Limits...
MagicFinger: 3D Magnetic Fingerprints for Indoor Location.

Carrillo D, Moreno V, Úbeda B, Skarmeta AF - Sensors (Basel) (2015)

Bottom Line: The resulting system does not rely on any infrastructure devices and therefore is easy to manage and deploy.Experimental evaluations carried out in two different buildings confirm the satisfactory performance of indoor location based on magnetic field vectors.These evaluations provided an error of (11.34 m, 4.78 m) in the (x; y) components of the estimated positions in the first building where the experiments were carried out, with a standard deviation of (3.41 m, 4.68 m); and in the second building, an error of (4 m, 2.98 m) with a deviation of (2.64 m, 2.33 m).

View Article: PubMed Central - PubMed

Affiliation: Department of Information and Communications Engineering, University of Murcia, 30100 Murcia, Spain. daniel.carrillo2@um.es.

ABSTRACT
Given the indispensable role of mobile phones in everyday life, phone-centric sensing systems are ideal candidates for ubiquitous observation purposes. This paper presents a novel approach for mobile phone-centric observation applied to indoor location. The approach involves a location fingerprinting methodology that takes advantage of the presence of magnetic field anomalies inside buildings. Unlike existing work on the subject, which uses the intensity of magnetic field for fingerprinting, our approach uses all three components of the measured magnetic field vectors to improve accuracy. By using adequate soft computing techniques, it is possible to adequately balance the constraints of common solutions. The resulting system does not rely on any infrastructure devices and therefore is easy to manage and deploy. The proposed system consists of two phases: the offline phase and the online phase. In the offline phase, magnetic field measurements are taken throughout the building, and 3D maps are generated. Then, during the online phase, the user's location is estimated through the best estimator for each zone of the building. Experimental evaluations carried out in two different buildings confirm the satisfactory performance of indoor location based on magnetic field vectors. These evaluations provided an error of (11.34 m, 4.78 m) in the (x; y) components of the estimated positions in the first building where the experiments were carried out, with a standard deviation of (3.41 m, 4.68 m); and in the second building, an error of (4 m, 2.98 m) with a deviation of (2.64 m, 2.33 m).

No MeSH data available.


Related in: MedlinePlus

Classification results of manually-clustered data in Building A.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4541928&req=5

f9-sensors-15-17168: Classification results of manually-clustered data in Building A.

Mentions: Classification is made using only the magnitude of the magnetic field and the intensities of its three components and performing feature selection on all filtered features. Furthermore, in choosing the best tuning parameters for each algorithm, the package caret [40] for R software was found useful. Once the best tuning parameters for each classifier were obtained through cross-validation, they are evaluated using test data that were used for training. As shown in Figure 6 and also described in the previous section, classification on automatically-clustered data can detect up to five zones with considerable precision if the chosen classifier is the one based on Gaussian processes with feature selection. However, this is not the case for manually-clustered data. To confirm this, the same algorithms are tried with our manually-selected zones (see Figure 9). Because our clustered samples are not optimally related (favoring more accurate estimation), the performance of the classifier is clearly worse. Depending on the configuration of technique + features, no more than two or three zones could be predicted when isolated. Therefore, rearranging clusters manually results in a less general solution.


MagicFinger: 3D Magnetic Fingerprints for Indoor Location.

Carrillo D, Moreno V, Úbeda B, Skarmeta AF - Sensors (Basel) (2015)

Classification results of manually-clustered data in Building A.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4541928&req=5

f9-sensors-15-17168: Classification results of manually-clustered data in Building A.
Mentions: Classification is made using only the magnitude of the magnetic field and the intensities of its three components and performing feature selection on all filtered features. Furthermore, in choosing the best tuning parameters for each algorithm, the package caret [40] for R software was found useful. Once the best tuning parameters for each classifier were obtained through cross-validation, they are evaluated using test data that were used for training. As shown in Figure 6 and also described in the previous section, classification on automatically-clustered data can detect up to five zones with considerable precision if the chosen classifier is the one based on Gaussian processes with feature selection. However, this is not the case for manually-clustered data. To confirm this, the same algorithms are tried with our manually-selected zones (see Figure 9). Because our clustered samples are not optimally related (favoring more accurate estimation), the performance of the classifier is clearly worse. Depending on the configuration of technique + features, no more than two or three zones could be predicted when isolated. Therefore, rearranging clusters manually results in a less general solution.

Bottom Line: The resulting system does not rely on any infrastructure devices and therefore is easy to manage and deploy.Experimental evaluations carried out in two different buildings confirm the satisfactory performance of indoor location based on magnetic field vectors.These evaluations provided an error of (11.34 m, 4.78 m) in the (x; y) components of the estimated positions in the first building where the experiments were carried out, with a standard deviation of (3.41 m, 4.68 m); and in the second building, an error of (4 m, 2.98 m) with a deviation of (2.64 m, 2.33 m).

View Article: PubMed Central - PubMed

Affiliation: Department of Information and Communications Engineering, University of Murcia, 30100 Murcia, Spain. daniel.carrillo2@um.es.

ABSTRACT
Given the indispensable role of mobile phones in everyday life, phone-centric sensing systems are ideal candidates for ubiquitous observation purposes. This paper presents a novel approach for mobile phone-centric observation applied to indoor location. The approach involves a location fingerprinting methodology that takes advantage of the presence of magnetic field anomalies inside buildings. Unlike existing work on the subject, which uses the intensity of magnetic field for fingerprinting, our approach uses all three components of the measured magnetic field vectors to improve accuracy. By using adequate soft computing techniques, it is possible to adequately balance the constraints of common solutions. The resulting system does not rely on any infrastructure devices and therefore is easy to manage and deploy. The proposed system consists of two phases: the offline phase and the online phase. In the offline phase, magnetic field measurements are taken throughout the building, and 3D maps are generated. Then, during the online phase, the user's location is estimated through the best estimator for each zone of the building. Experimental evaluations carried out in two different buildings confirm the satisfactory performance of indoor location based on magnetic field vectors. These evaluations provided an error of (11.34 m, 4.78 m) in the (x; y) components of the estimated positions in the first building where the experiments were carried out, with a standard deviation of (3.41 m, 4.68 m); and in the second building, an error of (4 m, 2.98 m) with a deviation of (2.64 m, 2.33 m).

No MeSH data available.


Related in: MedlinePlus