Skip to Main content Skip to Navigation

Collaborative Crowdsensing at the Edge

Yifan Du 1
Abstract : Mobile crowdsensing is a powerful mechanism to contribute to the ubiquitous sensing of data at a relatively low cost. With mobile crowdsensing, people provide valuable observations across time and space using sensors embedded in/connected to their smart devices, e.g., smartphones. Particularly, opportunistic crowdsensing empowers citizens to sense objective phenomena at an urban and fine-grained scale, leveraging an application running in the background. Still, crowdsensing faces challenges: The relevance of the provided measurements depends on the adequacy of the sensing context with respect to the phenomenon that is analyzed; The uncontrolled collection of massive data leads to low sensing quality and high resource consumption on devices; Crowdsensing at scale also involves significant communication, computation, and financial costs due to the dependence on the cloud for the post-processing of raw sensing data. This thesis aims to establish opportunistic crowdsensing as a reliable means of environmental monitoring. We advocate enforcing the cost-effective collection of high-quality data and inference of the physical phenomena at the end device. To this end, our research focuses on defining a set of protocols that together implement collaborative crowdsensing at the edge, combining: • Inference of the crowdsensor’s physical context characterizing the gathered data: We assess the context beyond geographical position. We introduce an online learning approach running on the device to overcome the diversity of the classification performance due to the heterogeneity of the crowdsensors. We specifically introduce a hierarchical algorithm for context inference that requires little feedback from users, while increasing the inference accuracy per user. • Context-aware grouping of crowdsensors to share the workload and support selective sensing: We introduce an ad hoc collaboration strategy, which groups co-located crowdsensors together, and assigns them various roles according to their respective contexts. Evaluation results show that the overall resource consumption due to crowdsensing is reduced, and the data quality is enhanced, compared to the cloud-centric architecture. • Data aggregation on the move to enhance the knowledge transferred to the cloud: We introduce a distributed interpolation-mediated aggregation approach running on the end device. We model interpolation as a tensor completion problem and propose tensor-wise aggregation, which is performed when crowdsensors encounter. Evaluation results show significant savings in terms of cellular communication, cloud computing, and, therefore, financial costs, while the overall data accuracy remains comparable to the cloud-centric approach. In summary, the proposed collaborative crowdsensing approach reduces the costs at both the end device and the cloud, while increasing the overall data quality.
Complete list of metadatas

Cited literature [225 references]  Display  Hide  Download
Contributor : Yifan Du <>
Submitted on : Monday, August 10, 2020 - 2:37:25 PM
Last modification on : Wednesday, August 12, 2020 - 3:15:17 AM


Files produced by the author(s)


  • HAL Id : tel-02913750, version 1


Yifan Du. Collaborative Crowdsensing at the Edge. Ubiquitous Computing. Sorbonne Université, 2020. English. ⟨tel-02913750⟩



Record views


Files downloads