ROLIPS: A Robust and Lightweight Wi-Fi CSI Indoor Positioning System

Abstract

Accurate and reliable indoor localization is essential for location-based services (LBSs). Wi-Fi channel state information (CSI)-based indoor localization has attracted increasing attention due to its low deployment cost and wide availability. However, CSI fingerprints are easily affected by environmental dynamics, hardware instability, multipath propagation, and nonline-of-sight (NLOS) conditions, resulting in unstable fingerprints, degraded localization accuracy, and an increased computational burden. To address these problems, this article proposes ROLIPS, a robust and lightweight CSI-based indoor localization system for underground parking garage environments. ROLIPS suppresses abnormal CSI amplitude fluctuations using Hampel-based outlier filtering and residual-adaptive Kalman smoothing. It further reduces packet redundancy and feature dimensionality through improved, adaptive, weighted K-means clustering and principal component analysis (PCA)-based feature reduction. Finally, an improved Bayesian, probability-weighted, localization fusion strategy is developed to combine different localization estimates and improve positioning robustness. Experiments were conducted in real underground parking garage scenarios, including both line-of-sight (LOS)-dominant and NLOS environments. Under NLOS conditions, ROLIPS achieves an average positioning error (APE) of 1.39 m, a root mean square error (RMSE) of 1.59 m, and a standard deviation (STD) of 0.77 m, outperforming the compared baseline methods. Under the default two-access-point (AP) configuration, online localization requires only five CSI packets from each AP and achieves a total latency of approximately 0.94 s, demonstrating a practical tradeoff among localization accuracy, robustness, and response latency.