by considering only visually important cells. Examples of this in-
clude using Triangulated Irregular Networks (TINs) (De Floriani and
Magillo 1994), or filtering based on topographic features (Rana and Mor-
ley 2002). These essentially look to simplify the terrain complexity, or re-
duce the number of observer-target pairs considered in viewshed genera-
tion.
The ‘golden case’ may be maintained whilst offering rapid retrieval of
visibility details by using a Complete Intervisibility Database (CID)
(Caldwell et al. 2003), also referred to as a visibility graph (O'Sullivan and
Turner 2001), or visibility matrix (Puppo and Marzano 1997). These ap-
proaches store the pre-calculated viewshed results from every possible lo-
cation in a study region, meaning future users only require a simple lookup
to return the stored viewshed result. The computational cost of producing
a CID is very high, being an O(n2) calculation, therefore parallel or grid
processing techniques are often used (Llobera et al. 2004).
By storing the complete set of viewsheds a number of additional attrib-
utes are available for analysis, such as the Cumulative Visibility (Wheatley
1995), or Visual Magnitude (Llobera 2003). These depict the total number
of times a cell can be viewed from elsewhere, indicating highly visible re-
gions. Topographically prominent areas such as ridges and peaks often fea-
ture highly, however visually prominent landscapes may not necessarily be
topographically prominent, such as the high intervisibility which occurs in
valleys (Llobera 2003).

2.1 The Urban Cumulative Viewshed

In the context of rural DTMs, peaks and ridgelines form important spaces
which act as barriers, and occupiable vantage points. In urban DSMs these
ridgelines correlate to building roofs, and are not generally occupiable.
This has a number of implications with regard to previous research fo-
cussed on reducing observer-target pairs, which consider ridgelines to be
significant vantage points (e.g. Rana and Morley 2002). To faithfully rep-
licate the situation in urban space, a cumulative visibility map must be
based on a DSM and restrict observer locations to those areas accessible
by pedestrians.
Fig. 4.1 shows cumulative visibility for a section of Christchurch, New
Zealand. This was produced by calculating viewsheds from 20,000 loca-
tions selected at random from all publicly accessible spaces, and summing
the results to indicate how many times a cell can be seen. The DSM was
created at 1 metre resolution from LiDAR return information. To reduce
the impact of edge-effects the analysis was carried out over a larger extent
than that shown (Llobera 2003).

Fig. 4.1. Cumulative Visibility Map for Christchurch, NZ

A number of observations can be made:
• Tall buildings have high cumulative visibility values
(e.g. Christchurch Cathedral - A)
• Open spaces have high cumulative visibility (e.g. Cathedral Square - B)
• The building frontages adjacent to streets have high visibility values for
the entire face (e.g. building C) whilst those faces surrounded by low
rise development only receive high scores for the uppermost sections of
the face (e.g. building D)
• Low rise buildings only visible from a single street receive mid-range
scores (e.g. E)
• Street intersections receive a high score, as they can be viewed from a
number of directions (e.g. F)
• The roofs of buildings have low scores as they cannot be viewed from
street level (e.g. G).

Depending on the intended purpose, the computational cost of cumula-

tive viewsheds may render them inappropriate. De Floriani (1994) sug-
gested that the cost will not be repaid if there are a minimal number of ob-
servation locations. In terms of an LBS where viewshed information is
only required for a single user location, it would therefore follow that a
cumulative viewshed would not be appropriate. However the cumulative
viewshed indicates which areas in an urban scene are important for visibil-
ity analysis, and this information could be used in a strategy to reduce the
number of target locations for real time visibility analysis conducted on a
mobile device.

3 Calculating Egocentric Visibility for LBS

Whilst GISs typically consider the world from above with all areas equally
important, an LBS takes an egocentric viewpoint (Meng 2005; Reichen-
bacher 2005).
A location based service is defined by Jiang and Yao (2006) as an ap-
plication which is both location-aware, and context-aware, therefore re-
quiring information on the user’s position and surroundings. Currently
LBSs use proximity as a spatial filter to retrieve relevant contextual infor-
mation, a notable exception is the Edinburgh Augmented Reality System
(EARS) (Bartie and Mackaness 2006) which is able to filter information
based on the visibility of FOIs from the user’s location.
EARS accesses a database of pre-calculated visibility results for 86
FOIs located around the city of Edinburgh. These results were calculated
using ESRI ArcInfo and stored in a relational database management sys-
tem for rapid sub-second retrieval while on location. As the user explores
the city, Global Positioning System (GPS) values are used to locate the
user, the application reports what can be seen from the current location.
There are a number of drawbacks to this approach including:
• the user’s height is fixed to 1.74m for all visibility calculations
• minimal quantitative information is available regarding feature visibility
• the system is unable to accommodate user or community contributed
FOIs, as viewshed functionality is not provided on board
• any updates to the DSM require all the visibility calculations to be re-
run.

An LBS able to provide the user with information about the visibility of
FOIs would be able to guide the user by referring to landmarks ( Michon
and Denis 2001; Raubal and Winter 2002; Goodman et al. 2004; Ross et
al. 2004; May et al. 2005; Millonig and Schechtner 2007), or inform the
user about the current surroundings in a natural way. It is therefore neces-
sary to provide an LOS algorithm for use in an LBS which can provide
real time quantitative information on the visibility of FOIs.

4 Visibility Implementation for LBS

The visibility algorithms implemented in GISs are often the subject of de-
bate; Fisher (1991) reported that different packages gave significantly dif-
ferent results. Source code for Open Source GIS applications are in the
public domain available for scrutiny, whilst the algorithm implementations
of commercial software is unknown. A useful survey of the visibility func-
tionality in GIS can be found in a publicly available report to the US Army
Line of Sight Technical Working Group (US Army Corps of Engineers
2004).
Riggs and Dean (2007) showed through field trials that predicted view-
sheds and surveyed results had lower discrepancies when using higher
resolution DSMs. For urban studies using LiDAR datasets it is hoped that
discrepancies will be small, although it is acknowledged that a DSM is a
2.5D dataset, and will not report true visibility values under bridges, over-
passes, or under vegetation canopy.
Llobera (2003) introduced the concept of ‘visual exposure’, and sug-
gested that this dynamic aspect of visibility had been overlooked within
previous research. Visual exposure focuses on how much of a feature can
be viewed from the surrounding space, enabling the creation of surfaces to
show in which direction a viewer would need to move to view the target
more, or less, clearly. This technique can be used to find visual corridors,
or visual ridges, and forms a useful basis for considering LOS in the con-
text of LBS.
For this research a toolkit was written to allow experimentation with the
‘golden case’ LOS algorithm. The source datasets were a surface model,
an observer location, and a database of feature locations. For this study the
test area selected was the city of Christchurch, New Zealand. All calcula-
tions were carried out using New Zealand Map Grid (NZMG), with the fa-
cility for a user height to be specified. LiDAR data was sourced from the
Christchurch City Council, and a DSM rendered at 1 metre resolution. For
simplicity vegetation was treated as a visual barrier, although the concept
of partially obscured views through vegetation has been examined
(Llobera 2007).

4.1 The Database Model

A database model was designed such that each FOI entity could be divided
into component parts. For example Christchurch Cathedral (Fig. 4.2) could
be divided into three parts to represent the spire, the main building, and
café annex. Each FOI part could be assigned a number of target locations
so the visually important aspects of the structure could be explicitly mod-
elled. It is necessary to place targets around the base of FOIs so that as
higher targets are obscured by the building’s walls on approach, the LBS
does not consider the FOI to be out of view (Fig. 4.3). It therefore follows
that a greater proportion of the targets on an FOI should be visible as the
observer moves away, although the target will occupy a smaller part of the
field of view.

Part 1 – Spire
Part 2 – Main body
Part 3 – Café

Fig. 4.2. Christchurch Cathedral, NZ (3D model by ZNO, sourced from Google
3D Warehouse)
Fig. 4.3. Target Placement (3D model by ZNO, sourced from Google 3D Ware-
house)

Provision was made within the database structure (Fig. 4.4) to scan tar-
gets according to a visual significance hierarchy. This allowed the LBS to
perform a number of scans at varying target densities, firstly to detect the
visibility of FOIs, secondly to quantify the visibility information. For this
research performance was not a primary consideration, therefore all 2568
cells within the building boundary were used as target locations, at a reso-
lution of 1 target per square metre.

FOIID
FOI FOIName
FOIDescription
1 FOILink

M FOIPartID
FOIID
FOIPart FOIPartName
FOIPartDescription
FOIPartLink

1 N

M M
FOITarget FOICategory
FOITargetID FOICategoryID
FOIPartID FOICategoryDescription
FOITargetLocation
FOITargetLevel

Fig. 4.4. Entity Relationship Diagram for FOI Database

 The allocation of category classes at FOI Part level allows for aspects of
an FOI to be excluded or included in the results, depending on the user’s
preferences. Each FOI Part could belong to more than one category class.

4.2 Line of Sight Metrics

When exploring the urban environment a pedestrian’s view is filled with

features competing for attention. Some distant FOIs may be clearly visible,
yet close items may be partially obscured. If these qualities are to be mod-
elled then a set of corresponding metrics are required. These include a met-
ric for proximity to the object, a metric for the amount of a feature that is
visible, the field of view occupied, and to indicate if the FOI is on the sky-
line.
The approach used here was to consider an LOS from the observer to
each of the FOI targets in turn, recording the vertical visible extent, and the
location of close and distant horizons. These results are then combined to
form a number of metrics for each FOI part. The intention is that future
LBS devices could use these values to deliver information to a user in a
meaningful order, filtering out details not relevant to the current context.

4.2.1 Close Horizon

The ‘Close Horizon’ is calculated by locating the feature which creates the
steepest viewing angle between observer and target, not including the FOI
itself. The elevation of this object, known from the DSM, is used to calcu-
late the intercept of a line of sight with the target, and deduce how much of
the target is visible (Fig. 4.5 - TH1). The obscured area is also recorded
(Fig. 4.5 - TH2).

4.2.2 Distant Horizon

By extending the LOS ray beyond the target it is possible to discover if the
target makes the skyline, or is overshadowed by a more distant object. The
search continues until either it intercepts the DSM, or reaches the same
elevation as the maximum elevation in the DSM dataset. If the ray inter-
cepts the DSM then the ‘Distant Horizon’ location is recorded along with
the elevation value at that point, and both distance behind target and extent
of the object showing (Fig. 4.5 - HD1) are calculated. If no feature is found
the target is designated as being on the skyline.
Fig. 4.5. Line of Sight Details; the Close Horizon is used to calculate how much of
the target is visible; the Distant Horizon information indicates whether the target is
on the skyline or overshadowed by a taller object

The values stored from each of the LOS calculations between the ob-
server and each target point are summarised in Table 4.1.

4.3 Visibility at Feature Level

For an LBS to make use of the target visibility attributes a number of

summaries at the FOI Part level are required. These summaries (Table 4.2)
are intended to provide an LBS with the facility to filter spatial databases,
and to sort results according to various quantitative measures. As the LBS
is able to determine automatically which aspects of a feature are visible, it
can customize the information delivered. For example it is able to report
details of the highly visible spire, but not mention the currently obscured
entrance lobby.
Table 4.1. Line of Sight Target Return Values
Criteria Data Type Details
Visible Boolean Whether a target point can be
(True or False) seen or not from current location

Target Location Point (x,y) The location of the target point

Target Elevation Metres (vertically) The height of the target point

(TH1+TH2)
Distance from Observer Metres (horizontally) The distance from observer to
(Distance O-T) target point

Close Horizon Point (x,y) The location of the tallest object

Location in the line of sight from the ob-
server to the target

Close Horizon Metres (vertically) The height of the tallest object

Height between observer and target
(HC)
Close Horizon Metres (vertically) The amount of target which
Intercept With Target shows above the tallest near ob-
(TH1) ject

On Skyline Boolean Whether the target has a taller

(True or False) visible object behind it, or sky

Distant Horizon Point (x,y) The location of the intercept

Location with a taller object behind the
target (if any)

Distant Horizon Metres (horizontally) The distance from the target to

Distance Behind Target the horizon
(Distance T-DH)
Distant Horizon Metres (vertically) The elevation of the item on the
Elevation (HD1+HD2) horizon visible behind the target

Distant Horizon Metres (vertically) The amount of the horizon that

Intercept (HD1) is visible above the target

Elevation Showing as Ratio Considering any near horizons,

Ratio Of Distance to calculate the ratio of visible ver-
Observer Ratio tical extent divided by distance
(TH1 /Distance O-T) to observer
Table 4.2. Visibility Metrics at FOI Part level
Criteria Data Type Description
Visible Boolean Shows if part of a feature can be seen
(True or False)
Average Distance Metres The average distance to only those
of Visible Targets (horizontally) targets visible to the observer

Maximum Horizon- Degrees The field of view between the widest

tal Field of View (horizontally) visible targets on an FOI

Number of Visible Integer A count of the number of visible tar-

Targets gets

Percentage of Decimal Count of targets visible divided by all

Targets Visible targets on FOI Part

Total Face Area Square Metres The combined total area of feature
Visible frontage visible, when considering
close horizon
Total Face Area Square Metres The area on the frontage which is in
Blocked From View the shadow of near blocking objects

Percentage of Tar- Ratio Count of visible targets on skyline, di-

gets On Skyline vided by all visible targets

Average Visible Metres Average elevation calculated from all

Target Height (vertically) visible targets

Minimum Visible Metres Lowest visible target elevation

Target Height (vertically)

Maximum Visible Metres Highest visible target elevation

Target Height (vertically)

4.3.1 Field of View

Most of these metrics are self-explanatory, however ‘Field of View’ and

‘Total Face Area’ require clarification. The ‘Field of View’ is calculated
between the most extreme targets visible, which make the widest angle. If
an object obscures a side of the FOI frontage then the FOV angle will de-
crease. However if an object blocks a portion in the middle of the FOI with
the outside targets still visible, the FOV will return the widest viewing an-
gle calculated from the outside points, ignoring the obscured portion of the
FOI frontage.
4.3.2 Total Face Area – Visible / Blocked

The ‘Total Face Area’ visible is calculated by summing the area visible
under each target after considering the area obscured by obstacles between
the observer and target, as illustrated in Fig. 4.6. The difference between
DTM and DSM establishes FOI height, removing topography from the re-
sulting area. The Total Face Area Blocked reports the area which is shad-
owed by near objects.

Fig. 4.6. Total Face Area Visible for a Feature of Interest

A number of derived metrics may also be useful such as the ratio of dis-
tant horizon height showing (HD1) over distance behind target, to give an
indication of dominance of any distant features. Also the total visible face
area divided by the average distance to the targets would give an indication
of the presence of an FOI from the user’s viewpoint.

5 Implementation and Evaluation

An initial demonstration of the method was carried out in the city of

Christchurch, New Zealand. Christchurch Cathedral was selected as an
FOI, and divided into 3 component parts as outlined in Fig. 4.2. A number
of observation points were selected to give different views of the Cathe-
dral, photographs were taken at each site, and the GPS locations passed to
the algorithm to return the visibility metrics. Fig. 4.7 shows the location of
the test sites, along with corresponding photographs.
Fig. 4.7. Map of Test Sites in Christchurch, New Zealand (LiDAR data sourced
from Christchurch City Council)
 Table 4.3 shows the results from running the visibility calculations. The
third part of the FOI was not visible from either location A or B, and has
been removed from the table.

Table 4.3. Results from Sites A and B

Location A B
Part 1 Part 2 Part 1 Part 2

Visible Targets (%) 18.2 0.8 17.7 0.6

Percentage Elevation Showing 13.4 8.2 12.9 7.9

of the Visible Targets (%)
Total Face Area Visible (sq m) 202.5 33.4 191.5 28.0

Total Face Area Blocked From 1325.9 376.1 1316.5 318.7

View by Near Horizon (sq m)
Average Distance of All Visible 1636.5 1620.5 721.6 706.4
Targets (m)
Percentage of Targets on Skyline 3.0 0 2.5 0
(%)
Visible Target Height Average: 41.3 25.6 41.9 26.6
(m) Maximum: 59.1 29.0 59.1 29.0
Minimum: 20.1 22.0 26.1 22.0

Maximum Horizontal Field of 0.25 0.11 0.56 0.24

View for Entire FOI (degrees)

The majority of the results are as expected with the more distant site
showing the FOI to occupy a narrower field of view, that the spire (FOI
Part 1) is visible, and none of main building (FOI Part 2) makes the skyline
due to the tall surrounding buildings.
However the results show that the percentage elevation of visible targets
(13.4% at A, 12.9% at B), percentage of targets on the skyline (3% at A,
2.5% at B), and total face area visible (202m2 at A, 191m2 at B) for the
spire (FOI Part 1) go against the expected trend and are slightly greater at
Location A than B. It is also noticeable that a greater extent of the spire is
visible at location A (20.1m to 59.1m at A, and 26.1m to 59.1m at B).
In fact, although counter-intuitive, these values match the real world ex-
perience as seen in the photographs taken from these sites. At Location A
the vertical extent of the spire visible is greater with the majority of the left
side making the skyline, as annotated in Fig. 4.8. At Location B the distant
skyscraper blocks the sky behind the spire, and trees in the foreground ob-
scure the lower aspects of the spire. This is in agreement with the output
values from the algorithm.

Location A Location B
Fig. 4.8. Comparison of Cathedral Spire from Locations A and B

5.1 Additional Analysis

Five further test locations (Fig. 4.9) were selected and the metric results
calculated. Values for total visible area frontage along a transect line pass-
ing from Location C, through D, and E are shown in Fig. 4.10.
A statue and several trees block the view of the Cathedral along this ap-
proach, as reflected by the metrics.
Fig. 4.9. Cathedral Square Test Sites (LiDAR data sourced from Christchurch City
Council)

Considering the percentage elevation of targets visible, at Location C

39.9% of the spire is visible, at Location D the main body of the Cathedral
receives a score of 17.2%, whilst the spire receives a value of 11.8%, indi-
cating the spire is obscured more than the main body. At Location E the
main body received a score of 20.4%, while the spire receives 34.8% indi-
cating the prominence of the spire once more.
The main difference in the view between locations F and G is the prox-
imity of a Police hut (Fig. 4.11). At Location F, the Café (FOI Part 3) is
not visible, whilst at Location G it scores 6.5% (percentage elevation of
visible targets).
Fig. 4.10. Frontage Area Visible against Distance from Feature of Interest

The total face area of the spire at Location F is 615.1m2, whilst at Loca-
tion G it receives a value of 818.3m2. These are in agreement with the pho-
tographic evidence.

Location F Location G

Fig. 4.11. Site F and G photographs

There is an interesting relationship between the field of view, and total

area visible. Fig. 4.12 illustrates this by considering 26,753 locations
around Cathedral Square spaced 1 metre apart.
Fig. 4.12. Field of View and Face Area Visible (LiDAR data sourced from Christ-
church City Council)

Close to a feature the field of view metric scores high, and the total
frontage face area visible is low. As can be seen from the graph the points
which contribute to the peak of face area value points (Set 1) are located
away from the Cathedral on the edge of the square, whilst the highest FOV
values (Set 2) are near the Cathedral. From a viewing experience the face
area values may be considered the most appropriate metric to reflect ‘how
much’ of an FOI can be seen, and should be considered with distance to
quantify the presence of an FOI.

5.2 Mapping the Visibility Metrics

The values from the LOS implementation may be mapped to indicate how
a user’s experience of an FOI would vary across space (Fig. 4.13). In this
example the area in front of the Cathedral between test locations C and D
(Fig. 4.9) enjoys the highest visible percentage of targets, essentially the
clearest view.

Fig. 4.13. Maps of Key Metrics

 Calculating the gradient from any of these datasets will show the magni-
tude and direction of change of the metric, and may be used to indicate in
which direction a user should move to see more, or less, of the FOI.

5.3 Using the Visibility Metrics

These metrics combined with other GIS datasets can provide an LBS with
the ability to be context-aware when delivering information to a user.
When searching spatial databases the visibility metrics may be used to
rank the results showing the most visible items first. In wayfinding, the
visible area values may be used to lead the user towards good viewpoints
for nominated landmarks. Any information delivered to the user would
need to be supported by additional datasets, such that attributes including
the FOI’s name, address, usage, and history could be conveyed to the pe-
destrian.
Although not implemented at this stage, a fuzzy logic layer will be in-
troduced in the next phase of the research so that a natural language engine
may select the most appropriate English terminology to describe the scene.
This will allow the LBS to take on the role of a virtual city guide. The
class limits will be set after conducting user trials to evaluate perceived ob-
ject visibility against the metric results.
To ensure the LBS remains responsive to the user’s movements, the
metrics must be available in real time. This can be achieved in a number of
ways. One method is to pre-cache the results for an area of interest, such
that the mobile client requires only a simple lookup of the corresponding
FOI visibility summaries.
An alternative approach, which will be used for the next phase of this
research, is to implement a client-server architecture whereby a mobile cli-
ent sends the user’s location information across a network to the server
which returns the visibility results for the surrounding region. Pre-caching
on the server side is also possible such that the most commonly visited ar-
eas can be held in memory for improved response times.

6 Conclusion and Future Work

This research has shown that a line of sight algorithm may be used to sup-
ply a number of useful contextual visibility metrics from a LiDAR dataset
in an urban environment. These results form part of a function of visibility
which can be used to prioritise information on features of interest from the
current observation location.
We have shown that it is possible to incorporate a wide range of visual
statistics into reports of object visibility, with consideration of the sur-
rounding cityscape, and the importance of close and distant horizons.
There are a number of areas for further research in this field. The algo-
rithm considers only the physical aspects of visibility which can be calcu-
lated from a DSM. It currently neglects to consider the time of day, or
weather in the visibility calculations.
It would also be beneficial to measure the texture, material, colour and
contrast differences between the target and any distant horizon objects,
such that a metric may be established to indicate how easily an FOI may
be resolved from its background. Aerial imagery might offer a partial solu-
tion suitable for identifying vegetation backgrounds, but roof top colours
will not assist in contrast and colour information from the user’s view-
point, so georeferenced street level photography would be preferable.
Partial visibility through vegetation (Llobera 2007) could be explored
such that lines of sight are able to pass through, and under, canopy layers.
The algorithm should be extended to accommodate seasonal tree canopy
and vegetation density variation. It would also be worthwhile to examine
raw LiDAR returns to produce detailed surfaces for the canopy layer.
Currently a scan in the vertical axis over the FOI returns information on
the distant horizon. This could also be applied across the horizontal axis so
that the surroundings may be used as context to determine if an FOI profile
is significantly different from its neighbours (e.g. skyscraper amongst low
rise buildings).
Referencing a number of FOIs in the metrics would allow an LBS to use
relative descriptions of space, such as “ the Chalice monument is visible to
the right of the Cathedral”. Accessing models from the Google 3D Ware-
house may also be useful such that pictorially the shape and texture of
FOIs could be displayed, allowing the user to more easily identify the tar-
get FOI from surrounding buildings.
The ultimate goal of this research would be a single visibility function
which considers weighted metrics based on visibility, architectural interest,
building form, and social factors such as building use or related interests to
the user, that would allow an LBS to rank in order from a scene those
items which a viewer would consider most interesting. The values would
then be passed through a fuzzy logic layer and a natural language engine to
generate appropriate sentences to convey to the user descriptions of their
surroundings, such that the LBS acts as a virtual city guide.
Acknowledgements

The authors would like to thank the Christchurch City Council for use of
their LiDAR dataset, and ZNO for the 3D model of Christchurch Cathe-
dral. The research would not have been possible without funding from the
University of Canterbury, New Zealand. We would also like to thank Dr
William Mackaness for comments on an early draft of the paper.

References

Baer W, Baer N, Powell W, Zografos J (2005) Advances in Terrain Augmented

Geometric Pairing Algorithms for Operational Test. Modelling and Simula-
tion Workshop, Las Cruces, NM
Bartie PJ, Mackaness WA (2006) Development of a speech-based augmented real-
ity system to support exploration of cityscape. Transactions in GIS 10:63-86
Benedikt ML (1979) To take hold of space: isovists and isovist fields. Environ-
ment and Planning B 6:47-65
Caldwell D, Mineter M, Dowers S (2003) Analysis and Visualisation of Visibility
Surfaces. http://www.geocomputation.org/2003/Papers/Caldwell_Paper.pdf
(accessed: 2nd November 2007)
Davidson DA, Watson AI, Selman PH (1993) An evaluation of GIS as an aid to
the planning of proposed developments in rural areas. Geographical Informa-
tion Handling: Research and Applications, pp 251–259
De Floriani L, Magillo P (1994) Visibility algorithms on triangulated digital ter-
rain models. International Journal of Geographical Information Systems 8:13-
41
De Floriani L, Marzano P, Puppo E (1994) Line-of-sight communication on ter-
rain models. International Journal of Geographical Information Systems
8:329-342
De Smith MJ, Goodchild MF, Longley P (2007) Geospatial Analysis: A Compre-
hensive Guide to Principles, Techniques and Software Tools. Troubador Pub-
lishing
Fisher-Gewirtzman D, Wagner IA (2003) Spatial openness as a practical metric
for evaluating built-up environments. Environment and Planning B: Planning
and Design 30:37-49
Fisher PF (1991) First experiments in viewshed uncertainty: the accuracy of the
viewshed area. Photogrammetric Engineering & Remote Sensing 57:1321-
1327
Fisher PF (1996) Extending the applicability of viewsheds in landscape planning.
Photogrammetric Engineering and Remote Sensing 62:1297-1302
Goodman J, Gray P, Khammampad K, Brewster S (2004) Using Landmarks to
Support Older People in Navigation. 6th International Symposium on Mobile
Human-Computer Interaction: MobileHCI 2004, Springer, Glasgow, UK
Jiang B, Yao X (2006) Location-based services and GIS in perspective. Comput-
ers, Environment and Urban Systems 30:712-725
Llobera M (2003) Extending GIS-based visual analysis: the concept of visual-
scapes. International Journal of Geographical Information Science 17:25-48
Llobera M (2007) Modelling visibility through vegetation. International Journal of
Geographical Information Science 21:799-810
Llobera M, Wheatley DW, Steele TJM, Cox S, Parchment O (2004) Calculating
the inherent visual structure of a landscape ('total viewshed') using high-
throughput computing. XXXII International Conference - Computer Applica-
tions in Archaeology 2004 - Computer Applications and Quantitative Meth-
ods in Archaeology, Beyond the Artifact: Digital Interpretation of the Past,
Prato, Italy
Lynch K (1976) Managing the sense of a region. MIT Press Cambridge
May AJ, Ross T, Bayer SH (2005) Incorporating landmarks in driver navigation
system design: An overview of results from the REGIONAL project. Journal
of Navigation 58:47-65
Meng L (2005) Ego centres of mobile users and egocentric map design. In: Meng
L, Zipf A, Reichenbacher T (eds) Map-based Mobile Services. Springer, Ber-
lin
Michon PE, Denis M (2001) When and why are visual landmarks used in giving
directions? Spatial Information Theory, Foundations of Geographic Informa-
tion Science, 292-305
Millonig A, Schechtner K (2007) Developing landmark-based pedestrian-
navigation systems. IEEE Transactions on Intelligent Transportation Systems
8:43-49
O'Sullivan D, Turner A (2001) Visibility graphs and landscape visibility analysis.
International Journal of Geographical Information Science 15:221-237
Palmer TC, Shan JA (2002) Comparative study on urban visualization using
LIDAR data in GIS. URISA Journal 14:19-25
Puppo E, Marzano P (1997) Discrete visibility problems and graph algorithms. In-
ternational Journal of Geographical Information Science 11:139-161
Rana S, Morley J (2002) Optimising Visibility Analyses Using Topographic Fea-
tures on the Terrain. Centre for Advanced Spatial Analysis, University Col-
lege London, London, UK
Raubal M, Winter S (2002) Enriching Wayfinding Instructions with Local Land-
marks. In: Egenhofer MJ, Mark DM (eds) Geographic Information Science.
Proceedings of the 2nd International Conference, GIScience 2002, Boulder,
CO, USA. Springer, Berlin Heidelberg
Reichenbacher T (2005) Adaptive egocentric maps for mobile users. In: Meng L,
Zipf A, Reichenbacher T (eds) Map-based Mobile Services. Springer, Berlin
Riggs PD, Dean DJ (2007) An investigation into the causes of errors and inconsis-
tencies in predicted viewsheds. Transactions in GIS 11:175-196
Ross T, May A, Thompson S (2004) The Use of Landmarks in Pedestrian Naviga-
tion Instructions and the Effects of Context. 6th International Symposium on
Mobile Human-Computer Interaction: MobileHCI 2004. Springer, Glasgow,
UK
 Rottensteiner F, Briese C (2002) A New Method for Building Extraction in Urban
Areas from High-Resolution Lidar Data. International Archives of Photo-
grammetry and Remote Sensing 34:295-301
Stucky JLD (1998) On applying viewshed analysis for determining least-cost
paths on Digital Elevation Models. International Journal of Geographical In-
formation Science 12:891-905
Tandy CRV (1967) The isovist method of landscape survey. Symposium: Meth-
ods of Landscape Analysis, pp 9–10
Turner A, Doxa M, O'Sullivan D, Penn A (2001) From isovists to visibility
graphs: A methodology for the analysis of architectural space. Environment
and Planning B: Planning and Design 28:103-121
US Army Corps of Engineers (2004) Line-of-Sight Compendium.
http://www.tec.army.mil/operations/programs/LOS/LOS Compendium.doc
Wheatley D (1995) Cumulative viewshed analysis: a GIS-based method for inves-
tigating intervisibility, and its archaeological application. Archaeology and
Geographical Information Systems: A European Perspective

View publication stats