User-Friendly Surveying Techniques for Location-Aware Systems

James Scott

Intel Research

Cambridge, UK

Mike Hazas

Lancaster University

Lancaster, UK

Both formerly at:

Laboratory for Communication Engineering,University of Cambridge, UK

Fine-Grained Location Systems — Ready for End Users?

• Systems with <25cm accuracy exist– Ultrasonic, radio (UWB), vision

• Problem: many of these are deployed only in research labs– They require a PhD to install/maintain!– Developed for research, so this wasn’t an issue

• Motivation: address [one of] the issues preventing deployment of fine-grained location systems for end-users.

Deployment = Installation + Configuration + Surveying

• Installation: simplify using engineering tricks• Configuration: simplify using software tricks• Surveying = determining the location of the

environmentally-placed components in the system.– Not so easy to automate.

Unit X Y Z

0 1.52 0.22 2.53

1 5.25 3.87 2.52

2 3.69 3.85 2.53

Current Surveying Techniques

• Manual techniques: e.g. tape measureVery tedious

Inaccurate over large distances

Lots of chances for human error

• Partly-automated techniques: e.g. theodolite

Automate the measurement bit

Potentially very accurate

Still has large user and hardware requirements

Self-Surveying

• The ability of a location system to gather survey data for itself

• How is this possible?– Location systems determine location using

data from a number of collected “sightings”– Many location systems collect surplus

“sightings” at least some of the time– This surplus data can be used for surveying

Self-Surveying Framework

• The framework for conducting self-surveying has 3 stages:– Data Gathering using the location system– Processing of the data to determine survey data– Combination of surveys from multiple rooms into a

single coordinate space

• This framework is applicable to a wide range of location systems

• Various data-gathering, processing and combination techniques can be used

Data Gathering

• “People” method– Gather data from people walking around with mobile unitsCan be fully “transparent”: users behaving as normal Each location sighting is independent so it’s hard to

determine good ones for surveying

• “Floor” method– Place mobile units on the floor of the room Can cull bad readingsThe height of the tags is identical, thus providing more

“surplus data” for surveying Less “transparent”

Data Gathering (cont.)

• “Frame” method– Use a rigid frame to mount mobile unitsMobile units’ relative positions fully known, so more

surveying data is gathered Requires the use of extra hardware

Processing

• Simulated Annealing– Model the locations of fixed and mobile units– Search for locations which best fit data gathered– Avoids local minima in solution spaceVery general method, potentially applicable to many

location systems/data-gathering methods

• “Inverted” Location System Algorithm– For “frame” data, survey problem looks like an inverted

version of normal location-finding problem• find fixed unit location using known mobile unit locations

– Can use location system’s own algorithmAlgorithm is optimised for location system

Combination of Survey Data

• Data gathered from multiple rooms must be combined to form a single large survey

• Can be done at pre-processing or post-processing stage, depending on the data-gathering method.

Experimental Results

Values are 90th percentile errors in centimetres

• Frame method gives accuracy comparable with accuracy of underlying location system (Bats).

• But every technique is useful, depending on app• Tradeoff between effort and accuracy• Can always resurvey

Method Small Rm 1

Small Rm 2

Small Rm 3

Large Rm 1

Large Rm 2

Mean Mean (Central Units)

People 20 20 24 17 13 19 15

Floor 12 21 26 9 6 15 5

Frame (SA) 9 9 2 10 5 7 4

Frame (Inv) 3 3 2 4 3 3 2

Results: Plan View of Room

• Plan view shows that errors are higher at sides of room • Due to fewer mobile units at edge of room (lower “PDOP”)• Implications for data-gathering: bias survey towards edges

Conclusions and Broader Picture

• Self-surveying shown to be viable• Can facilitate rollout of location systems to

end users• The framework and methods presented are

likely relevant to other location systems

• Related problem: surveying of environment (e.g. walls, furniture) and of important objects (computers, phones)– c.f. Rob Harle’s work at UbiComp 2003

QUESTION TIME

Results: S-curves

Coordinate Space Transformation