Real-Time Trajectory Publishing in Mobile Crowdsensing Under Local Differential Privacy

Mobile crowdsensing (MCS) leverages the sensing capabilities of mobile devices carried by a vast number of participants to accomplish large-scale urban sensing tasks. In MCS systems, participants continuously generate trajectory streams while uploading sensing data, which raises severe risks of location privacy leakage. Local differential privacy (LDP) provides strong guarantees for user-side privacy protection. However, existing approaches for real-time trajectory publishing in MCS remain inadequate; they fail to fully leverage the continuously evolving spatial density and temporal structures within trajectory streams, which leads to suboptimal utility and distorted spatiotemporal representations under strict privacy constraints. To address this issue, we propose AdaGrid-Syn, a two-layer adaptive grid framework for real-time trajectory publishing under LDP. AdaGrid-Syn dynamically adjusts spatial granularity based on changes in density distributions and transition patterns. In addition, we develop a new adaptive privacy budget allocation strategy that distributes privacy budgets according to the temporal variation of Markov transition matrices, enabling better capture of time-sensitive behavioral characteristics. Experimental results demonstrate that our framework significantly outperforms existing approaches.