Cooperative Target Tracking and Signal Propagation Learning Using Mobile Sensors

Abstract

Target tracking refers to positioning mobile objects over time. The targets may be hospital patients, park visitors, mall shoppers, warehouse assets, etc. We consider a novel cooperative system to track targets, where a target carries low-cost RF tag which not only beacons its ID, but also receives and rebroadcasts beacons of tags within a certain hop away. Mobile sensors, equipped with localization and communication modules, are used to capture and forward the beacons to a server to track the targets. Such multi-hop approach greatly extends the sensing range of the mobile sensors, or equivalently, the beaconing range of the tags, leading to cost-effective deployment. We propose Mosent, a highly accurate multi-hop system using mobile sensors for target tracking. To account for complex signal propagation in different indoor and outdoor environment, we represent the received signal strength (RSS) matrix overcoming the assumption on propagation model. Given sensor locations, beacons detected by the sensors and RSS matrix, Mosent jointly considers temporal and spatial information to track targets using a modified particle filter. Mosent has an optional, independent and offline module to learn spatial signal propagation in terms of RSS matrix using cooperative mobile sensors equipped with beaconing transceivers. We have implemented Mosent and conducted extensive experiments. Our results show that Mosent achieves 4.37m and 9.46m tracking error in the campus and the shopping mall, respectively, which outperforms other state-of-the-art approaches with significantly lower tracking error (often by more than 30%).