Scribd
Upload
15 views18 pages

IPIN Presentation2021

Uploaded by

quezada
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
15 views18 pages

IPIN Presentation2021

Uploaded by

quezada
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 18

Lightweight Wi-Fi Fingerprinting

with a Novel RSS Clustering


Algorithm
Authors: Darwin Quezada-Gaibor, Joaquín Torres-
Sospedra, Jari Nurmi, Yevgeni Koucheryavy and
Joaquín Huerta
Outline

1 Introduction
“Everything is theoretically impossible,
2 Related Work
until it is done.”
3 Algorithm Robert A. Heinlein

4 Experiments and Results


5 Conclusions
1. Introduction
Indoor Technologies and
Number of wearables WW? Techniques
?
~ 1 Billion BLE, UWB, Light …
Wi-Fi
Fingerprinting
Wearable Big dataset
+ Positioning/Localization Low performance
- Resource-constrained Not suitable for real-
devices time App
Require Propousal
Efficient algorithms Lighweight clustering
algorithm
2. Related work
Algorithm based on To join nearest FP Combined WKNN and Combined signal Based on C-Means
k-Medoids based on the physical affinity propagation weighted Euclidean
distance (AP) distance and PL- Optimize the radio
New distance metric map, common area
Combined k-NN and assisted clustering
AP to determine the between FP
k-Means best centroid
Detect outliers and Better performance
well classification Low positioning
Reduction in the than k-Means
Robust method to error and fast time
positioning error form the radio map response
Low positioning error Low positioning
error than WKNN

Tao et al. Xue et al. Cui et al. Wang et al. Ren et al.
3. Algorithm
Cluster 1
AP1 Cluster 3 AP3

Cluster 1 Max. RSS


POS6
AP3
Max. RSS Max. RSS
POS3 POS7
AP1 AP3
Max. RSS
POS20
AP3
RSS based AP2
Cluster 2
AP4
Max. RSS → Same AP POS4
Max. RSS
AP2
Cluster 4

Max. RSS Max. RSS


POS5 POS1
AP2 AP4
Max. RSS Max. RSS
POS8 POS2
AP2 AP4
Max. RSS
POS9
AP2

Centroids with high Join


level of similarity clusters
3. Algorithm
A B
C

Data Initial Join Small


preparation Clusters Clusters

• Powed, positive, exponential • Join FP → Max. RSS • If num. of samples <


values average
• Remove features with zero
values • Centroid • Similarity
• Update the centroid.
3. Algorithm

Input : Dataset, data representation, num. max. RSS.


Output : IDX, centroid (IDX: cluster indexes)
4. Experiments and Results
Configuration Hardware Software

Intel® Core i7-8700T


@ 2.40GHz Windows 10
and 16 GB of RAM Matlab
k-NN as a core IPS
Best configuration
Best k for k-NN
Best parameters:
K-Means, C-Means and DBSCAN Datasets Combinations
(minPts and Eps)
Plain kNN
17 Wi-Fi and 3 BLE open- kNN + K-Means
access datasets.
U. Jaume I, Tampere kNN + C-Means
U., U. of Minho, U. of kNN + DBSCAN
Mannheim and U. of kNN + FPC
Extremadura.
4. Experiments and Results

𝛿= Number of samples in the dataset


𝛾 = Number of feature (Access points)
Num. max. RSS = Number of maximum RSS values
used to join the FP. (1 to 4 max. RSS)
4. Experiments and Results

𝜖3𝐷 : Mean 3D positioning error (m)


𝜖2𝐷 : Mean 3D positioning error (m)
𝜏 : Execution time KNN + Clustering (sec.)

𝜖3𝐷
ǁ : Normalized 3D positioning error (m)
𝜖2𝐷
ǁ : Normalized 3D positioning error (m)
𝜏ǁ : Normalized Execution time KNN + Clustering (sec.)
4. Experiments and Results
kNN + K-Means
𝜖3𝐷
ǁ : 13.8%
𝜖2𝐷
ǁ : 18.8%
𝜏ǁ : 57.2%
kNN + C-Means kNN + FPC
𝜖3𝐷
ǁ : 59.5% 𝜖3𝐷
ǁ : 42.5%
𝜖2𝐷
ǁ : 72.5% 𝜖2𝐷
ǁ : 58.5%
𝜏ǁ : 72.5% 𝜏ǁ < 75.2%
kNN + DBSCAN
𝜖3𝐷
ǁ > ~2.5 times
𝜖2𝐷
ǁ > ~3 times
𝜏ǁ < 18.4 %
4. Experiments and Results

K-Means is better than FPC, K-Means and C-Means in term of positioning accuracy

FPC is better than C-Means and DBSCAN (approx. 58% and 55% -- 3D --)

kNN + K-Means by 84%

FPC is faster than kNN + C-Means by 85%


kNN + DBSCAN by 70%
4. Experiments and Results
Time execution clustering algorithms
104
k-Means
C-Means
DBSCAN
FPC
102
Time (sec)

100

10-2
I1
DS DS
I2 IB1 IB2 N1 N2 T1 001 UT1 UT2 UT3 UT4 UT5 UT6 UT7 XB1 XB2 XB3 UJI1 UJI2
L L MA MA MIN SIM T T T T T T T U E UE UE

FPC is faster in 93% of the cases


4. Experiments and Results

FPC and K-Means have a similar distribution


Imbalanced distribution of samples
5. Conslusions

• FPC is faster than K-Mean, DBSCAN and C-Mean in 93% of the cases.
• Mean positioning error (2D and 3D) lower than DBSCAN and C-Mean
• Suitable for resource-constrained devices

Future Work
• Identify outliers → reduce the positioning error
Thank you for listening!
Bibliography
• Y. Tao and L. Zhao, "A Novel System for WiFi Radio Map Automatic Adaptation and Indoor Positioning," in
IEEE Transactions on Vehicular Technology, vol. 67, no. 11, pp. 10683-10692, Nov. 2018, doi:
10.1109/TVT.2018.2867065.
• W. Xue, X. Hua, Q. Li, W. Qiu and X. Peng, "Improved Clustering Algorithm of Neighboring Reference Points
Based on KNN for Indoor Localization," 2018 Ubiquitous Positioning, Indoor Navigation and Location-Based
Services (UPINLBS), 2018, pp. 1-4, doi: 10.1109/UPINLBS.2018.8559874.
• H. Cui and K. Liu, "Indoor Positioning and Fingerprint Updating Based on Affinity Propagation Clustering,"
2018 Eighth International Conference on Instrumentation & Measurement, Computer, Communication and
Control (IMCCC), 2018, pp. 226-230, doi: 10.1109/IMCCC.2018.00055.
• B. Wang, X. Liu, B. Yu, R. Jia, and X. Gan, " An Improved WiFi Positioning Method Based on Fingerprint
Clustering and Signal Weighted Euclidean Distance". Sensors (Basel, Switzerland), 19(10), 2300.
https://doi.org/10.3390/s19102300.
• J. Ren, Y. Wang, C. Niu, W. Song and S. Huang, "A Novel Clustering Algorithm for Wi-Fi Indoor Positioning," in
IEEE Access, vol. 7, pp. 122428-122434, 2019, doi: 10.1109/ACCESS.2019.2937464.
http://a-wear.eu
http://a-wear.eu

This project has received funding from the European Union’s Horizon 2020 (H2020)
Marie Sklodowska-Curie Innovative Training Networks H2020-MSCA-ITN-2018 call,
under the Grant Agreement no 813278.

You might also like