Improvement of RSS-based indoor positioning systems by enhancing location sensing algorithms

Advances in mobile technologies have changed the way users interact with devices and other users. These new interaction methods and services are offered by the help of intelligent sensing capabilities, utilizing context, location and motion sensors. However, indoor location sensing is mostly achieved by measuring radio signal (WiFi, Bluetooth, GSM etc.) strength and nearest neighbor identification. The most common algorithm adopted for Received Signal Strength (RSS)-based location sensing is K Nearest Neighbor (KNN), which calculates K nearest neighboring points to estimate location. Nevertheless, fluctuation on the received signal strength is one of the crucial problems in the RSS-based KNN algorithm. Adopting the fluctuated signals for positioning may lead to inaccurate results. In this study, the accuracy of the RSS-based KNN algorithm is targeted to be enhanced by eliminating the effects of fluctuated signals. For this purpose three separate enhancements are applied to the KNN algorithm. In the first proposed system, the accuracy of the KNN algorithm is attempted to be incremented by exploiting wireless mesh network capabilities. This approach is to share the location data among devices and utilize them into the location estimation. However, the results showed that the proposed systems could not provide the intended accuracy improvement. The second proposed system aims to apply k-means clustering to improve the KNN algorithm by enhancing the neighboring point selection. In the proposed method, k-means clustering algorithm groups nearest neighbors according to their distance to mobile user. The evaluation results indicate that the performance of clustered KNN is closely tied to the number of clusters, number of neighbors to be clustered and the initiation of the center points in k-mean algorithm. The third system aims to improve the KNN algorithm by integrating a short term memory (STM) where past signal strength readings are stored. In this proposed approach, the signal strength readings are refined with the historical data prior to comparison with the environment's radio map. The results indicate that the performance of enhanced KNN-STM outperforms the KNN algorithm. Moreover, as an application, the proposed location sensing system is utilized in a location-aware system that accesses patient records.