無標記式動作擷取技術運用於體感互動遊戲之研究

本文提出了通過使用低成本的靜態攝影機進行無標記人體動作捕捉。我們利用鉸鏈式人體骨架代表複雜的人體，並且不需要任何深度傳感器的輔助來取得深度資訊。首先將靜態照相機進行校準，取得適合的顏色和亮度。再根據靜態畫面與動態畫面的兩者差異，辨別出人體的輪廓。隨後將圖像劃分為多個大小一致的區域以增加辨識的效率。此外，姿勢決策樹已在初始化階段定義，目的是用於姿勢轉變的判斷，再根據不同的姿勢轉換不同的動作追蹤策略。最後，採用標準的BVH格式來存儲連續的運動數據，並提供其他系統進一步的使用。實驗結果顯示本論文所提出的方法，可以在成本大幅低於穿戴式動作擷取器的情況之下，擷取可接受準確度的人體連續動作。

The paper presents an easy-to-implement technique of performing a markerless humanoid motion capture by using a low-cost and stationary camera. We make use of articulated simplified skeleton to represent a complicated human body and do not require any auxiliary multiple depth sensors for obtaining depth information. First, a static camera is calibrated to obtain applicable color and luminance. A T-pose of a character is required to distinguish which part is the humanoid character within an image and to facilitate the motion tracking. Subsequently, we divide the image into a variety of uniform regions for rapidly estimating a human’s pose. A pose decision tree has been defined in advance to benefit pose transitions while the character is changing a motion. Finally, a standard BVH color format is adopted to store the continuous motion data for further usage. Experimental results show that the proposed method is practical and feasible for capturing human motions. In particular, the cost of the proposed method is less expensive in comparison with the marker-based motion capture systems.