3D人體動作之分析與檢索

The Analysis and Retrieval of 3D Kinematical Motions

楊勝雄(Sheng-Hsiung Yang)

指導教授 ： 施國琛

淡江大學/工學院/資訊工程學系碩士班/碩士(2006年)

https://doi.org/10.6846/TKU.2006.00079

摘要

人體運動的自動分析為近年來很熱門且極具有挑戰性的研究主題，但是，絕大部分的分析數據及研究成果都是設計在2D影像資訊上。因此，提出了一個新的研究方向，針對3D的人體運動分析為主軸，透過VICON Motion Capture System (動態捕捉系統)所輸出的純3D的檔案"CSM"，作為實驗分析的資料依據，本論文所設計的搜尋系統，提供給使用者手動調整權重值，可以動態調整至使用者想要看的結果，而這套系統將人體身上的關節點分為身體、頭部、手、腳...等。
本論文的研究方法是比對動作軌跡，利用補點運算，將三維檔案內的移動軌跡修正、再將其轉換成VRML格式之3D動畫運動，資料庫中大約分為四大類運動：排球、壘球、網球，與羽毛球，根據這些動作粹取出具特徵的關鍵動作，依據分類結果，再將關鍵動作加入權重值，接著比對關鍵動作之間的差異性並將其差異量化，透過數據資料可找出相似動作並以相似度做排序。

關鍵字

虛擬實境語言 ； VICON動態捕捉系統 ； 動作比對 ； 3維 ； 動態演算法

並列摘要

Automatic analysis of motions is an interesting and challenge research topic. But, most existing motion analysis systems are designed based on 2D video information. This thesis proposes a 3D motion retrieval system which supports the efficient analysis of 3D motions of sports made by VICON motion capture system to substitute for the traditional 2D information. The developed system provides adaptive weights function which dynamically calculated according to user’s perception of motion features. Besides, the representation of human skeleton includes head, hip, knee, elbow, wrist, etc. further to aggregate important features in sport motions based on kinematics is proposed. Finally, with the motion capture of video technology, it is possible to automatically represent, analyze and adjust the 3D motions of user via Internet.

並列關鍵字

VRML ； VICON ； Motion Retrieval ； 3D ； Dynamic Programming

參考文獻

[5].嚴子翔. “VRML虛擬實境網頁語言”, 知城數位科技股份有限公司, ISBN 957-0435-48-8。

[6].Dejan V. and D. Saupe. "3D Model Retrieval", Proceedings of Spring Conference on Computer Graphics 2000, Comenius University Press, Bratislava, Slovakia, pp. 89-93, May 2000.

[7].Vranic, D. V., ”Content-based Search for 3D-objects” in Proceeding of Computational Intelligence and Multimedia Applications. Fourth International Conference on ICCIMA, pp. 266 –270, 2001.

[12].C. Zhang and T. Chen. “Active Learning for Information Retrieval: Using 3D Models as An Example", Carnegie Mellon Technical Report: AMP01-04.

[13].C. Zhang and T. Chen. "Efficient Feature Extraction for 2D/3D Objects in Mesh Representation", ICIP 2001,Thessaloniki,Greece, Oct. 7-10, 2001.