locating in fingerprint space: wireless indoor localization with little human intervention....
TRANSCRIPT
wine.korea.ac.kr
WINE
Locating in fingerprint space: wireless indoor localization with little human intervention.Proceedings of the 18th annual international conference on Mobile computing and networking. ACM, 2012.
RSSI FingerprintAutomatic Radio Map Generation
Presenter: Jongtack Jung
wine.korea.ac.kr
WINE 2
Localization technique where each location is associated with the RSSI Fingerprint of the location
Arbitrary fingerprint from an un-known location is matched with the radio map, and best fittingoption is selected
RSSI Fingerprint Method?
wine.korea.ac.kr
WINE 3
Site survey process Training phase a.k.a. calibration
Fingerprint A set of RSS values obtained at a location
Radio map The map of RSS fingerprints associated with the location
MDS (Multi-Dimensional Scaling) A method to map points into given dimensional space where only the dis-
similarities among the points are known Stress (MDS term)
How well the mapping expresses the dissimilarity matrix
Terminology
wine.korea.ac.kr
WINE 4
PROS All APs can be exploited
Including password pro-tected APs
Fast execution
Best accuracy of all
Pros and Cons of RSSI Fingerprint
CONS Necessary training period
Necessary maintenance
EXPENSIVE Training and maintenance
require human labor
wine.korea.ac.kr
WINE 5
The cost of RSSI Fingerprint method can be reduced using automated status update mechanism
The concept of automation is adopted Many methods have been attempted to auto-
mate the process of site surveying
RSSI Fingerprint
wine.korea.ac.kr
WINE 6
Main Idea Since the geographic distance does not really represent the actual walk-
ing distance of two positions, use walking distance to create a map
Concept Two position close together in walking distance means similar fingerprint The number of footsteps obtained from accelerometer provides the dis-
tance between locations Hybrid of fingerprint and dead reckoning
Locating in Fingerprint Space – Innovation!
wine.korea.ac.kr
WINE 7
Overview
wine.korea.ac.kr
WINE 8
Stress The accuracy of MDS If a distance map can be perfectly resolved
in given dimensions, the stress is 0 Given dataset, higher dimension means
less stress Draw 3D floor plan
Disparity between two locations is given with the number of footsteps
The distance between two nodes in the graph is the actual walking distance
Footstep recognition The number of footsteps is obtained from
accelerometer – only the #steps, not the distance
Stress-free Floor Plan
wine.korea.ac.kr
WINE 9
The distance between finger-prints can also be expressed with disparity map
MDS algorithm is tolerant to measurement errors on its own
If no user actually passes through a particular pair of fingerprints, the value is calculated with shortest path
Fingerprint Space
High dimension floor plan (top) and fingerprint space map(bottom)
wine.korea.ac.kr
WINE 10
With above equation as dissimilarity, two points having the value less than threshold are considered as the same point and merged together.
Pre Processing
wine.korea.ac.kr
WINE 11
Fingerprint space needs to be mapped on stress-free floor plan
Floor-level transformation Use simplest linear transformation and
shift between the two graphs
Room-level transformation Detect rooms with K-cluster method and
apply MDS to each room, and then match them
Space Transformations
MST of fingerprint space map
wine.korea.ac.kr
WINE 12
The virtual high dimensional data needs to be mapped on ac-tual floor plan
Corridor recognition MST betweenness
Room Recognition Clustering of nodes
Reference Point Mapping Point where values change largely
are considered as doors
Mapping
wine.korea.ac.kr
WINE 13
Betweenness Centrality Dis-tribution of all points
K-Means Clustering of all points
Evaluation Results
wine.korea.ac.kr
WINE 14
Evaluation Results
wine.korea.ac.kr
WINE 15
Fingerprints clusters vs. Floor plan rooms
wine.korea.ac.kr
WINE 16
The result is not so much impressive, but the values indicate the Fingerprint generation without site survey is possible
Fingerprint generation needs to be conveyed with human hands, but the required labor for the system is reduced a lot
Notes on High Dimension Fingerprint
wine.korea.ac.kr
WINE 17
RSSI Fingerprint method’s credibility has been widely accepted to be the best method
It shows slightly less accuracy than traditional fingerprint method, but the cost is reduced by much
Conclusion