A Hierarchical Indoor Localization Method Based on Location Clustering and Intra-Class WKNN

Abstract

Fingerprint localization methods based on Wireless Sensor Networks (Wi-Fi) have attracted extensive attention due to the widespread availability of infrastructure. However, traditional approaches are limited by significant signal fluctuations, complex and variable environmental conditions, and multipath effects, making it difficult to further improve localization accuracy. To address these challenges, this paper integrates the K-means clustering algorithm and proposes a region-aware improved Weighted K-Nearest Neighbors (WKNN) hierarchical localization algorithm, achieving a synergistic enhancement of both localization accuracy and time efficiency. Specifically, in the offline phase, the proposed method first clusters reference points based on their positional information at the location level, forming multiple spatially correlated sub-regions, and trains a dedicated WKNN model for each sub-region. In the online localization phase, coarse clustering is first applied to determine the region to which the target point belongs, after which fine-grained sample-level WKNN localization is performed only within that region. This two-stage processing mechanism effectively narrows the search space, reduces the probability of mismatches caused by similar RSSI vectors across different regions, and improves matching efficiency. Experimental results show that compared with the traditional WKNN algorithm, the proposed method reduces localization error by 4.48% and decreases realtime testing time overhead by 76.68%, which fully demonstrates the effectiveness of the proposed algorithm.