Pedestrians trajectory prediction in urban environments

Increasing adoption of cellular phones equipped with global positioning system (GPS) chips enables the exploration of pedestrians' mobility patterns. Tasks such as discovering hot-spots in large cities can be addressed through the usage of accumulated GPS coordinates. In this work we utilize spatiotemporal analysis on collected geo-location points to discover Zone of Interests (ZOIs) of pedestrians in large cities to understand people's dynamics. We design an adaptive Markov model to forecast long distance trajectories of pedestrians, which adapts it's behavior constantly by switching from a first or second order Markov chain based on the quality of trace data and users' mobility patterns. From the predicted trajectories, we further introduce a mechanism to predict congested trajectories by estimating the number of pedestrians, who may take the same trajectory in a future moment. We conduct comprehensive empirical experiments using a real-life dataset, namely the Mobile Data Challenge (MDC) dataset with 185 participants. Our mechanisms can deliver a satisfactory pedestrian trajectory prediction with a precision of 86% and a recall of 84%.