Enhanced WiFi CSI fingerprints for device-free localization with deep learning representations

With the increasing demands for location-based services in the indoor environments and the widespread deployment of WiFi devices, WiFi-based indoor localization techniques have attracted more attention, especially the device-free localization (DFL), which could estimate the location of the target without attaching any dedicated electronic devices. However, existing fingerprint-based DFL methods face the fingerprint similarity problem, which happens when some fingerprints corresponding to different locations are very similar and may result in the ambiguity in online location estimation phase where we cannot match the fingerprint of the target with the correct one, and severely degrades the localization performance. To address this problem, this article proposes a novel DFL method, in which the original fingerprints are replaced by the hidden layer parameters of the deep neural network. Specifically, the extracted raw data are first transformed into a discriminable feature space using the weights of multilayer extreme learning machine (ML-ELM). Next, we take the generated features as the fingerprints to build an ELM-based DFL model. Finally, we implement a prototype system and evaluate its performance in several indoor environments. Experimental results demonstrate the accuracy and robustness of the proposed method.