摘要
虛擬實境憑藉著快速成長的普及率與帶給使用者強烈共存感的全新互動模式，成為下一代社交平台將是未來發展之重要趨勢。有鑑於此，本研究設計了一個虛擬實境社交系統，不僅讓使用者跨空間交流，還著重於藉由提升虛擬人物寫實程度與還原社交過程中的非語言溝通行為，提供給使用者深度沉浸感與真實感。本研究對正臉照片進行高品質臉部建模，且以新建模之臉部模型作為基準，生成出各種表情的臉部模型。接著再運用經卷積神經網路訓練之嘴形辨識器分辨用戶嘴形，控制虛擬人物臉部表情與使用者達成同步，另外藉由手臂連動技術使虛擬人物即時重現使用者上半身的肢體動作。結合上述各項技術，還原表情、肢體語言這兩項非語言溝通行為，且提高虛擬人物的寫實感與細緻度，讓使用者在虛擬實境中能夠真實的與對方進行面對面互動，達到最佳的溝通社交效果。
並列摘要
Virtual Reality(VR) is becoming the mainstream of social interaction platform considering its rapid growth and innovative immersive user interaction experience. Hence, the paper proposes a VR social application which provides users great immersive experiences. It not only enables users in different locations to communicate in real time, but features reality of virtual characters and presentation of nonverbal communications. First, the paper implements high-quality face model reconstruction using single front face photo to generate the face model with neutral expression, and generates four face models with different facial expression based on it. Then, expressions of the user and the virtual character are synchronized using convolutional neural network(CNN) model which is trained by mouth images with different expressions. Last, arm movement synchronization technique is implemented to present the movement of users' upper body in real time. In conclusion, by reconstructing synchronizing users' facial expressions and upper-body movements, the proposed system not enables users to communicate with each other without geographical boundaries, but makes VR social more real.