Fingerprint-based Indoor Localization using Weighted K-Nearest Neighbor and Weighted Signal Intensity

In the past few decades, wireless indoor positioning systems, especially signal strength fingerprint technology, have become the subject of major research efforts. However, most proposed solutions require an expensive site survey to build a radio map, which can be used to match the radio signature to a specific location. In this study, we proposed a novel fingerprint-based indoor localization using weighted K-nearest neighbor and weighted signal strength, named WKNNS. We adjust the weight of samples by the strength of the signal: reduce the influence of strong signal samples, and increase the influence of weak signal samples. First, the strong signal samples were divided into multiple clusters by dividing. Then, the weak signal samples are divided into those clusters. Thus, multi-sample classification can be turned into a binary classification problem. This algorithm was applied to indoor positioning and obtained better accuracy. Compared with the traditional KNN and Bayesian algorithms, we found that the positioning accuracy of WKNNS after region division is higher than that of Bayesian algorithm. The cumulative error probability distribution of the WKNNS algorithms optimized by the reader is also higher than Bayesian algorithm. The positioning accuracy of the WKNNS algorithm based on the fingerprint conversion model is higher than that based on the signal strength fingerprint.