### 120820_Locating in Fingerprint Space

```LOCATING IN FINGERPRINT SPACE:
WIRELESS INDOOR LOCALIZATION
WITH LITTLE HUMAN
INTERVENTION
Zheng Yang, Chenshu Wu, and Yunhao Liu
MobiCom 2012
- Sowhat 2012.08.20
OUTLINE

Introduction

System Design

Evaluation

Discussion

Conclusion
OUTLINE

Introduction

System Design

Evaluation

Discussion

Conclusion
MOTIVATION


Site survey
Time-consuming
 Labor-intensive
 Vulnerable to environmental dynamics
 Inevitable

OBJECTIVE
Wireless Indoor Localization
Approach
Floor Plan
User
Movement
OUTLINE

Introduction

System Design

Evaluation

Discussion

Conclusion
LIFS, SYSTEM ARCHITECTURE
Geographical dist.
≠
Walking dist.
MULTIDIMENSIONAL SCALING (MDS)

Information visualization for exploring
similarities/dissimilarities in data
STRESS-FREE FLOOR PLAN



Geographical distance ≠ Walking distance,
Ground-truth floor plan –
conflict with measured distance
Sample grids in a floor plan (grid length l = 2m)
Distance matrix D = [dij],
dij = walking distance between point i and j
MDS

Stress-free floor plan – 2D & 3D
FINGERPRINT SPACE –
FINGERPRINT & DISTANCE MEASUREMENT

Fingerprints and distance collection
Record while walking
 Footsteps every consecutive steps by accelerometer
 Set of fingerprints, F = {fi, i = 1~n}
Distance(footsteps) matrix, D’=[d’ij]


Pre-processing

Merge similar fingerprints (δij<ε)
Twice integration  Distance: Noice
Local variance threshold method  Step count
 Stride lengths vary?  MDS tolerate measurement errors

FINGERPRINT SPACE –
FINGERPRINT SPACE CONSTRUCTION

1.
2.
10x sample locations in stress-free floor plan
First several days for training

d’ij unavailable  d’ij = d’ik + d’kj

Shortest path  update D’
all-pairs of fingerprints
 Floyd-Warshall algorithm


MDS  Fingerprint space 2D & 3D
MAPPING –
CORRIDOR & ROOM RECOGNITION

Corridor recognition (Fc)
Higher prob. on a randomly chosen shortest path
 Minimum spanning tree
 Betweenness
Classify fingerprints into
 Watershed

1.
2.

the corridor or rooms
Size(corridor) / Size(all)
Large gap of betweenness values
Room recognition (FRi)

k-means algorithm
(k = number of rooms)
MAPPING –
REFERENCE POINT


Fingerprints collected near “doors”
PD = {p1, p2, …, pk}, stress-free floor plan
FD 1. Map near-door fingerprints
, fingerprint space
to real locations (FD → PD)
2. Map roomsNear-door
to rooms
fingerprints, FD,
labeled with real locations

distance matrix D and D’ 
cosine similarity
l = (lp1, lp2, …, lp k-1)
l’ = (lf1, l’f2, …, l’f k-1)
MAPPING –
SPACE TRANSFORMATION

Floor-level transformation
Stress-free floor plan ≠ Fingerprint space
∵ translation, rotation, reflection
 Transform matrix,



xi = coordinate of fi ∈ FD
yi = coordinate of pi ∈ PD
For fingerprint with coordinate x
real location = sample location closest to Ax + B
Room-level transformation
Room by room
 Doors and room corners as reference point
 Transformation matrix

OUTLINE

Introduction

System Design

Evaluation

Discussion

Conclusion
HARDWARE AND ENVIRONMENT
 Typical office building covering 1600m2

16 rooms,
5 large – 142m2, 7 small, 4 inaccessible
 26 Aps, 15 are with known location
 2m x 2m grids, 292 sample locations

EXPERIMENT DESIGN
5 hours with 4 volunteers
 Fingerprints recording – every 4~5 steps (2~3m)
 Accelerometer –
work in different frequency based on detecting
movement
 600 user traces, with 16498 fingerprints
 Corridor, >500 paths
Small rooms, >5 paths
Large rooms, >10 paths
 Half of data used for training,
half …………………... in operating phase

THRESHOLD VALUE OF
FINGERPRINT DISSIMILARITY
STEP COUNT

5 ~ 200 footsteps

Error rate = 2% in number of detected steps

Accumulative error of long path
Unobvious performance drop
 ∵ only use inter-fingerprint step counts

FINGERPRINT SPACE

795 fingerprints when ε = 30
CORRIDOR RECOGNITION

Refining
Perform MST iteratively
 Sift low betweenness
 Until MST forms a single line

ROOM RECOGNITION
REFERENCE POINT MAPPING
POINT MAPPING
• 96 percentile < 4m
• Average mapping error = 1.33m
LOCALIZATION ERROR
Emulate 8249 queries using real data on LiFS
 Location error

Average,
LiFS = 5.88m
 Percentile of LiFS
80 < 9m / 60 < 6m
 Caused by
symmetric structure
 Fairly reasonable!


Room error = 10.91%
OUTLINE

Introduction

System Design

Evaluation

Discussion

Conclusion
DISCUSSION

Global reference point
Last reported GPS location
Locations of APs
Similar surrounding sound signature
…
 Could be added in LiFS for more robust mapping
 Key for symmetric floor plans / multi-floor fuildings


Large open environment
OUTLINE

Introduction

System Design

Evaluation

Discussion

Conclusion
CONCLUSION

LiFS
Spatial relation of RSSI fingerprints + Floor plan
 Low human cost

