ML-MultiLoc: Device-Free Passive Multi-target Indoor Localization Using Multi-label Learning

Zhiliang Yang, Weiqing Huang, Siye Wang, Shuai Wang, Di Gao

Abstract: With the rapid development of the Internet of Things (IoT) and artificial intelligence (AI), indoor localization technology has become increasingly important in areas such as smart warehousing and security management. However, multi-target localization still faces challenges such as multipath interference and insufficient localization accuracy. In this paper, we propose the ML-MultiLoc, a device-free passive multi-target indoor localization system based on multi-label learning. Utilizing the RFID antenna array, ML-MultiLoc is capable of collecting signals from multiple angles, obtaining rich spatial features. In addition, we introduce a lightweight improvement based on the ResNet and use a dual-branch structure to extract fused fingerprint features from received signal strength indication (RSSI) and phase data. By incorporating a multi-label learning framework, we transform the multi-target localization problem into a multi-label classification task, thereby improving the system’s ability to distinguish multiple targets. Experimental results show that ML-MultiLoc achieves localization accuracies of 96.55%, 96.14%, and 96.46% in single-target, dual-target, and triple-target environments, respectively, outperforming existing RFID multi-target localization methods.

External IDs:dblp:conf/icic/YangHWWG25