EVALUATION OF AN INDOOR NAVIGATION APPROACH BASED

ON APPROXIMATE POSITIONS

Ory Chowaw-Liebman, Uta Christoph, Karl-Heinz Krempels and Christoph Terwelp

Informatik 4, Intelligent Distributed Systems Group, RWTH Aachen University, Aachen, Germany

Keywords:

Indoor navigation, Mobile computing, Device whispering, Context-based services.

Abstract:

Until now navigation aids have primarily focused on outdoor scenarios, whether driving on highways or, more

recently, walking through cities. These systems use the Global Positioning System (GPS) for position infor-

mation. Indoor navigation however cannot rely on GPS data, as the signals do not penetrate building structure.

Thus other techniques are required to provide position information indoors. In this article the approach to an

indoor navigation system based on the position information provided by the Device Whispering technique is

presented. The position information acquired by Device Whispering is less precise than information acquired

by the Fingerprinting technique but more robust. To compensate the deﬁcit of precision the position informa-

tion is combined with a movement model. This movement model is automatically generated from the maps

which are already required for navigation.

1 INTRODUCTION

Tools for navigational assistance have become an es-

sential element in today’s traveling society. Interac-

tive software, available for mobile phones, is capable

of guiding users who are driving cars, riding bicycles

or walking. Until now such tools have focused on out-

door environments and are hence based on the precise

data of the Global Positioning System (GPS) to deter-

mine the current position of the device, and thus the

user. But indoor navigation bears several challenges.

First, GPS positioning information cannot be used

for indoor scenarios since GPS radio signals do not

propagate into buildings.

Second, within buildings navigation does not rely

on streets or footpaths but on traversable areas and

certain connections between such, e.g. corridors,

rooms, staircases and elevators. Thus such areas and

also the altitude of the user, i.e. the ﬂoor he is cur-

rently standing on, and possible connections to other

ﬂoors have to be considered.

Third, the description of a path needs to be intu-

itively understandable for humans instead of precise

distance instructions as they are used in cars or other

devices with odometers.

The ﬁrst issue, a positioning technique for indoor

scenarios, was addressed among others in (Wallbaum

and Spaniol, 2006). The third issue on intuitive navi-

gational instructions is considered for outdoor envi-

ronments in (Dale et al., 2003). Section 2 gives a more

detailed overview. Based on this current state of the

art a prototype indoor navigation system was devel-

oped and implemented as a software client for a mo-

bile device with WLAN interface (Chowaw-Liebman

et al., 2009), (Chowaw-Liebman, 2009), which is de-

scribed in Section 3. Central data structure of the

system is the Map, described in Section 4, which

also speciﬁes how the Map is generated from user in-

put. Section 5 summarizes the text generation method

used by the prototype to generate natural language in-

structions for routes. Finally, Section 6 concludes this

paper and shows some directions for future work.

2 STATE OF THE ART

This section brieﬂy reviews previous work on indoor

positioning and natural language generation as basis

for our indoor navigation approach.

2.1 Fingerprinting

Fingerprinting (Ohlbach et al., 2006) is an indoor po-

sitioning method based on wireless networks. The

main goal of the Fingerprinting project is to pro-

vide precise position information. The approach is

basically a pattern matching of the received signal

strengths (RSS) for all APs at a position. The RSS at

195

Chowaw-Liebman O., Christoph U., Krempels K. and Terwelp C. (2010).

EVALUATION OF AN INDOOR NAVIGATION APPROACH BASED ON APPROXIMATE POSITIONS.

In Proceedings of the International Conference on Wireless Information Networks and Systems, pages 195-201

DOI: 10.5220/0003032601950201

 SciTePress

the user’s current position are matched to a database

of RSS measurements at known locations, the ﬁnger-

prints. When combined with user tracking and a ”no

teleporting” policy, the achieved precision is reported

to be approximately 1 meter.

2.2 Device Whispering

The Device Whispering technique is also an indoor

positioning approach based on wireless network in-

frastructures. The main idea is to reduce the number

of access points (APs) considered for position estima-

tion to those that are closest in range. This is achieved

by controlling the transmitting power of the WLAN

interface: The device is set to minimum power before

performing an active network scan. The closest AP is

deﬁned as the one answering to the request with the

lowest transmission power used, which is the mini-

mum information possible.

APs can be tagged with geopgraphical position in-

formation, and if multiple APs are available these tags

can be used to approximate the current position of

the user more precisely than just assigning the closest

AP. Various caveats are discussed in (Krempels et al.,

2009).

The method’s lack of precision makes it neces-

sary to use a novel navigation approach, as current

solutions assume precise information. This paper de-

scribes an approach speciﬁcally designed for use with

the Device Whispering technique.

2.3 CORAL

The CORAL system (Dale et al., 2003) is a natural

language instruction generator designed to enhance

existing outdoor navigation tools, accepting route de-

scription in standard formats. In addition, CORAL is

dependent on a geographic information system (GIS)

for additional data about the geography traversed by

the route. The GIS is required to provide landmarks

(like trafﬁc lights, churches and other easily visible

objects) with which the text decription is augmented.

3 PROTOTYPE NAVIGATION

SYSTEM

Any navigation system must convey the route to the

user, typically using a combination of graphical map

(with highlighted path) and textual instructions.

A convenient way to provide route information to

users are natural language instructions (NLI), which

offer rich and ﬂexible means of describing paths.

Figure 1: Components of the navigation system.

NLI, when created by people, are heavily based

on landmarks. Landmarks are distinctive features of

the local geography, e.g. churches, malls, fountains

and trafﬁc lights outdoors, and speciﬁc shops, foun-

tains and staircases indoors. To that effect landmarks

are also used by the CORAL system in its descrip-

tion of (outdoor) routes. One important distinction

of indoor scenarios are different ﬂoors in a building,

which add a third dimension to the geography of a lo-

cation. Landmarks can also represent connections be-

tween ﬂoors, e.g. stairs and elevators, which cannot

be displayed at all in a purely two dimensional map.

This is the approach taken by the Map data structure

of our system.

This prototype system is shown conceptually in

Figure 1, it is centered around the ”Map” data struc-

ture, which makes all navigation information avail-

able to the client software. The dashed parts, repre-

senting the use of a local server are not implemented

by the prototype implementation. In a commercial

system this server would be responsible for providing

information to visitors of the location, most important

the Map, as well as location speciﬁc services and con-

text sensitive information and commands (CSCI). See

(Schilit et al., 1994) for further information on these

features.

In our decentralized approach every location is re-

sponsible for providing the information locally, for in-

stance by using the local WLAN infrastructure which

can be assumed to be available since it is used by the

whispering technique for localization. Thus, it is not

required to establish a network link to the user’s mo-

bile provider to equip the mobile devices with up-to-

date Maps. The prototype implementation however

provides the Map through a data ﬁle saved on the mo-

bile device.

The Map is generated by a preprocessor, which

provides a GUI to create and maintain the Map data

structure. The ”Admin” in Figure 1 is the person re-

sponsible for maintaining a location’s Map, by pro-

viding the data which is needed by the preproces-

sor. In the Map data structure locations are geomet-

rically described by polygons, inside which the APs

and landmarks are positioned. Section 4 describes

how this input data is converted to a location represen-

WINSYS 2010 - International Conference on Wireless Information Networks and Systems

196

tation based on hierarchical graphs. This conversion

allows for an efﬁcient use of the location data which is

of particular importance for mobile clients. The Map

is made available to the client software which is re-

sponsible for all interactions of the system with the

actual end users.

The client software is supposed to run on mobile

devices, which do not have the computational power

and memory available to notebooks or desktop com-

puters. Therefore, a preprocessor was developed to

perform the computationally intensive tasks before-

hand. The client is left with the tasks of comput-

ing routes in the provided Map, communicating these

routes to the user both graphically and verbally per-

form positioning.

4 MAP

The Map data structure is the central element of the

system which encodes all known information about

a location. The two main goals of the data structure

are adaptation to the imprecise position returned by

the Device Whispering technique and support of the

generation of NLI.

Both landmarks and geometry, can be assigned

descriptive ”features”, which provide nouns to de-

scribe elements of the location. The prototype im-

plementation uses a small ontology including words

like ”corridor”, ”room”, ”elevator” and so on for ob-

jects and speciﬁc parts of locations typically encoun-

tered within (public) buildings. These features are

used during text generation to assign nouns, verbs and

articles , so that these landmarks and polygons can de-

scribe a route in natural language.

4.1 Data Structure

The Map is basically represented by a hierarchical

graph, where nodes can contain a sub-graph. Our

Map uses a hierarchy of four levels summarized in

Table 1. For indoor navigation, ”Building” is an in-

tuitive title for the root node, which contains ﬂoors.

The child nodes of the ﬂoors represent the traversable

geography of the location in a two dimensional way.

The geometry is represented by “sectors” which also

take into account the local distribution of access

Table 1: Levels of the Map’s hierarchical graph.

Building

1 Floors

2 Sectorization (Rooms, Corridors)

3 Micro Paths (Skeletons)

points. The sectorization process will be described

below.

Finally, the lowest level of the hierarchy contains

skeletons of the sectors (which are polygons), used

mostly to describe traversal of concave sectors which

occur at corners and intersections. These skeletons

form the basic elements of the paths for which text

instructions are generated.

Two nodes in different sub-graphs can be con-

nected if their parent nodes are connected. This re-

lationship is also invariant in the other direction, i.e.

if two nodes are connected, at least one connection

between the sub-graphs exist. This ensures that a

path on one level of the hierarchy also implies a path

through the sub-graphs of the path’s nodes.

This is an important property when the paths are

to be computed by reﬁning paths along the lower lev-

els of the hierarchy. This incremental approach to

route ﬁnding can reduce the time required to com-

pute routes, as large parts of the graph need not be

considered in each step. For example, if a route on

hierarchy level 1 goes from ﬂoor 1 to ﬂoor 5 using

an elevator, there is no need to consider any other

ﬂoors during route ﬁnding on level 2. The nice prop-

erties of this approach come at the cost of complex

book-keeping, which can be avoided by computing

the complete route on the skeleton level directly.

In any case, start and goal nodes on the skeleton

level have to be found (from there the hierarchy can

be traversed to the coarser levels). The starting node

is chosen to be the skeleton node closest to the users

position. To determine the goal node, the skeleton

edge closest to the goal landmark is considered: from

the two nodes terminating the edge, the one farther

away from the user’s position is used to ensure the

edge under consideration is part of the route.

The prototype implementation selects the node

with the greater Euclidean distance to the current po-

sition, ignoring the geography. This worked well for

the test Map, which represented Terminal 1 of the

Cologne/Bonn Airport. Another option is to just take

any of the two nodes, and add the edge to the route

if the wrong node was selected. The latter approach

may be more stable for some geometries.

The Device Whispering technique deﬁnes a map-

ping from points in space to the closest AP. This prob-

lem

has been studied in various ﬁelds under various

names. In computational geometry this kind of map-

ping is known as Voronoi diagrams, where it plays an

important role in various applications (Aurenhammer,

1991; Aurenhammer and Klein, 2000). Consequently,

efﬁcient algorithms to compute Voronoi diagrams

have been devised, the best known being Fortune’s

Also known as the post-ofﬁce problem.

EVALUATION OF AN INDOOR NAVIGATION APPROACH BASED ON APPROXIMATE POSITIONS

197

Figure 2: A simple location, access points are the circled

dots. The upper image shows the Voronoi regions gener-

ated from the access points. In the bottom picture, the re-

gions have been clipped to the corridors. Artifact regions

generated during clipping can be seen top center below the

cross corridor.

algorithm described in (Fortune, 1986). The voronoi

diagram assignes a polygon (called the Voronoi re-

gion) to every AP such that the mapping describes at

the paragraphs beginning can be solved by point-in-

polygon queries.

These voronoi regions, computed from the APs

positions, are clipped to the polygons used to describe

the location. The sectorization process is illustrated

in Figure 2. The polygons resulting from the clip-

ping stage are also called “sectors” and are assigned

to nodes on level two of the Map. Nodes are con-

nected if their associated sectors share an edge.

Several algorithms to clip simple polygons ex-

ist, all are based on ﬁnding intersections between the

polygons and then traversing the resulting contours,

jumping between the two polygons in such a fashion

that the resulting shape is the desired combination of

the two polygons. One such algorithm is described in

(Weiler and Atherton, 1977).

The second goal, the generation of NLI for a route,

needs to address two requirements:

1. While traversing the route, a user passes by sev-

eral landmarks, namely those contained in the

sectors which are part of the route. The land-

mark’s order of traversal depends on the direction

in which the user is walking within the sector.

2. Sectors can have complex shapes which are dif-

ﬁcult to describe with words. Speciﬁcally cor-

ners and intersections are difﬁcult to extract from

a polygon alone, and difﬁcult operations are to

be avoided on mobile devices. The Map should

therefore be able to describe a concave polygon’s

shape.

Both issues are addressed in the Map data struc-

ture by the inclusion of polygon skeletons on the low-

est level of the hierarchy.

4.2 Map Generation

The generation of the Map proceeds by issueing a

node and sub-graph for every “ﬂoor” level, which are

connected if landmarks inside the ﬂoor are ﬂagged as

connected. This is done by the administrator when

creating a Map with the preprocessor. The ﬂoors’

immediate child nodes are associated with the poly-

gons generated during sectorization. Two sector level

nodes are connected by an edge if their associated

polygons share an edge. The sectors polygons are cre-

ated by clipping the polygons describing the locations

geometry to the voronoi regions computed for the ac-

cess points.

The nodes on the sector level are also associated

with those landmarks and APs which are located in-

side the associated polygon. In the case of APs this

is at most one AP, the one contained inside the sector

(a polygon), which is unique due to construction. But

sectors which do not contain any APs can also occur

as can be seen in Figure 2.

Such “artifact” sectors are treated in the next step

of the algorithm by simply associating them with

the access points from those connected sectors which

do contain an AP, based on the fact that whispering

should classify all those APs among the closest ones.

This assignment strategy is used instead of using all

neighbors, which includes neighbors which were as-

signed APs in this step, and avoids sets of APs depen-

dent on the order in which they are created.

The nodes on the lowest level, representing the

skeleton, are associated with single points, the posi-

tion of the associated skeleton node. The skeleton

nodes are connected naturally through the skeleton

edges. In addition, the connections between the un-

derlying nodes need to be passed to the higher level.

This is done by comparing all child nodes of one sec-

tor with all child nodes of a neighboring sector. For

every edge shared by the associated polygons, the

closest pair of these skeleton nodes is chosen, on one

condition: the line segment connecting the two nodes

WINSYS 2010 - International Conference on Wireless Information Networks and Systems

198

must intersect the polygon edge under consideration.

This ensures that the skeleton edge stays within the

traversable area.

Figure 3: Example sectorization and skeleton of a simple

Map generated by the preprocessor. The skeleton includes

edges across polygons.

Note the triangle conﬁguration of edges close to

the center of Figure 3, where a polygon without a

skeleton node had to be skipped. This is an ugly case

occurring when generating the skeletons from the ge-

ometry. It can be avoided by generating skeletons for

the sectors instead of the geometry.

Finally, landmark-based connections are created

connecting the ﬂoors on level one of the hierarchy.

The plurals in the previous sentence are intentional:

multiple sectors in a ﬂoor can have connections to

multiple sectors in another ﬂoor, represented by dis-

tinct landmarks. In fact, even a single sector can con-

tain both a staircase and an elevator. To this end,

edges on level one of the hierarchy (between distinct

ﬂoors) are assigned sets of landmarks.

When assigning sets of landmarks to the edges,

the edges are assumed to be directed (as available in

paths) to distinguish between the landmarks of the

two ﬂoors. This is done in order to accommodate

the descriptions of routes which use these landmarks,

navigating to the landmark on the ﬂoor where the user

currently is, and continuing at the landmark on the

other ﬂoor where the path resumes.

The resulting generation of the Maps, leaving out

some details discussed above, follows these steps:

1. For each zone:

(a) Compute sectorization according to Figure 2:

i. Generate Voronoi diagram of APs.

ii. Clip Voronoi regions to geometry. Resulting

sectors inherit their features from the geome-

try they are generated from.

(b) Find (and remember) shared polygon edges,

connect the sectors.

(d) Associate those sectors who where not assigned

an AP with those APs of all direct neighbors

Figure 4: Simple polygon with bisectors and resulting

skeleton.

which where assigned an AP in the previous

step.

(e) Compute skeleton for every geometry polygon,

place resulting points in containing sector and

connect.

(f) Connect points from different geometry poly-

gons if a shared polygon edge exists (remem-

bered in step 1(b). Two candidate points are le-

gal if the line connecting the pair intersects the

polygon edge. See Figure 3.

(g) Place landmarks in the containing sectors.

Landmarks connecting zones are remembered

for connection after all zones have been gener-

ated.

2. For the landmarks remembered in step 1(g), cre-

ate the according edges between ﬂoors, and asso-

ciate all landmarks connecting those ﬂoors with

the created edge.

It remains to describe the algorithm which com-

putes the polygon skeletons. This is a novel al-

gorithm which was developed during the course of

this work, and is originally described in (Chowaw-

Liebman, 2009). The algorithms underlying idea is

based on the observation that the medial axis of the

polygon contains the lines bisecting the convex cor-

ners of the polygon, and that a node exists where two

of these lines meet.

The approximation of the medial axis computed

by this skeletonization algorithm computes bisectors

for all vertices of the polygon. These bisectors are

then clipped to the polygon, and intersections be-

tween these bisectors are computed to create the

nodes of the skeleton. Every bisector is assigned the

closest point of intersection it caused (bisectors can

intersect multiple others).

These nodes are connected by traversing the poly-

gon contour: the bisectors associate nodes of the con-

tour with one skeleton node (the closest one kept in

above paragraph), which are simply connected in or-

der of traversal along the contour. Figure 4 shows a

polygon, the bisectors of it’s contour nodes and the

resulting skeleton.

EVALUATION OF AN INDOOR NAVIGATION APPROACH BASED ON APPROXIMATE POSITIONS

199

The skeleton algorithm follows these steps:

1. Create a “contour list” of segments bisecting the

inward angles of all vertices, originating at a node

of the polygon contour and ending at the earli-

est intersection with a polygon edge (length and

polygon edges intersected are recorded). This re-

stricts the segments to lie completely inside the

polygon, according to the Jordan Curve Theorem.

This prevents intersections outside the polygon

and ensures the creation of points guaranteed to

lie inside the polygon. All bisectors are initially

marked as “intersects polygon”.

2. Find the earliest intersection between all bisect-

ing segments (again record length and other edge

intersected). The intersection kept has the lowest

length for both segments crossing. All such edges

are marked as “intersects segment”, yielding pairs

of intersecting segments.

3. All segments still marked as “intersects polygon”

are assigned an opposing node by splitting the

polygon edge at the point where the segment inter-

sects, inserting a ”bisector” pointing opposite to

the original vertex. This ”balancing” is required to

assert the correctness of the next step. Some care

must be taken during insertion to keep the contour

in a consistent ordering, e.g. counter-clockwise.

4. The ”contour list” is walked through. Every seg-

ment creates a skeleton vertex, which is connected

to the previously generated vertex (which is ini-

tialized to the last segments vertex). For segments

intersecting another segment, the vertex created is

at the intersection point determined in step 2. For

polygon intersection segments, the vertex is cre-

ated in the middle of the segment, and by exten-

sion equidistant to the closest polygon edges. Be-

cause the segments lie completely inside the poly-

gon the node is also inside.

5 TEXT GENERATION

Generation of natural language instructions is based

on the skeleton level of the hierarchy. The order of

nodes in the path deﬁnes a direction for the edges,

and with the nodes’ positions each edge deﬁnes a line

segment. Landmarks are now conveniently passed by

in the order of the closest point on the line segment.

Using parametric lines of the form ~y = ~p + x ·

d with

origin ~p and direction

d, the closest points on the line

can be expressed as a scalar value.

The line segments are constructed such that ~p is

the point of the originating node and ~p + 1 ·

d is the

Figure 5: Screen shots of the client in action.

point of the terminating node. That is, the line seg-

ments is deﬁned for values of the parameter x inside

the interval [0, 1]. If the closest point on the line is

outside this intervall it is not on the line segment, and

the landmark can be discarded. Sorting the remain-

ing landmarks based on the parameter for the closest

point arranges them in order of traversal.

These edges with landmarks in order of traversal

form the basic element for text generation: instruc-

tions consider one edge after the other, with some

connectives (e.g. “and” or “then”) and landmarks for

orientation, while the skeleton takes care of corners

and intersections deﬁnition. Text generation is based

on three functions:

traverseBy instructs the user to follow the com-

plete skeleton edge, which is described as passing

by the last landmark along the edge. This land-

mark is used to reassure the user about being on

the right track, so it is left out if the last landmark

is too close to the beginning of the edge.

traverseTo describes movement to a landmark

somewhere along the edge. This is used for the

last edge of the path, along which the target land-

mark is located, a distance which is communi-

cated to the user in approximate terms, see below.

The side on which the landmark should be is also

used in the description.

rotateToward turn to face along the next edge.

This function is responsible for describing the ac-

tion needed at an intersection or corridor.

Directions and distances are discretized into a few

approximate phrases like ”to your left”, ”ahead” or

WINSYS 2010 - International Conference on Wireless Information Networks and Systems

200

”close by”, ”in the middle” and ”at the far end” to

give a few representative examples. These discretiza-

tions provide relative information which is expected

to be of more use to people than precise distances.

The difference is that ”in 68 meters” can mean the

same as ”halfway down the corridor”, the latter being

of better utility as long as users are not equipped with

odometers.

The prototype text generator simply uses

traverseBy followed by rotateToward for every

skeleton edge of the path, except for the last which

is handled by traverseTo. These functions are

implemented based on the ”ﬁll the blanks” approach,

generating sentences of a ﬁxed structure with the

appropriate landmark, geometry type (e.g. corridor or

room) and direction placed in their positions. Figure

5 shows a screen shot of the client software, showing

a route and the generated NLI.

6 CONCLUSIONS

A novel indoor navigation system adapted to the de-

vice whispering technique was presented. The cen-

tral element of the system is the Map data structure,

which was presented and it’s applicability to natural

language instruction generation was shown, see Fig-

ure 5. Tests of the prototype system uncovered that

common WLAN interface cards (i.e. hardware de-

vices) do not modulate transmission power when in-

structed to do so: all access points which are detected

already respond to the lowest power request. Pending

further inquiry, we are currently assuming that trans-

mission power modulation is not implemented, possi-

bly to reduce the chip’s area.

The implemented text generation system produces

NLI which are quite satisfactory, especially concern-

ing its minimalistic simplicity. In both cases, out-

put and generation complexity, the system compares

favourably with CORAL: the generated text is not

quite as eloquent, but is generated with much less ef-

fort. This indicates that our approach of skeletons an-

notated with landmarks provides adequate informa-

tion for the NLI generator.

The very simple approach to text generation

leaves open many possibilities for extension, e.g.

adding more rules to improve both the natural lan-

guage as well as for selecting landmarks to use in the

text.

REFERENCES

Aurenhammer, F. (1991). Voronoi diagrams: A survey of a

fundamental geometric data structure. ACM Comput-

ing Surveys, 23:345–405.

Aurenhammer, F. and Klein, R. (2000). Voronoi diagrams.

In Handbook of Computational Geometry, pages 201–

290. Elsevier Science Publishers B.V. North-Holland.

Chowaw-Liebman, O. (2009). Context-Sensitive Informa-

tion Systems in Wireless Networks. Master’s thesis,

RWTH Aachen University.

Chowaw-Liebman, O., Krempels, K.-H., von St

ulpnagel,

J., and Terwelp, C. (2009). Indoor navigation using

approximate positions. In (Obaidat and Caldeirinha,

2009), pages 168–171.

Dale, R., Geldorf, S., and Prost, J.-P. (2003). CORAL:

Using Natural Language Generation for Navigational

Assistance. In Conferences in Research and Practice

in Information Technology, Vol. 16.

Fortune, S. (1986). A sweepline algorithm for voronoi dia-

grams. In SCG 1986: Proceedings of the second an-

nual symposium on Computational geometry, pages

313–322. ACM Press.

Krempels, K.-H., Patzak, S., von St

ulpnagel, J., and Ter-

welp, C. (2009). Evaluation of directory-less wlan

positioning by device whispering. In (Obaidat and

Caldeirinha, 2009), pages 139–144.

Obaidat, M. S. and Caldeirinha, R. F. S., editors (2009).

WINSYS 2009 - Proceedings of the International Con-

ference on Wireless Information Networks and Sys-

tems, Milan, Italy, July 7-10, 2009, WINSYS is part

of ICETE - The International Joint Conference on e-

Business and Telecommunications. INSTICC Press.

Ohlbach, H., Rosner, M., Lorenz, B., and Stoffel, E.-

P. (2006). NL Navigation Commands from Indoor

WLAN Fingerprinting position data. Technical report,

REWERSE.

Schilit, B. N., Adams, N., and Want, R. (1994). Context-

Aware Computing Applications.

Wallbaum, M. and Spaniol, O. (2006). Indoor position-

ing usingwireless local area networks. John Vincent

Atanasoff Modern Computing, International Sympo-

sium on, 0:17–26.

Weiler, K. and Atherton, P. (1977). Hidden surface removal

using polygon area sorting. In SIGGRAPH ’77: Pro-

ceedings of the 4th annual conference on Computer

graphics and interactive techniques.

EVALUATION OF AN INDOOR NAVIGATION APPROACH BASED ON APPROXIMATE POSITIONS

201