RSSI-based Hybrid Centroid-K-Nearest Neighbors localization method

Various ambient intelligent environment applications are based on location-based services. To obtain the location, many popular localization methods use the Received-Signal-Strength-Indicator (RSSI) measurement, as it can be obtained from almost any wireless communication device technology and does not require additional hardware. However, physical phenomena, such as reflection and diffraction, affect the radio signal propagation and lead to imperfections in the RSSI measurements, impacting the localization performance. In this paper, the K-Nearest Neighbors (KNN) machine learning technique is used to improve the localization accuracy of the RSSI-based geometric localization methods. First, in the training step, we proposed to model the RSSI measurement by a distance interval that accounts for the RSSI imperfections due to the signal propagation noise in the environment of interest. Then, in the online step, a Hybrid Centroid-KNN (HCK) localization method based on the defined distance intervals is proposed to calculate the position of the target node. To validate the performance of the proposed localization method, we used three datasets from three testbeds which are a residence equipped with WiFi technology, a library, and office space, both equipped with BLE technology. The obtained results show that the proposed method significantly reduces the localization error in these environments in comparison with the well-known geometric localization methods Multilateration (ML), Min–Max (MM), and Weighted Centroid Localization (WCL). In particular, the average localization errors of the HCK method are reduced compared to the ML, MM, and WCL methods in the residence by 50%, 37%, and 14%, respectively, when the inhabitant is performing daily activities and by 44%, 31%, and 14%, respectively, when the inhabitant is walking.