INDOOR NAVIGATION USING APPROXIMATE POSITIONS

Ory Chowaw-Liebman, Karl-Heinz Krempels

Janno von St

ulpnagel 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:

Navigation aids have usually concentrated on the great outdoors, whether driving on highways or, more re-

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

mation. Indoor navigation cannot rely on GPS. In order to provide position information indoors, a technique

called device whispering was developed.

The following presents an indoor system currently being developed, which is suited to the imprecise position

information provided by device whispering.

1 INTRODUCTION

Computerized aids to navigation are the natural exten-

sion to printed maps. Since the advent of GPS, these

tools have become interactive, instead of being lim-

ited to providing precomputed routes with graphics.

The ﬁrst systems where targeted at supporting vehicle

navigation. Early interactive devices used dedicated

hardware, but mobile phones are starting to be used

as alternative hardware.

This has in turn motivated navigation systems for

pedestrians, e.g. Qiro

or Nav4All

. Qiro actually

uses only the current cell as information, and only if

it is not ambiguous; users can manually input their

current position (see FAQ on Qiro’s web page). The

use of trilateration, or even multilateration could be

a possibility, but are computationally expensive, most

likely preventing a satisfactory user experience.

Further, tri- or multilateration based on neighbor

cells’ signal strengths have difﬁculty coping with the

various propagation effects present in cell based net-

works. For instance, reﬂection of radio waves often

allows a connection to the base station without a di-

rect line of sight - breaking the assumption lateration

techniques are based on. In case of indoor systems,

which is considered here, GPS cannot be used for po-

sitioning, because the radio waves used by the GPS

system do not propagate through roofs (or even dense

foliage). However, cell based networks in the form of

www.qiro.de

www.nav4all.com

wireless local area networks (WLAN) can be used for

positioning.

Currently, research considering indoor navigation

concentrates on improving the position data com-

puted. One possibility, used in (Ohlbach et al., 2006),

is ﬁngerprinting, which provides detailed informa-

tion about local propagation effects. Another tech-

nique for indoor positioning named Device Whisper-

ing (Krempels and Krebs, 2008; Patzak, 2009), was

developed at the Department of Computer Science,

Informatik 4, of the RWTH Aachen. This method

is based on wireless local area networks, and trades

accuracy for speed. After a short description of the

technique, a navigation method is presented, which

is currently being developed as a real life application

based on whispering.

2 DEVICE WHISPERING

The main idea of the Device Whispering technique

is to reduce the access points considered for posi-

tion estimation. This is done by controlling the trans-

mitting power WLAN interface: The device is set to

minimum power, then queries access points for man-

agement information. The closest access point is de-

ﬁned as the one answering to the request with the least

amount of transmission power used, which is the min-

imum information possible.

Access points can be tagged with positions, and

if multiple access points are available these tags can

168

Chowaw-Liebman O., Krempels K., von Stülpnagel J. and Terwelp C. (2009).

INDOOR NAVIGATION USING APPROXIMATE POSITIONS.

In Proceedings of the International Conference on Wireless Information Networks and Systems, pages 168-171

DOI: 10.5220/0002189301680171

 SciTePress

be used to approximate the current position some-

what more precisely than just giving the closest ac-

cess point. Various caveats are discussed in (Patzak,

2009), but this work will assume only knowledge of

the closest access point. The whispering method is

also robust against signal multi-path propagation and

power oscillations or automated adaption of access

points’ transmitting power.

Since the method is based on indoor infrastruc-

ture, reception should not be a problem, assuming ex-

isting infrastructure. But WLAN for Internet access

is more and more provided as a service to guests.

The method’s lack of precision makes it necessary

to use a novel approach, as current solutions assume

precise information. One such approach is presented

here, which is based on a sectorization of the map

based on the positions of the access points.

3 MAP GENERATION AND

REPRESENTATION

The whispering technique deﬁnes a mapping from a

point in space to the closest element in a set of spe-

cial points, the access points. This type of mapping,

known as Voronoi Diagrams, is an important concept

in computational geometry (Aurenhammer and Klein,

2000; Aurenhammer, 1991). Efﬁcient algorithms to

compute the voronoi diagram for a set of points ex-

ist, see (Aurenhammer and Klein, 2000; Barber and

Dobkin, 1996). Thus, a natural, though idealized, sec-

torization of a ﬂoor plan can be generated automati-

cally by calculating the voronoi diagram of the access

points, and clipping the resulting sectors to the ﬂoor

plan.

Based on the assumption that Whispering acts as

a ﬁlter against propagation effects, such an idealized

sectorization should be sufﬁcient. Technically this

idealization can be reduced by using a so called ap-

proximate Voronoi Diagram, which has fuzzy edges.

Instead, the system adds way-points on or close to

edges (in the next sector), to lead the user out of am-

biguous areas. 4 discusses how way-point instruc-

tions are communicated to the user.

The actual map used for navigation operates on

a graphs of sectors with edges connecting adjacent

sectors. Features of the geometry are positioned in-

side the containing sectors. These can be Targets (e.g.

shops or cafs), Landmarks (e.g. statues or fountains,

see 4.1) or connective features like stairs or elevators.

Connective features are associated with edges con-

necting locations not connected by normal sectors.

Thus map representation is supplied to client de-

vices by a local server, which can be contacted using

the WLAN infrastructure required for positioning.

Figure 1: A simple location, access points are the circled

red dots. The center image shows the voronoi regions gen-

erated from the access points. In the bottom picture, the

regions have been clipped to the corridors. Artifact regions

generated during clipping can be seen top center below the

cross corridor.

4 NAVIGATION COMMANDS

FROM APPROXIMATE

POSITIONS

The generation of navigation commands must distin-

guish two types of navigation tasks:

INDOOR NAVIGATION USING APPROXIMATE POSITIONS

169

Long Paths have endpoints located in different sec-

tors. Finding such paths in the graph representa-

tion is a simple application of the A* algorithm,

(Russel and Norvig, 2003; S

anchez-Crespo Dal-

mau, 2003). This type of path is represented as a

series of adjacent sectors. Long paths can be dis-

played graphically, laid over the map.

Short Paths remain inside a sector. As long paths

can be expressed as a series of short ones,

generating succinct instructions for the latter is

paramount.

For long paths, position information can deter-

mine the sector where the mobile device is located.

Using this information, it is possible to restrict the

navigation instructions to the current location. For

short paths, position information is at best an approx-

imation and ﬁne grained instructions like “take the

next corridor to the right” are not feasible. Natu-

ral language allows the use of coarse grained

rela-

tive instructions: Instead of alerting the user when no

misunderstanding is possible, a navigation instruction

should unambiguously describe the target, the corri-

dor in this case.

Natural Language Instructions are attractive for

this application (Reiter, 2000; Kray and Blocher,

1999), as their rich descriptive abilities are able to sin-

gle out targets from a group, if instructions are con-

structed with care. This can be be done using adjec-

tives which apply to the target (e.g. “the green Col-

umn” or “the leftmost door”), or by giving a spatial re-

lationship to a Landmark (see e.g. (Elias et al., 2005;

Tversky and Lee, 1999; Lazkano et al., 2005)).

The lack of rigidly detailed instructions may seem

inappropriate when navigation instructions are con-

sidered from the point of view of a car’s driver. For

example the instruction “take the next turn left” al-

ways refers to a road for cars, but a pedestrian could

turn left into a store. As pedestrians are less con-

strained than cars in their movement, usually travers-

ing areas (Gaisbauer and Frank, 2008), the construc-

tion of “short paths” is predominated by avoiding

clumps of people, fountains and other landmarks as

well as adaption the movement of other pedestrians,

forming ﬂows and eddies

. These considerations

show that ﬁne detailed instructions are not as desir-

able for pedestrians as they may be for cars: Navigat-

ing across a plaza ﬁlled with a milling crowd is best

left to a humans cognitive abilities.

The remainder of this section elaborates on these

topics.

That is, without relying on up-to-date exact positions.

Flocking in AI terms.

4.1 Landmarks

Landmarks can be anything, as long as it is easily

seen. Stores and other corporate entities (preferably

with a nice iconic logo) are useful. In cities, Stat-

ues, fountains, buildings, the cologne cathedral and

the tour d’Eiffel, and other elements of the surround-

ings are used which are less interesting for indoors ap-

plications. Staircases, elevators, emergency exits and

phone booths (while they exist) are exemplary of ar-

chitectural features which make for good indoor land-

marks. Navigational targets can also serve as land-

marks, if they are not targets of the current path. Nat-

ural language nicely distinguishes this as e.g. “go to

the booth by the stairs” versus “climb the stairs by the

booth”

Landmarks can be displayed on the map by draw-

ing icons and logos or even images at the appropriate

places. This gives users hints for in-sector naviga-

tion where the approximate positions can not be used

to generate micro-instructions graphically without be-

coming confusing. The graphical representations of

landmarks need to be provided together with the map

information.

Landmarks are regularly used by humans in nat-

ural language instructions. Human languages of-

fer a plethora of ways to express absolute directions

(straight ahead, turn left, in and out), relative locations

(By the stairs, next to the record store) and descrip-

tion (colors, shapes, labels), which can further aid in

in-sector route description.

4.2 Natural Language Instructions

Natural Language is an familiar means to convey in-

formation. It has the capability to convey naviga-

tion information with closer emphasis on signiﬁcant

or detail. For instance, in the above example (booth

and stairs), natural language can indicate whether the

booth or the stairs, or both, are landmarks, based on

the next way-point. It is also able to generate static

instructions, which where identiﬁed as a necessity de-

scribing short paths inside a sector.

Instructions can be ﬁne grained when using spatial

and temporal constructs to aid navigation inside a sec-

tor. Relative instructions can try to determine what is

on the left and what is on the right side, for example

when entering a large hall from a corridor. In gen-

eral, assuming the user walked in a straight line from

the last sector to the current one, directions relative to

the users orientation (like “left” or “right”). As men-

tioned in the example at the beginning of this section,

context based information can be used to place em-

phasis on important elements of the Locations (stairs

WINSYS 2009 - International Conference on Wireless Information Networks and Systems

170

which must be used) and delegating lesser ones to de-

scriptive items, based on the currently scheduled ac-

tivity. Relative descriptions can be used to associate

a target with a mote prominent feature in the close

vicinity.

Annotating landmarks or prominent features of

the locations (e.g. statues or fountains) with descrip-

tive adjectives makes it possible to convey succinct

descriptions of a location or feature.

For a given language, a grammar deﬁnes how sen-

tences are formed. While computers are mostly con-

cerned with parsing (Aho et al., 2007), that is under-

standing a language, the generative aspect of gram-

mars is well known. Based on the current way-point

and the next one, as well as the knowledge provided

by the extended map, instructions can be generated by

expanding productions of the grammar based on the

available information. The expansion of productions

is controlled by a set of rules, which take the available

information into account.

Finally, natural language instructions can be used

to convey information related to the users current ac-

tivity, but not the geometry itself. Platform changes,

deadlines like the trains’ departure, and gotchas like

the fact the train will be split at a later stop (only the

back half going to the users destination), can not be

conveyed on a map, and thus beneﬁt from a natural

language interface.

5 CONCLUSIONS

A system for indoor navigation based on the Whisper-

ing technique, which is currently being implemented,

was described. While imprecise, it counteracts prop-

agation effects. This motivates the use of voronoi

diagrams to represent an idealized sectorization of

locations, based on access points provided by local

WLAN infrastructure.

User guidance without precise position informa-

tion relies heavily on natural language instructions.

The graphical representation, while simple and fa-

miliar, provides only an overview of the path. Nat-

ural language instructions are used to provide static

path descriptions where dynamic descriptions would

require a position with more precision than available

using Whispering.

REFERENCES

Aho, A. V., Lam, M. S., Sethi, R., and Ullman, J. D. (2007).

Compilers: Principles, Techniques and Tools, 2nd Ed.

Pearson Education.

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.

Barber, C. B. and Dobkin, D. P. (1996). The quickhull algo-

rithm for convex hulls. ACM Transactions on Mathe-

matical Software, 22:469–483.

Elias, B., Pelke, V., and Kuhnt, S. (2005). Concepts for the

cartographic visualization of landmarks.

Gaisbauer, C. and Frank, A. U. (2008). Wayﬁnding model

for pedestrian navigation.

Kray, C. and Blocher, A. (1999). Modeling the basic mean-

ings of path relations. In Proceedings of the 16th IJ-

CAI, pages 384–389. Morgan Kaufmann.

Krempels, K.-H. and Krebs, M. (2008). Improving

directory-less wlan positioning by device whispering.

Lazkano, E., Astigarraga, A., Sierra, B., inez Otzeta, J.

M. M., and Ra

o, I. (2005). Natural landmark based

navigation.

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

(2006). Nl navigation commands from indoor wlan

ﬁngerprinting position data.

Patzak, S. (2009). Aktive wlan positionierung. Master’s

thesis, RWTH Aachen University.

Reiter, E. (2000). Building natural language generation sys-

tems.

Russel, S. and Norvig, P. (2003). Artiﬁcial Intelligence: A

Modern Approach, 2nd Ed. Pearson Education.

anchez-Crespo Dalmau, D. (2003). Core Techniques and

Algorithms in Game Programming. New Riders Pub-

lishing.

Tversky, B. and Lee, P. U. (1999). Pictorial and verbal tools

for conveying routes. Springer.

INDOOR NAVIGATION USING APPROXIMATE POSITIONS

171