Addressing Temporal RSSI Fluctuation in Passive Wi-Fi-Based Outdoor Localization

Passive outdoor localization is valuable in understanding pedestrian mobility patterns for city planning and other purposes. While there are various approaches to Wi-Fi localization, few demonstrate robustness to temporal fluctuations in RSSI measurements. In this paper, we continue the work on the Mobility Intelligence System (MobIntel) and address RSSI temporal fluctuation via two approaches. First, given a rectangular testbed bounded by four sensors, we reformulate the problem from localizing the emitters from anywhere within the testbed to classifying whether they originate from the North or South strip and estimating their East-West position on the identified strip. This change simplifies the problem, reduces the impact of RSSI fluctuations, and better resembles how pedestrians move in city streets. Second, we present a multi-stage model (InXModel) to perform localization based on the reformulated problem. Within this model, we compare the performance of four data transformation methods that further dampen the effect of RSSI fluctuation—basic standardization (STD), Kernel Principal Component Analysis (KPCA), Transfer Component Analysis (TCA), and Semi-Supervised TCA (SSTCA). Data resistant to improvements via transformation are discarded to preserve model performance. We observe that the InXModel with STD is the fastest and most accurate model, achieving a localization error $\le 4$ m in 94.3% of the cases with a 30.2% data discard rate. Finally, we compare the InXModel with a recently published method (EMDT-WKNN) specialized for handling RSSI temporal fluctuations and find that the former outperforms the latter. We discuss the causes of the difference in performance.