混合實境（Mixed Reality, MR）的遠端渲染需要在有限的網路資源下，有效分配頻寬以提供沉浸式體驗。本論文提出一個名為「串流圖論優化」（Graph-Theoretic Optimization for Streaming, GOS）的創新框架，旨在應對此項挑戰。GOS框架將傳統上各自獨立的可見性、分群與品質決策，整合成單一且整體的優化過程。此方法將MR環境模型化為一個動態圖形，其中邊的權重編碼了空間與動態的相互關係。GOS採用一種迭代式的協同優化迴圈，讓這些相互依存的決策能夠彼此提供資訊並相互完善。這種作法從根本上突破了靜態、序列式流程的限制，以有原則、數據驅動的機制取代了過往的特定啟發式演算法，從而無需手動調整參數。實驗評估證明，GOS的表現顯著優於基於啟發式演算法的方法，不僅能額外減少34%的頻寬消耗，同時還能將定位精準度提升56%。此框架在多樣化且動態的場景中皆展現出穩健的效能，證實其作為下一代MR系統高效能、可擴展解決方案的潛力。;Mixed Reality (MR) remote rendering demands efficient bandwidth allocation to deliver immersive experiences under constrained network resources. This paper proposes a novel Graph-Theoretic Optimization for Streaming (GOS) framework that addresses this challenge by unifying the traditionally separate decisions of visibility, clustering, and quality into a single, holistic optimization process. By modeling the MR environment as a dynamic graph where edge weights encode spatial and motion-based relationships, GOS employs an iterative co-optimization loop where these interdependent decisions mutually inform and refine one another. This approach fundamentally breaks from the limitations of static, sequential pipelines, replacing ad-hoc heuristics with principled, data-driven mechanisms that eliminate the need for manual parameter tuning. Experimental evaluations demonstrate that GOS significantly outperforms heuristic-based approaches, reducing bandwidth consumption by an additional 34% while simultaneously improving positioning accuracy by 56%. The framework achieves robust performance across diverse and dynamic scenarios, confirming its potential as an efficient and scalable solution for next-generation MR systems.
