Practical evaluation of a crowdsourcing indoor localization system using hidden Markov Models

The extensive deployment of wireless infrastructure provides a low-cost approach to tracking of mobile phone users in indoor environments using received signal strength (RSS). Crowdsourcing has been promoted as an efficient way to reduce the labor-intensive site survey process in conventional fingerprint-based localization systems. Despite its stated advantages, use of crowdwourcing for localization has issues of accuracy and reliability in indoor applications, in large part because of multipath propagation. This paper discusses and evaluates a Bayesian approach to localization of mobile users based on a crowdsourced fingerprint, in which environmental constraints as well as dynamic property of the mobile are incorporated as priors. Both a Markov chain and a semi-Markov chain approach are applied for modeling the transition and duration statistics of the mobile user across different location cells. Field test results demonstrate the effectiveness of introducing this additional information for localization in a real-world wireless network.