WiNCDN: A Novel WiFi-Assisted Non-Cooperative Indoor Localization System via Dropout-Based Neural Network

Abstract

The increasing demand for indoor location-based services (LBS) and the widespread use of mobile devices have spurred the development of accurate indoor localization systems for smartphones. Implementing these systems requires real-time responsiveness, non-cooperative communication with targets, and reliable localization capabilities. However, conventional active and passive localization systems face challenges, such as the necessity of target cooperation or latency caused by low measurement frequency. Additionally, the uncertainty in localization evaluation is often overlooked. In response to these issues, we present a WiFi-assisted non-cooperative fingerprint localization system (WiNCDN) that utilizes a dropout-based neural network. WiNCDN quickly gathers a large number of responses from WiFi devices by employing a hiding mechanism in the 802.11 protocol. Moreover, the system can make informed decisions based on localization confidence. Our findings demonstrate that WiNCDN enables real-time target tracking in various scenarios, including line-of-sight (LoS) and non-line-of-sight (NLoS) situations. Compared to existing methods, the results indicate that WiNCDN achieves a better balance between accuracy and robustness.