Design and implementation of an occupancy monitoring method for indoor public sensing applications

Abstract

Recently, new architectures of the modern smart phones come with integrated powerful platform sensors. Moreover, indoor environments are typically rich with such smart devices. By considering these advantages, the mobile phones can be exploited for collecting measurements through a public sensing (PS) system. As a motivation for the mobile users to participate in the data sensing process, we should restrict the energy drawn by the sensing tasks. Several energy consumers may negatively affect the participating mobile devices while collecting measurements. Among these consumers are: (1) data acquisition, (2) measurements uploading, (3) overhearing superfluous sensing queries, and (4) position fixing and updating. Actually, a significant amount of energy is dissipated, on the mobile devices, due to updating the PS server with the current position. Several research work has devoted to reducing this overhead through either piggybacking the update messages or through reducing the number of position updates. However, there exists still an overhead on the mobile devices due to estimating the position and reporting it to the corresponding PS server. In this thesis, we consider an alternative approach to overcome this energy overhead. We propose to opportunistically exploit the already-existent WiFi traffic to monitor the occupancy of a certain area by the participating mobile devices. Accordingly, we move the burden of estimating the position and reporting it to a PS server, from the mobile devices to the PS servers. At the outset, we investigate the applicability of our proposed method through experimentally studying the WiFi traffic. The traffic analysis confirmed that plenty of WiFi messages are exchanged between mobile devices and the WiFi access points (AP). Based on these findings, the proposed occupancy monitoring method is divided into two scenarios: (1) multiple APs localization scenario; which is applied when there are enough uplink traffic from the mobile devices, hence the user’s location can be estimated using RSSI from multiple APs; (2) a single AP localization scenario when there is no enough WiFi traffic but the server still receives the RSSI measurement of mobile device from the associated AP. Specifically, two algorithms constitutes our proposed algorithm, namely device detection and data collection algorithm and position estimation algorithm. The former runs at the AP level to detect the existing devices in their coverage area whereas the latter algorithm runs on the server side to gather the RSSI measurement and to calculate the current user location. We adopt the fingerprinting strategy as our position fixing method. Generally, fingerprinting has two phases including an offline and an online phase. For the offline phase, we adopt affinity propagation to cluster the collected RSSI measurements. For the online phase, we provide a comparative study between adopting the K-nearest neighboring algorithm and the compressive sensing algorithm. For evaluating the performance, we construct a testbed and several experiments have been carried out. The findings show a localization accuracy of circa two meters which is achieved via adopting compressive sensing.