A Learning-Based Inversion Strategy for Passive Wireless Detection of Crowds

Passive indoor localization is an emerging technology for the detection and tracking of transceiver-free entities using wireless infrastructures. Customized wireless sensors are often used for the acquisition of suitable wireless signals. In this work, neither hardware customization nor dedicated deployment are introduced. The wireless signal transmitted by standard Wi-Fi access points is processed to solve an inverse problem for the real-time detection of crowds. Towards this end, a wavelet decomposition of the acquired data is combined with a learning-by-example (LBE) strategy in order to learn the complex relation between crowd presence and signal perturbations. Experimental results show the capabilities of the proposed solution in detecting people within a real Wi-Fi enabled test site. The obtained performance points out that a standard Wi-Fi network can be profitably adopted as a low-cost, and scalable solution to support crowd management in many applicative fields.