English  |  正體中文  |  简体中文  |  Items with full text/Total items : 56577/90363 (63%)
Visitors : 11890156      Online Users : 67
RC Version 7.0 © Powered By DSPACE, MIT. Enhanced by NTU Library & TKU Library IR team.
Scope Tips:
  • please add "double quotation mark" for query phrases to get precise results
  • please goto advance search for comprehansive author search
  • Adv. Search
    HomeLoginUploadHelpAboutAdminister Goto mobile version
    Please use this identifier to cite or link to this item: http://tkuir.lib.tku.edu.tw:8080/dspace/handle/987654321/35133


    Title: Classification and retrieval on human kinematical movements
    Other Titles: 人體運動行為之分類與搜尋
    Authors: 黃俊宏;Huang, Chun-hong
    Contributors: 淡江大學資訊工程學系博士班
    施國琛;Shih, Timothy K.
    Keywords: 骨架辨識樹;人體運動;人體運動搜尋重現;變形動態演算法;虛擬實境模組語言;Skeleton Discrimination Tree;human movement;Human movement retrieval;Mutative Dynamic Programming;VRML
    Date: 2007
    Issue Date: 2010-01-11 06:03:29 (UTC+8)
    Abstract: Motion retrieval是一個相當有趣且具挑戰性的研究議題,然而大多數的Motion retrieval系統是建立在2D視訊影像的基礎上。但是隨著3D動作捕捉器影像技術及3D VRML動畫的呈現,使得真實世界中人體運動的軌跡路徑得以利用3D的電腦技術呈現並加以分析及辯識。在本篇論文裡,我們提出一個3D人體運動動作搜尋重現系統,此系統可以使得使用者找出相似的3D人體運動動作,對於動作的分析與比較上,系統主要包含兩個主要的單元,第一個單元是動作類別的分類辨識,其依據的方法是以 Skeleton Discrimination Tree為基礎,Skeleton Discrimination Tree 根據人體運動時,四肢能量的分佈來做動作型態的初步分類,在非跳躍的動作類別型態上的辨識具有明顯的效果,另外可以利用腳部在y軸上能量明顯劇烈的變化以及雙腳是否離開地板等條件來判斷辨識跳躍與非跳躍兩種不同動作群組,此動作類型辨識單元可以過濾掉非相關類別的運動動作。第二個單元包含動作與時間相似度的比較,比較的方法是以mutative dynamic programming演算法為基礎,動作相似度比較是以兩個個別的動作上相對應到的關節點的運動軌跡做比較,我們詢問了數位體育界教授的意見及參考了許多文獻資料,根據教授們的意見及文獻資料訂定了人體上16個做為軌跡追蹤的重要關節點,包括手腕、手肘、腳踝、膝蓋、頸部、頭及其他重要的人體關節點等,做為比較的軌跡都是由這16重要特徵關節點所擷取出,每一段軌跡都是由連續的點座標所組成的,這些連續的點座標會被轉換成並以連續的向量方式呈現,我們就以不同空間中的兩個相對應軌跡上的向量所形成的夾角大小程度來判斷是否是屬於一樣的序列元素,以mutative dynamic programming求出兩軌跡路徑的最長共同序列,有愈長的共同序列表示動作相似度愈高。求出兩軌跡路徑的最長共同序列後,我們進一步計算出最長共同序列相對應元素的時間差異度,避免系統將雙手同時舉起與雙手沒有同時間舉起的動作誤判為一樣的動作。
    在我們所擷取出的連續特徵點座標是一連串的數字資料,一連串的數字資料對於使用者而言並不能提供及滿足視覺化的效果與享受,使用者很難以連續的數字資料感受到整個人體運動動作在空間與時間上實際的變化,因此在我們所獲得的軌跡座標資料會被匯入到VRML格式的3D人體模組,使用者只要給予一個3D VRML人體動作模組,系統就會自動依據相似的程度排列找出相似的動作。我們並協請三位體育老師及碩士班學生為本系統所執行出來的結果做佐證,實驗證明此系統所搜尋的結果具有良好之成效。
    另外我們系統中並提供調適性參數能夠讓使用者根據自己的觀點去調整參數以找出使用者所需要之動作。此外查詢的動作物件也可以跟資料庫中標準的動作做比較以找出動作之間的差異。我們希望此系統可以提供動畫設計師所需要的動作並套用不需再重新製作一個新的動畫,以減少動畫製作上的時間與成本,並且也可以提供教練或運動員做為運動技術上動作改進的輔助工具。
    Motion retrieval is a quite interesting but challenging research topic. However, for the most motion retrieval systems are designed based on 2-D video information. But, with the motion capture of video technology and presented by VRML animation, it is possible to automatically represent, analyze and adjust the 3D motions of real person. In this study, we propose a 3D human movement retrieval system, which allows users to retrieve 3D kinematical movements. The system includes two major components for movement analysis and comparison. The first one is a recognition unit of movement types which is based on Skeleton Discrimination Tree. The Skeleton Discrimination Tree can judge movement types in the field of Un-Jump. According to the conceptions of violent variation of energy of foots in Y axis and the feature information that if both foots are stuck on the ground, we can distinguish the movement types which belong to groups of “Jump” or “Un-Jump”. This recognition component can filter unallied human movements. The second unit includes movement and synchronization similarity. The comparing approach is based on mutative dynamic programming that considers the degree of the included angles of the vectors which belong to individual feature tracks. There are 16 track points include head, knee, elbow, wrist, etc. and further aggregate important features of human joints. The trajectories that can be used to comparison are extracted from these 16 feature joints. Each trajectory is composed of serial coordinates. The serial coordinates would be transformed and represented as successive vectors. We use the succession of vectors as the feature information to compute the similarity based on mutative Dynamic Programming.
    The forms of the feature coordinate points are variational numeral data. The variational numeral data can not provide satisfied for human sense of sight. Furthermore, only based on the variational numeral data, it is difficult to experience the motion variation of whole human body parts with a spatial-template domain. To solve this problem, the feature coordinates points of human body parts’ trajectories are transformed into a 3-D human body model as VRML animations. Users may give a VRML human movement object, and find the similar human movements via the system. As a result, the system can automatically retrieve similar actions. The results are tested by three professors of physical education and master students with a good satisfaction. Besides, our system provides adaptive parameter which dynamically calculated according to user’s perception of motion features. The query object also can compare with standard human kinematical motion to find the difference in each joint.
    Appears in Collections:[Graduate Institute & Department of Computer Science and Information Engineering] Thesis

    Files in This Item:

    File SizeFormat
    0KbUnknown425View/Open

    All items in 機構典藏 are protected by copyright, with all rights reserved.


    DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU Library & TKU Library IR teams. Copyright ©   - Feedback