1 Introduction
2 Background
2.1 Fingerprint-based localization technique
2.1.1 Training phase
2.1.2 Localization phase
2.2 GSM RSS fingerprint
Band | ARFCN |
---|---|
Extended GSM-900 | 0–124, 975–1023 |
GSM-1800 | 512–885 |
2.3 Desired localization accuracy
3 Experimental setup
3.1 Experimental sites
3.2 Data acquisition and processing device
3.3 Data collection schemes
3.3.1 Space sampling scheme
3.3.2 Time sampling scheme
3.4 Datasets
4 Methodology
4.1 Classification algorithms
4.1.1 Pairwise classifier
Training data window (days) | Training examples | Total training time (min) |
---|---|---|
1 | 23,871 | 16.5 |
2 | 38,556 | 28.7 |
8 | 91,044 | 72.7 |
4.1.2 Decision rules for multiclass discrimination
4.2 Bayesian filter
4.2.1 Recursive Bayesian filtering
4.2.2 Prior information
4.2.3 Observation model
4.2.4 The filtering procedure
-
Making room predictions from the previous location estimation based on the node-path room layout model and their probabilities P(x k |x k − 1) from relation (9).
-
Inputting the current fingerprint measurement into SVM classifiers and obtaining the room label, which is used to calculate the observation probabilities P(y k |x k ) based on relation (10).
-
The filtering location output of a current fingerprint measurement is by updating the predictions based on observation probabilities using (7).
5 Results
5.1 Results of SVM classification
True room | SVM output 1 (%) | SVM output 2 (%) | SVM output 3 (%) | SVM output 4 (%) | SVM output 5 (%) | SVM output 6 (%) | SVM output 7 (%) |
---|---|---|---|---|---|---|---|
1 | 85.4 | 9.1 | 4.4 | 0 | 0.1 | 0.1 | 0.9 |
2 | 11.5 | 71.4 | 16.3 | 0.1 | 0.2 | 0.3 | 0.2 |
3 | 3.4 | 17.1 | 78.8 | 0.2 | 0.2 | 0.2 | 0.1 |
4 | 0 | 0 | 0.4 | 88.2 | 7.9 | 1.9 | 1.6 |
5 | 0 | 0 | 0.2 | 8.3 | 80 | 1.6 | 9.9 |
6 | 0.1 | 0.1 | 0.6 | 0.8 | 1.8 | 93 | 3.6 |
7 | 1.2 | 0.1 | 0.3 | 1.4 | 13.3 | 4.3 | 79.4 |
5.2 Results of Bayesian filtering
Test dataset | Accuracy without Bayesian filtering (%) | Accuracy with Bayesian filtering (%) |
---|---|---|
December 24, 2012 | 74.3 | 98.8 |
December 29, 2012 | 70.6 | 99.1 |
December 30, 2012 | 72.4 | 99.4 |
January 2, 2013 | 77.1 | 99.6 |
January 3, 2013 | 75.9 | 98.9 |
January 4, 2013 | 70.0 | 98.4 |
January 6, 2013 | 83.2 | 99.7 |
January 7, 2013 | 68.3 | 97.1 |
February 3, 2013 | 77.7 | 99.5 |