A Bayesian Approach to Detect Pedestrian Destination-Sequences from WiFi Signatures
In recent years, interest in crowd dynamics and pedestrian modeling is reviving due to urban growth and its pressure on urban infrastructure. Crowd and pedestrian simulation is emerging as a tool for designing new infrastructures and optimizing the use of current infrastructures. Innovative data collection techniques and realistic experiments are vital in estimating the demand for these infrastructures. Our goal consists in first developing a methodology to collect data about activities, and then model the observed behavior. In most cases, cost and privacy prohibit from installing high precision sensors such as cameras covering an entire pedestrian infrastructure (e.g., airport or a railway station). The large size of an airport or a railway station implies either precise sensors with incomplete coverage (e.g., cameras or bluetooth sensors in intersections), or full coverage with imprecise long range sensors (e.g., cellular network data, traces from WiFi infrastructures). As a result, localization data are either scarce, fuzzy, or both. We propose a methodology to build destinations sequences from scarce data, directly modeling the imprecision in the measure, and using prior knowledge of the infrastructure. We present a case study on a campus, with results of a sensitivity analysis.