[Zurück]

S. Wilker, H. Rupkatha, R. Indra:
"Towards an Evaluation Framework for Indoor Localization using Directional Antennas";
Vortrag: Workshop on Computer Science And Information Technologies, Wien (eingeladen); 02.10.2019 - 04.10.2019.

Using Received Signal Strength(RSS) for indoor localization is presently an area of vibrant research activities, and the use of directional antennas seems to hold the promise of improving the performance as well as reduce the effect of disturbances during the localization process. In order to reduce the complexity of the endeavor using directional antennas, several approaches have been adopted so far, out of which, those involving "clustering" are perhaps the most prominent ones. In this paper, we propose an evaluation framework for such clustering algorithms, which can be used to tune the parameters for the clustering process, as well as compare the performance of different clustering algorithms. Using the said framework, we demonstrate the effect of selected clustering algorithms, as well as the hyper-parameters within each algorithm, on the overall localization performance.
The localization procedure we have implemented has two phases- the offline phase and the online phase. In the offline phase, an RSS map is created by measuring the RSS at predetermined points, followed by classification of the RSS map into regions of different ranges of signal strength. Points at each of these regions are then clustered based on their spatial coordinates as well as RSS value, in order to reduce the search space. In the online phase, the target point is first assigned a class based on its RSS fingerprint. A position estimation algorithm runs on the reduced search space defined by the cluster centers of the required class. The evaluation framework is developed in order to make the overall process more efficient and increase position estimation accuracy. The hyper-parameters of the clustering algorithms are tuned to improve the quality of the clustering algorithm using internal metrics, as well as to minimize the average position estimation error obtained from a random sampling of known target points, by optimizing external metrics. Comparison of the selected clustering algorithms being used during the procedure is another aspect of the evaluation framework.
We improved the process of RSS based indoor localization using directional antennas and the observations obtained from the proposed evaluation model.

https://publik.tuwien.ac.at/files/publik_285206.pdf