Toward a Dynamic K in K-Nearest Neighbor Fingerprint Indoor Positioning

K-nearest neighbor (KNN) fingerprint positioning is a promising solution for WLAN-based indoor positioning that has received much attention over the past ten years. In order to achieve good positioning results, much effort has been made to develop advanced KNN algorithms and to find the optimal K. So far most of the work has concentrated on using a fixed K for a given positioning system. A drawback is that the positioning system would become unstable since the best K at one place may not be the best for another. In order to address this problem, we propose a dynamic KNN algorithm that can adjust the value of K dynamically to offset noises of different levels. The value adjustment is made based on the WiFi signals measured in real-time, therefore, it does not require any prior knowledge of the WLAN or the indoor environment. Analysis on field measurement data shows that, for a large percentage of the positioning area, the best K is 1 instead of 3 or 5 found in previous studies. The field experiment also shows, by setting K=1, the proposed method can achieve better positioning accuracy compared with the classical KNN method.