A Weighted Algorithm Based on Physical Distance and Cosine Similarity for Indoor Localization

Abstract

The weighted K-nearest neighbor (WKNN) is a main algorithm based on fingerprint for indoor localization. The weight is usually calculated by the inverse of the received signal strength indication (RSSI) distance between reference point and the test point, which does not take the exponential relationship between RSSI and physical distance into account. Some weighted algorithms (called physical distance algorithms), while considering their relationship, do not consider the physical distance of the reference point from the test point and the characteristics of the RSSI. Since the range of RSSIs received from different Bluetooth transmitting devices (Beacons) in the same area is different, the cosine similarity difference is extremely small, and the influence of the reference point on the positioning result cannot be fully expressed. Therefore, in order to improve the positioning accuracy, this paper proposes a method of normalizing RSSI and a new weighted algorithm based on the physical distance and on the cosine similarity of the processed RSSI. Experiments in the actual environment show that the proposed algorithm has better positioning accuracy, and the average positioning accuracy is 1.816m, which is 36.41% higher than NN, 14.54% higher than KNN, 12.27% higher than wKNN and 12.78% higher than the physical distance algorithm.