標題: 利用物件軌跡之自動化壓縮視訊資料擷取系統之設計與實作
Automatic Access to Compressed Video Data via Object Motion: Design and Implementation
作者: 張佑鼎
Chang Yu Ting
楊維邦
Dr. Wei-Pang Yang
資訊科學與工程研究所
關鍵字: 視訊資料庫;運動物件;壓縮視訊;軌跡比對演算法;視訊內容存取;Video Database;Object Motion;MPEG;Trajetory Matching;Query by Example
公開日期: 1998
摘要: 近年來在儲存硬體容量與作業系統技術的突破使得以往無法管理與存取的大量資料如視訊資料,已經逐漸受到重視。視訊影片(資料)是由場景與其中移動的物件所組成的。這些移動的物件通常即為整部影片中最容易引起人們注意的焦點與影片內容的重心。如何利用移動物件來作視訊資料內容的存取即為本篇論文所要探討的課題。 本論文設計與實作出一個利用物件移動軌跡來存取視訊資料的系統。整個系統的流程由移動物件特徵的擷取,利用物件特徵來表示物件運動(object motion),物件運動的比對,並以舉例方式查詢。其中物件特徵的擷取是由壓縮視訊資料中自動化處理後直接取得的。在物件運動的表示法與比對查詢上,我們提出了一個新的方法,利用物件影像與空間時間上的特徵來完整的表示一個移動物件。在比對的演算法上,我們加入了物件移動距離的考量,使得物件運動軌跡的比對更加正確,並以數個軌跡相似度的量化標準實驗來佐證其成效。最後,使用者可以舉例的方式,在系統的界面上描繪出物件運動的軌跡與其他影像特徵,以最自然與具描述力的方式來查詢視訊資料的內容。
Recent progress in storage device and operating system technologies has made it possible to manage multimedia data such as text, image, sound and video. Video data is the most informative type of data among these. It consists of a scene and some moving objects conveying particular meaning in the scene. The moving objects are usually the most interested and important content in video data. Thus, to access video data via object motion is what we focused in this thesis. We present a system that retrieves video data via an object motion. The procedure of the system is: (1) moving object feature extraction, (2) modeling object motion from feature extracted, and (3) matching of object motion. Feature extraction is processed automatically from compressed video data. We propose a new object motion representation and matching algorithm using both image and spatio-temporal feature of moving objects so that the query result would be more precise. Finally, user can issue query in the most descriptive and natural way by specify a hand-drawn trajectory together with other features of objects.
URI: http://140.113.39.130/cdrfb3/record/nctu/#NT870394006
http://hdl.handle.net/11536/64143
顯示於類別:畢業論文