近年來在視訊資料庫中探勘頻繁樣式已有許多研究成果,但由視訊資料庫中探勘出有意義頻繁樣式仍具有兩方面的問題。一是如何表示視訊的頻繁樣式,其中一種方法是以視訊中物件間所具有的時空關係來表示,過去研究曾提出9DST視訊表示法來表示視訊中已標示為最小矩形物件間的時空關係,研究成果亦顯示可適用於視訊中物件時間頻繁樣式的探勘;另一個問題則是使用何種探勘方法,現有大部份的研究成果均使用Apriori-like的探勘方法,其中9D-SPA影像探勘演算法針對影像中物件間空間關係進行探勘,並提出推導方法減少候選樣式的數量,且利用索引結構減少掃描資料庫的成本。所以在本研究中利用9DST視訊表示法、9D-SPA影像探勘演算法及9DST視訊物件間時間關係探勘方法,更進一步提出視訊中物件間時空關係頻繁樣式探勘演算法,以由視訊資料庫中找出更精確的頻繁樣式。
Recently, there are many researches stressing on mine the frequent patterns in video databases. There are two main issues in finding the meaningfully frequent pattern from the video databases. The first issues is the presentation of frequent patterns of videos. One of the methods is the spatial-temporal relations that exist between the objects in a video. The 9DST presentation method of video uses the minimum bounding rectangle to represent spatial-temporal relations between the objects in a video. Therefore, 9DST presentation can be used in frequency pattern mining in video databases. The mining approach is another issues. Most proposed video pattern mining algorithms use Aprio-ri-like algorithms. The 9D-SPA image mining approach is based on Apriori-like algorithm and use spatial relation in image database mining. Accordingly, in this research, we propose a new video mining method combining 9DST presentation and the 9D-SPA image mining approach to find the frequent patterns of spatial-temporal relations be-tween objects in the video database.