隨著混合實境（Mixed Reality, MR）在沉浸式元宇宙體驗中的日益普及，其對容積影像串流（volumetric video streaming）帶來了巨大挑戰，尤其是在受限的網路條件下支援六自由度（six-degrees-of-freedom, 6DoF）移動。本論文提出一個創新的信心驅動（confidence-driven）串流框架，以優化資源分配進而提升整體串流效能。我們將視野（Field of View, FoV）預測重構成一個分類問題，使模型不僅能預測使用者視口，更能產生一個對應的信心分數。此分數是建構一個新穎三維可能性圖（3D Likelihood Map）的關鍵，從而實現可擴展且精準的多用戶資源分配策略。
我們的方法展現了優異的效能，其 FoV 預測準確率可達 97.92 %，並證明信心分數是預測正確性的可靠指標。此高預測保真度直接轉化為系統效益；多用戶模擬結果顯示，即便在高密度場景下，我們提出的系統效能仍穩定優於基準方法，帶來了 12.9 % 至 27.2 % 的效益（utility）提升。本研究為在真實世界環境中部署高品質、多用戶的 MR 串流，提供了一個具可擴展性的解決方案。;The growing adoption of Mixed Reality (MR) for immersive metaverse experiences creates significant challenges for volumetric video streaming, particularly in supporting six-degrees-of-freedom (6DoF) movement under constrained network conditions. This paper introduces an innovative, confidence-driven streaming framework to optimize resource allocation to improve overall streaming performance. We reformulate Field of View (FoV) prediction as a classification problem, enabling our model to not only predict the user′s viewport but also to generate a corresponding confidence score. This score is integral to constructing a novel 3D Likelihood Map, which enables a scalable and precise multi-user resource allocation strategy. Our approach demonstrates strong performance, achieving a peak FoV prediction accuracy of 97.92\% and proving the confidence score to be a reliable indicator of correctness. This high predictive fidelity translates directly into system benefits. Multi-user simulations show that our proposed system consistently outperforms baseline methods, yielding a utility improvement of 12.9\% to 27.2\% even in high-density condition. Our work presents a scalable solution for deploying high-quality, multi-user MR streaming in real-world environments.
