考量使用者暈眩之基於圖塊分割的適應性360度全景影音串流

Abstract

隨著各種虛擬實境裝置的出現，360度影片在生活中會有著越來越重要的地位。與傳統影片相比，360度影片串流所需的頻寬消耗為數倍甚至十多倍，因此圖塊(tile)分割式串流被用來有效率的傳輸這類有高頻寬與低延遲需求的360度影片。傳統的圖塊選擇與串流演算法在測量用戶體驗特質(Quality of Experience)時並沒有考慮到一項重要的因素：虛擬實境暈眩(cybersickness)，因此本文制定一個新的最佳化問題：圖塊選擇與暈眩控制(TSCC)，在設計有暈眩減輕方法之360度影片串流系統中，最小化選擇圖塊與使用暈眩減輕方法的成本。為了解決TSCC，我們將圖塊與暈眩減輕方法之參數的選擇建構為馬可夫決策過程(Markov Decision Process)來找到小型案例之最佳解，接著抽取其中的特質來設計有效且快速的演算法：有效圖塊選擇與暈眩減輕演算法(ETSCAA)來解決大型案例。模擬結果顯示在各種頻寬、圖塊數量、視區(viewport)的變化下，我們的演算法皆能勝過傳統演算法。

With the emergence of various virtual reality (VR) devices, 360° videos are guaranteed to play increasingly important roles in the near future. In contrast to traditional 2D videos, the bandwidth requirement for streaming a 360° video is an order of magnitude larger. The tile-based streaming algorithm has been proposed to effectively deliver bandwidth-hungry and delay-restricted 360° videos. However, previous algorithms do not consider cybersickness, one of the significant factors of Quality of Experience (QoE), as part of the QoE metric. In this thesis, we investigate a new optimization problem, named Tile Selection with Cybersickness Control Problem (TSCC), in a tile-based adaptive 360°video streaming system with cybersickness alleviation. We aim to minimize the cost function by choosing the best tile set for prefetching and deciding how to use the cybersickness alleviation methods. To solve TSCC, we model the tile and parameter selection as a Markov Decision Process to find the optimal solutions in small cases. Then we extract the intrinsic ideas behind to devise an efficient and effective heuristic algorithm, named Effective Tile Selection and Cybersickness Alleviation Algorithm (ETSCAA), for large cases. Simulation results manifest that our algorithm can outperform baselines no matter bandwidth condition, number of tiles, or viewport size.