Introduction
The rapid advancement of mobile sensing technologies and the proliferation of smart connected vehicles have made vehicular crowdsensing (VCS) a promising solution for large-scale data collection in urban environments. Unlike traditional mobile crowdsensing (MCS), VCS leverages vehicles equipped with advanced sensors, reliable communication capabilities, and substantial computational power, making them ideal for diverse sensing applications such as traffic monitoring, smart parking, and air quality assessment. However, while VCS offers significant benefits in terms of efficiency and coverage, it also introduces critical privacy challenges, particularly concerning the exposure of users’ sensitive temporal data during task allocation.
Most existing privacy-preserving mechanisms in VCS focus primarily on protecting location-based information while neglecting the risks associated with temporal sensing inferences. Attackers can exploit users’ actual dwell times in specific sensing regions to deduce sensitive behavioral patterns, daily routines, and movement trajectories. To address this gap, this paper introduces a novel privacy-preserving task allocation scheme that ensures the indistinguishability of sensing times while maintaining efficient task distribution among participating vehicles.
System Architecture and Problem Formulation
The proposed VCS system consists of two primary entities: the cloud platform and vehicle participants. The cloud platform acts as a centralized coordinator responsible for collecting historical call records, processing dwell time data, and assigning sensing tasks to vehicles. Participants, on the other hand, are vehicles willing to contribute sensing data. They communicate with the cloud platform, receive assigned tasks, and upload obfuscated dwell time data after task execution.
The core challenge lies in balancing privacy protection with task allocation effectiveness. Traditional approaches often fail to account for the privacy risks associated with revealing precise dwell times, which can be exploited by adversaries. The proposed solution tackles this by incorporating differential privacy techniques to obscure real dwell times while ensuring that task allocation remains feasible under temporal and spatial constraints.
Privacy Protection Mechanism
To safeguard users’ temporal privacy, the scheme employs differential privacy—a robust mathematical framework that guarantees privacy by injecting controlled noise into sensitive data. Specifically, Laplace noise is added to the actual dwell times reported by vehicles before they are transmitted to the cloud platform. The amount of noise is governed by a privacy budget parameter, which determines the trade-off between privacy strength and data utility. A smaller privacy budget results in higher noise injection, enhancing privacy but potentially reducing task allocation accuracy.
The key innovation lies in transforming the problem of verifying whether a vehicle’s dwell time meets task requirements into a probabilistic assessment. Instead of relying on exact dwell times, the platform computes the probability that a vehicle’s noise-obfuscated dwell time exceeds the minimum sensing duration required for a task. This probabilistic approach ensures that attackers cannot reverse-engineer the original dwell times while still allowing the platform to make informed task allocation decisions.
Task Allocation Strategy
The task allocation process is designed to maximize both coverage and fairness while adhering to privacy constraints. The system considers three critical conditions before assigning a task to a vehicle:
- Call Probability: The vehicle must have a non-zero probability of making at least one call within the specified sensing period and region. This ensures basic participation feasibility.
- Dwell Time Sufficiency: The vehicle’s noise-masked dwell time must probabilistically satisfy the minimum sensing duration required by the task.
- Travel Time Feasibility: The time taken for the vehicle to travel between consecutive task locations must not conflict with task start and end times.
Tasks are allocated in a round-robin fashion to prevent any single vehicle from being overloaded, thereby ensuring load balancing. The algorithm prioritizes tasks based on their deadlines and dynamically adjusts assignments to accommodate real-time constraints.
Performance Evaluation
The proposed scheme was evaluated through extensive simulations comparing its performance with existing approaches such as UBTA, LBTA, and MPPTA. Key metrics included task completion rate (TCR), load fairness index (LFI), and average task completion time (AvTCT).
Results demonstrated that the scheme achieves a high TCR while maintaining superior load balancing, particularly when the number of tasks scales up. The privacy budget was found to significantly influence TCR—higher privacy budgets (less noise) improved task allocation success but reduced privacy guarantees. Notably, the scheme outperformed baseline methods in scenarios with large task volumes, proving its robustness in real-world VCS deployments.
Conclusion
This paper presents a comprehensive solution to the underexplored challenge of temporal privacy preservation in vehicular crowdsensing. By integrating differential privacy with a probabilistic task allocation framework, the scheme effectively obscures sensitive dwell time data while ensuring efficient and fair task distribution. Experimental results validate its superiority over existing methods in terms of both privacy protection and operational efficiency. Future work may explore adaptive privacy budgeting and decentralized task allocation to further enhance scalability and resilience.
DOI: 10.19734/j.issn.1001-3695.2024.09.0320
Was this helpful?
0 / 0