Spatial-Temporal Gaussian Visit Model for Mobile Crowd Sensing

Jixuan Zhou, Mingliu Liu, Deshi Li

Abstract: Currently, mobile crowdsensing (MCS) is a promising solution for urban information collection. Generally, whether the participants can arrive at the specific location in time is the foundation of recruitment; hence exploring spatial-temporal restrictions according to their mobility are essential in MCS. However, the evaluation of spatial-temporal fitness between participants and sensing tasks is quite challenging due to laborious granularity control. To this end, we first propose a spatial-temporal Gaussian model to characterize visiting distribution in this paper. Specifically, the spatial and temporal boundaries can be expressed through the probability with two Gaussian random variables, which do not need to define the granularity in advance. Accordingly, to describe the participant's daily visiting behavior, the proposed model can be extended to a mixed spatial-temporal Gaussian formation. Particularly, in order to extract the participant's daily visiting behavior efficiently, a two-step clustering method is developed for analyzing the historical trajectory. Finally, the tasks can be assigned to participants intelligently without disturbing their behavior patterns. Experimental results show that the proposed spatial-temporal Gaussian visit model can well adapt to MCS tasks with different spatial and temporal granularities.