An experimental investigation of indoor localization by unsupervised Wi-Fi signal clustering

Indoor localization is an important research challenge for context aware computing. Although various solutions and products are developed based on Wi-Fi received signal strength indicator (RSSI) fingerprinting-based indoor localization, the construction of a complete and high quality database of fingerprints is a challenge, because in particular human involvement is usually required in such methods. We have developed the unsupervised Density-based Clustering Combined Localization Algorithm (DCCLA) to address this problem. The fingerprints of meaningful locations in people's daily lives can be automatically learned without requiring explicit data labelling from users. In this paper an extensive experimental investigation of the performance of DCCLA is presented. These experiments focus on Wi-Fi signal propagation conditions, including the distance to an Access Point (AP), the damping of the signal, and Fast Fading. The parameters used in DCCLA are determined based on these performance analysis and evaluation.