A Brownian Motion Restricted K-Nearest Neighbor Algorithm for Indoor Positioning

In the indoor positioning method based on traditional KNN, the Received Signal Strength Indicator (RSSI) is commonly utilized as fingerprint information for measuring similarity, with the selection of the most matching K reference points (RPs) for positioning. However, ensuring the accuracy of the KNN fingerprint positioning method requires the collection of a substantial amount of fingerprint information and is susceptible to the complexity and stability of the indoor environment. Consequently, we propose a novel algorithm called Brownian Motion Restricted K-Nearest Neighbor (BMR-KNN). In the BMR-KNN method, we leverage the assumption that the tester's activity exhibits a degree of adherence to the principles of Brownian motion. We utilize this assumption as prior knowledge to correct the results obtained from the KNN positioning algorithm based on RSSI. Furthermore, we propose a dynamic K value allocation algorithm (DKAA) for automatic optimization of the K value within the KNN positioning algorithm. Despite utilizing the previous location and time information, BMR-KNN achieves real-time positioning without requiring knowledge of the user's exact moving speed and direction. Experimental evaluations conducted on two public datasets demonstrate that the new algorithm outperforms other advanced methods, including the optimal traditional KNN, and reduces the average positioning error to 3.31 m to the greatest extent.