HSM-SMCS: Task Assignment Based on Hybrid Sensing Modes in Sparse Mobile Crowdsensing

Xiaohui Wei,Zijian Li,Chenghao Ren,Tao Guo,Shang Gao

DOI: https://doi.org/10.1109/jiot.2022.3150804

IF: 10.6

2022-01-01

IEEE Internet of Things Journal

Abstract:Sparse mobile crowdsensing (Sparse MCS) is an emerging paradigm for urban-scale sensing applications, which recruits suitable participants to complete sensing tasks in only a few selected cells and then infers data of unsensed cells for saving sensing costs and obtaining high-quality sensing maps. In Sparse MCS, one crucial issue is task assignment, in which the platform selects cells whose sensing data can reduce inferred sensing maps errors (i.e., cell selection) and recruits the participant set with the maximum contribution for performing tasks (i.e., participant recruitment). The research on participant recruitment mainly focuses on single participatory-based or single opportunistic-based sensing mode. Due to the complementarity of two sensing modes, recruiting participants by only one sensing mode would result in wasting sensing resources and compromising the quality of task completion. Thus, combining the advantages of two sensing modes, we propose a task assignment framework based on hybrid sensing modes in Sparse MCS (HSM-SMCS) for achieving a good tradeoff between sensing quality and cost. Specifically, we propose a heuristic two-stage search strategy that simultaneously recruits opportunistic and participatory participants to perform tasks in significant cells within the constraint of total costs, considering their contributions to sensing map inference. Thereinto, for opportunistic participants, mobility prediction greatly affects task assignment effectiveness. However, existing prediction algorithms lead to unsatisfactory outcomes when the historical trajectory data of opportunistic participants are scarce. To effectively improve the predictive accuracy, we design a mobility prediction model based on transfer learning. The experimental evaluation on real trajectory data sets and sensor data sets of corresponding areas demonstrates that our framework outperforms state-of-the-art methods with higher quality reconstructed sensing maps.