Online Spatiotemporal Modeling for Robust and Lightweight Device-Free Localization in Nonstationary Environments

Recent advances in WiFi-based device-free localization (DFL) mainly focus on stationary scenarios and ignore the environmental dynamics, hindering the large-scale implementation of the DFL technique. In order to enhance the localization performance in nonstationary environments, in this article, a novel multidomain collaborative extreme learning machine (MC-ELM)-based DFL framework is proposed. Specifically, the whole environment is first divided into several subdomains depending on the distributions of the collected data using a clustering algorithm, and a corresponding number of local DFL models are then built to represent these subdomains separately. Finally, a global DFL model is achieved by seamlessly integrating all the local DFL models in a global optimization manner. The created MC-ELM-based DFL model also can be incrementally updated with sequentially coming data without retraining to track the environmental dynamics. Extensive experiments in several indoor environments demonstrate the robustness and generalization of the proposed MC-ELM-based DFL framework.

I. Introduction

Due to the rapid development of wireless communication techniques and ubiquitous deployment of Internet-of-Things (IoT) infrastructures, location-based services are powering and reshaping our daily life [1], [2], [3]. Prior studies have shown that the existence of the target posing influence on surrounding wireless signals can be utilized to estimate target's location [4], [5], [6]. Sequentially, WiFi-based device-free localization (DFL) techniques were proposed and have attracted great attention in both academia and industry, due to the ubiquitous, low-cost, and nonintrusive characteristics of WiFi signals [7], [8], [9]. Different from device-based and computer vision-based localization techniques, WiFi-based DFL does not need the target to equip with dedicated electronic devices and the target's location can be estimated passively, and has enough sensing coverage and privacy protection capability without lighting requirements, enabling it a promising candidate technique for several intelligent IoT applications [10], [11], [12], [13]. Early WiFi-based DFL methods mainly utilize received signal strength (RSS), but the localization performance of such set of methods is seriously limited due to the unpredictable fluctuations of RSS. In recent years, channel state information (CSI) can be retrieved from network interface cards (NIC) with CSITOOL [14], and CSI has the better characterization of the frequency response of the wireless channel at orthogonal frequency-division multiplexing (OFDM) subcarriers, which can enhance the localization performance of WiFi-based DFL.