RWKNN: A Modified WKNN Algorithm Specific for the Indoor Localization Problem

This paper proposes a restricted weighted
${k}$
-nearest neighbor algorithm (RWKNN) specific for indoor environments. The traditional WKNN method determines the locations by calculating the difference between the current received signal strength (RSS) and fingerprint RSS. However, the limitations of the traditional WKNN positioning method include RSS instability and spatial ambiguity. With this focus, the proposed RWKNN considers indoor moving constraints and uses searching rectangular and trajectory restriction to reduce spatial ambiguity. In addition, to mitigate the effects of RSS instability on the iteration-based method, a confidence number is introduced. Through simulation of office environments, field experiments, and verification based on public datasets, numerical results show the superiority and effectiveness of the proposed RWKNN over other constraint-based algorithms in terms of robustness and accuracy. Specifically, RWKNN outperformed the traditional WKNN method by 43% and 20% in the simulation and field experiment tests respectively, and by 22% on the Tampere open dataset.