A new device-free localization method for RSS data with considering correlations

Abstract

Device-free localization (DFL) is an emerging intelligent sensing technique with challenging issues due to the highly correlated data. Localization performance has been demonstrated from algorithmic as well as modelling perspectives, for example, by including various sparsity constraints. The general sparse regularization method simultaneously estimates parameters and selects features, but its performance relies heavily on the correlations among features, which leads to generalization errors. Our goal in this study is to formulate a DFL method based on independently interpretable sparse coding to improve the accuracy and efficiency of localization. The scheme uses a novel correlation constraint of sensing matrix columns to handle the high correlation of the received signal strength (RSS) data, overcoming not only the fluctuation of RSS due to environmental changes but also avoiding overfitting through identification of informative variables that provide interpretable results. Meanwhile, the minimax concave penalty (MCP) is proposed as a sparse regularizer to obtain strong sparsity and an exact estimate. We propose using the difference of convex functions algorithm (DCA) and non-convex proximal splitting to address the non-convexity problem. Extensive experiments were performed in a cluttered indoor laboratory and open outdoor environment to verify the efficacy of the proposed method.