Crowdsensing is a promising sensing paradigm to efficiently collect and monitor the physical world by using the embedded sensors in mobile devices. However, the observations (sensory data) submitted by mobile device users may not be reliable. For the same sensing task, users with different reliabilities may submit conflicting information. Thus, we need to estimate the truth based on the submitted observations. Temporal and spatial correlations among tasks are widely observed in crowdsensing applications. However, most of the existing truth discovery mechanisms assume that the tasks are independent, which is not suitable for all crowdsensing applications. To solve this problem, we propose an efficient truth discovery mechanism for crowdsensing tasks with temporal and spatial correlations. To improve the reliability of the estimated truth, we first filter the outliers based on the temporal correlations among tasks, then estimate the truth based on the weighted observations, and finally refine the estimated truth based on the spatial correlations among tasks. Our experiments on a real transportation dataset show the efficiency of the proposed mechanism.
展开▼
