WiFi-Based Device-Free Passive Multi-Targets Localization Using Multi-Label Learning

In this letter, we propose MTFLoc, a novel device-free passive fingerprinting localization system based on CSI, leveraging multi-label learning to overcome challenges in multi-target localization. The MTFLoc system divides the localization area into multiple training point areas, each representing a unique class or label. This formulation transforms the multi-target localization problem into a multi-label classification problem. It is crucial to obtain high richness and resolution fingerprint features. MTFLoc uses novel data pre-processing techniques to obtain high-resolution fused multi-target fingerprint features (FMTF) from CSI amplitude and phase information, improving localization accuracy and quality. Finally, the FMTFs are inputted into a deep learning-based multi-label classification framework for parameter training and location estimation. Experimental results clearly demonstrate the outstanding performance of MTFLoc compared to existing multi-target localization approaches.