運用邊緣運算預測人體動作之多人無線虛擬實境串流

Human motion prediction for edge-assisted multiuser wireless virtual reality

蕭大哲(Ta-Che Hsiao)

指導教授 ： 廖婉君

國立臺灣大學/電機資訊學院/電機工程學研究所/碩士(2020年)

https://doi.org/10.6342/NTU202003275

摘要

無線虛擬實境有著低延遲、高頻寬的需求，而邊緣運算技術被視為一個解決方案，利用放置於使用者較近的邊緣伺服器，將渲染畫面等龐大的計算移植到邊緣伺服器上提供較低的計算延遲。然而在多人虛擬實境的應用中，異地使用者之間端到端延遲仍然是一個很大的挑戰，有著資料傳輸的物理限制，使得異地使用者彼此傳輸資料的延遲無法忽略，同時於虛擬實境中互動的使用者之間不同的延遲使得同步變得困難。在此論文中我們利用人體動作預測技術放置於邊緣運算伺服器，即時的預測遠端參與者動作並渲染於本地使用者畫面上，以滿足本地使用者之端到端低延遲的需求與同步需求，同時最小化預測造成的高計算成本。我們首先制定一個新的最佳化問題，稱為虛擬實境動作預測選擇問題(VMPS)，藉由選擇一部分畫面中的遠端使用者預測其動作來最小化預測成本同時滿足延遲與同步需求，接著我們證明NP-困難並提出一種新的算法，稱為成本節省預測選擇(CEPS)，通過對偶最佳理論來最小化成本並滿足延遲與同步需求。根據模擬結果表明，與傳統的減少延遲之資源分配相比，CEPS可以有效的降低超過40% 的成本。

關鍵字

虛擬實境 ； 端到端延遲 ； 同步 ； 人體動作預測 ； 邊緣運算 ； 近似演算法 ； 對偶理論

並列摘要

With virtual reality (VR) service becomes popular, edge computing has been seen as a potential solution to provide low latency for VR application, but it is still challenging for remote multi-user virtual reality to provide low end-to-end latency between users while synchronizing the states of the users. In this paper, we envisage a scenario of leveraging edge computing to predict human body motion of remote VR user, which brings major advantages over multi-user VR network: 1) hide the large end-to-end latency caused by the propagation transmission delay, and 2) proactively predict motion state of the user with worse uploading latency to synchronize interacting remote participants. However, the motion prediction task is computationally complex, which incurs large computation costs to VR service operator. Accordingly, we formulate the VR Motion Prediction Selection (VMPS) problem to select a subset of remote participants for motion prediction in multi-user VR. We prove the NP-hard and propose a new approximation algorithm, Cost-Efficient Prediction Selection (CEPS), based on the primal-dual optimization to select the proper subset of remote participants for motion prediction. Simulation results show that CEPS can effectively decrease the prediction cost by more than 40% compared with the baseline schemes.

並列關鍵字

Virtual Reality ； end-to-end latency ； synchronization ； human motion prediction ； edge computing ； approximation algorithm ； primal dual method

參考文獻

Markets and Markets. (2019) Virtual reality market - analysis and forecast to 2024.[Online]. Available: https://www.marketsandmarkets.com/Market-Reports/reality-applications-market-458.html 1

M. S. Elbamby, C. Perfecto, M. Bennis, and K. Doppler, “Toward low-latency and ultra-reliable virtual reality,” IEEE Network, vol. 32, no. 2, pp. 78–84, 2018. 1

A. Becher, J. Angerer, and T. Grauschopf, “c,” Virtual Reality, pp. 1–15, 2019. 1, 3

X. Hou, J. Zhang, M. Budagavi, and S. Dey, “Head and body motion prediction to enable mobile vr experiences with low latency,” in 2019 IEEE Global Communications Conference (GLOBECOM). IEEE, 2019, pp. 1–7. 2

Z. Lai, Y. C. Hu, Y. Cui, L. Sun, N. Dai, and H.-S. Lee, “Furion: Engineering high-quality immersive virtual reality on today’s mobile devices,” IEEE Transactions on Mobile Computing, 2019. 2