Decomposable Intelligence on Cloud-Edge IoT Framework for Live Video Analytics
Yi Zhang, Jiun-Hao Liu, Chih-Yu Wang, Hung‐Yu Wei
Abstract
With the rapid development of deep learning technology, the modern Internet-of-Things (IoT) cameras have very high demands on communication, computing, and memory resources so as to achieve low latency and high accuracy live video analytics. Thanks to the mobile-edge computing (MEC), intelligent offloading to the MEC nodes can bring a lot of benefits, especially when the decomposable pipeline is adopted in the cloud-edge architecture. In this article, we provide decomposable intelligence on a cloud-edge IoT (DICE-IoT) framework to support joint latency- and accuracy-aware live video analytic services. Specifically, the intelligent framework enables the pipeline-sharing mechanism to reduce MEC resource usage. A Nash bargaining is proposed to incentivize cooperative computing provision between the MEC and the cloud, and a generalized benders decomposition (GBD)-based approach is utilized to optimize the social welfare. The results show that the proposed DICE-IoT framework can achieve a win–win–win solution to the IoT device, the MEC, and the cloud stratum.