高品質多視角影像對於分析現代競技運動中運動員的技術動作至關重要。然而，部署專業多視角攝影系統往往成本高昂；即便採用Full HD影像，為了涵蓋寬廣場域，包含運動員的前景區域通常仍為低解析度。本研究聚焦於從單目影片合成運動員高品質新視角影像的問題。現有基於神經輻射場（NeRF）的方法在此情境下表現受限，尤其在低解析度前景、大幅且快速動作，以及複雜戶外光照等條件下更顯不足。

為克服上述挑戰，我們提出一套新穎的整合框架，結合密集人體輻射場（Dense Human Radiance Field）與方向感知多層感知器（Direction-aware MLP）來解決這些問題。密集人體輻射場在低解析度限制下增強場景表示能力，方向感知多層感知器則改善對視角相依顏色變化的建模。與既有方法不同，本方法無需高解析度訓練資料，能直接合成逼真的人體外觀。

在具挑戰性的戶外運動場景大量實驗中，結果顯示本方法持續產生高品質新視角影像，性能超越最先進技術，為真實環境中的運動員表現分析提供低成本且有效的解決方案。

High-quality multi-view images are essential for analyzing athletes’ technical movements in modern competitive sports. However, deploying professional multi-view camera systems is often cost-prohibitive, and even with Full HD footage, the foreground regions containing athletes typically remain low-resolution due to the need to cover wide fields. This work addresses the problem of synthesizing high-quality novel-view images of athletes from monocular videos. Existing NeRF-based methods struggle in this setting, especially under low-resolution foregrounds, large and rapid motions, and complex outdoor lighting.

We propose a novel integrated framework that overcomes these challenges by combining a Dense Human Radiance Field, which enhances scene representation under low-resolution constraints, with a Direction-aware MLP, which improves the modeling of view-dependent color variations. Unlike previous methods, our approach does not require high-resolution training data and is capable of directly synthesizing realistic human appearances.

Extensive experiments on challenging outdoor sports scenarios demonstrate that our method consistently produces high-quality novel views, outperforming state-of-the-art techniques. This offers a low-cost and effective solution for athlete performance analysis in real-world settings.