多應用視頻分析的智能資源分配的 5G 邊緣計算實驗

5G Edge Computing Experiments with Intelligent Resource Allocation for Multi-Application Video Analytics

趙慈暄(Tzu-Hsuan Chao)

指導教授 ： 魏宏宇

國立臺灣大學/電機資訊學院/電機工程學研究所/碩士(2021年)

https://doi.org/10.6342/NTU202103356

摘要

第五代移動網絡的特點是無線連接所有人和智能自動化的時代。從技術上講， 它需要低延遲和高質量的服務。因此，邊緣計算的概念最近逐漸興起，或者可以稱 為移動邊緣計算(MEC)。邊緣服務器是指靠近用戶設備(UE)的計算節點;換句 話說，數據傳輸延遲會比雲服務器減少。此外，在多個視頻直播分析場景中使用了 智能識別和檢測，這意味著 5G 網絡將很好地解決檢測服務中的時延要求和準確性 要求。論文提出了一種邊緣架構多應用智能資源管理方法(RMMAE)。該方法將計 算任務重新分配到邊緣服務器或云服務器。我們實現了算法來分配計算資源，並在 我們的 Edge 測試平台中與面部檢測、對象檢測和姿態估計配合使用。邊緣計算 和雲計算之間存在權衡。從實施結果來看，我們在測試平台上證明了令人印象深刻 的改進和性能。

關鍵字

邊緣計算 ； 霧計算 ； 資源分配 ； 實時視頻分析 ； 物件辨識 ； 姿態估計 ； 服務品質

並列摘要

The fifth-generation mobile network is characterized as the age of wireless connectivity for all and intelligent automation. Technically, it requires low latency and high-quality service. As a result, the concept of edge computing is gradually rising recently, or it may be called Mobile Edge Computing (MEC). The edge server means a compute node near the user equipment (UE); in other words, the data transmission delay will decrease than the cloud server. Moreover, intelligent recognition and detection are used in several live video analytic scenarios, implying that the 5G network will solve great with the latency requirement and accuracy requirement in the detection services. The thesis proposed an intelligent Resource Management method with Multiple Applications in Edge architecture (RMMAE). The method will reallocate the computing task to an edge server or a cloud server. We implement the algorithm to allocate computing resources and cooperate with facial detection, object detection and pose estimation in our Edge testbed. There is a trade-off between edge computing and cloud computing. From the implementation results, we prove impressive improvement and performance on our testbed.

並列關鍵字

Edge Computing ； Fog Computing ； Resource Allocation ； Live Video Analytic ； Object Detection ； Pose Estimation ； Quality of Experience

參考文獻

[1] M. Aledhari, R. Razzak, R. Parizi, and F. Saeed. Federated learning: A survey on enabling technologies, protocols, and applications. IEEE Access, 8:1–1, 01 2020. doi:10.1109/ACCESS.2020.3013541.

[2] J. Chen, K. Li, Q. Deng, K. Li, and P. S. Yu. Distributed deep learning model for intelligent video surveillance systems with edge computing. IEEE Transactions on Industrial Informatics, pages 1–1, 2019. doi:10.1109/TII.2019.2909473.

[3] Q. Cheng, H. Shan, W. Zhuang, L. Yu, Z. Zhang, and T. Quek. Design and anal­ ysis of mec­ and proactive caching­based 360 mobile vr video streaming. IEEE Transactions on Multimedia, 2021. Publisher Copyright: IEEE. doi:10.1109/ TMM.2021.3067205.

[4] Y. Chiang, Y.­H. Chao, C.­H. Hsu, C.­T. Chou, and H.­Y. Wei. Virtual network embedding with dynamic speed switching orchestration in fog/edge network. IEEE Access, 8:84753–84768, 2020. doi:10.1109/ACCESS.2020.2991986.

[5] Y. Chiang, C.­H. Hsu, and H.­Y. Wei. Collaborative social­aware and qoe­driven video caching and adaptation in edge network. IEEE Transactions on Multimedia, pages 1–1, 2020. doi:10.1109/TMM.2020.3040532.