由於資訊技術和儲存媒體的進步，視訊和影像的資料量與日俱增，使得如何從視訊資料庫找出頻繁樣式成為一項重要的研究課題，雖然近年已有許多相關研究被提出，但仍是一個具有相當挑戰性的議題。因由視訊資料庫探勘出有意義的時間空間頻繁樣式仍具有兩方面的問題，一個是如何表示影片中物件間時間空間的關係變化，因視訊是由一連串連續畫面組成，每一個畫面均具有多個物件，要如何簡單且完整的表示視訊物件間的時間空間關係且適用於所使用的探勘方法為第一個問題。第二個問題則是如何有效率的探勘視訊物件間的時空關係頻繁樣式。所以本研究首先針對視訊物件間時間關係，利用序列樣式的概念產生出新的視訊物件時間關係表示法，再利用PrefixSpan的探勘方法提出新的視訊物件時間關係序列頻繁樣式演算法，以能更有彈性及效率的由視訊資料庫中探勘出具有意義的視訊物件頻繁樣式。
Recently, information technology and storage media are developed rapidly; image and video data also grows with every day. It would be an important issue to find meaning patterns form video database. Although there are many related researches, there are still two questions to deal with this issue. First, question is how to simply and completely represent the relationships between the objects in a video.
Another is how to use efficiency mining methods to mine the frequent patterns from a video database. Accordingly, in this paper, we use the concept of sequence mining to represent the time relationships between video objects, and develop a new video mining method based on PrefixSpan mining approach called VTPrefix to find the temporal frequent patterns between objects in a video database.
