Passive radar based on Wi-Fi signals for crowd monitoring
This thesis aims at performing crowd monitoring by exploiting signals of opportunity from existing Wi-Fi access points to build a passive radar. Radar processing will be applied on signals that bounced on people (targets) in an environment to obtain a so-called range-Doppler map (RDM) containing the distance and speed information of the targets, as well as an angle spectrum (from MUSIC method) containing the angle-of-arrival of the signals that bounced on targets. Experiments will be performed with human targets using Universal Software Radio Peripherals (USRPs) to assess the feasibility of using the latest Wi-Fi standards (802.11ac, 802.11ax) for passive radar application, and the feasibility of building a multi-antenna passive radar with high bandwidth. Target tracking algorithm will be applied on measurements from the USRPs setup to track the target movements over time in cartesian coordinates in the environment. Since the relation between the measurements and the target tracking state in cartesian coordinates involves non-linear equations, the Unscented Kalman Filter (UKF) will be investigated in particular to tackle those non-linearities. Single-target tracking will first be considered, and then extended to multi-target tracking and group tracking. For the group tracking, clustering algorithms such as Density-based spatial clustering of applications with noise (DBSCAN) will be applied to combine the signal responses of several targets being close from each other in one single centroid in order to track that centroid only. Data association techniques such as the Probabilistic Data Association Filter (PDAF), the Joint-PDAF and possibly Multiple Hypothesis Tracking (MHT) will be investigated to link targets signals response with current tracks of the UKF. Machine learning-based algorithms will also be developped to count the number of people in a group based on the radar RDM and angle spectrum, by exploiting the high bandwidth provided by the latest Wi-Fi standard as well as the signals from multiple antennas at the receiver. This will include the study of Bayes classifiers, Support Vector Machines (SVM) and Neural Networks. The output of those algorithms will then be combined with the group tracking algorithms in order to simultaneously track groups in space and know the number of people in those groups.