基於事件資料探勘方式進行的體育視訊分析,檢索和查尋需求等在近年來變得越來越重要。體育項目中的足球,籃球,橄欖球與網球等運動則是觀眾最多的項目之一。因此基於語義分析方式的自動體育錄影總結並提供瀏覽、戰術分析等功能變成是一種視訊處理的重要之應用需求之一。針對此需求下的影像場景分類與球員或是球的標記和跟蹤等工作則是最重要的任務。但是由於如鏡頭的伸縮、視訊畫面拍攝時光線亮度的不同、球場與球員的陰影,照相機的位置移動,球員之姿勢變化及其他各種各樣的雜訊甚至多個攝影機等許多因素而造成視訊中各影像之內容極為不同因此使得自動視訊分析具有相當的難度。一般而言體育分析的錄影來源可能通常有二個類型,一個是廣播錄影的視訊,另一個則是個人單機方式的錄影視訊。廣播錄影的畫面由於具有較多的觀眾與內容因此更具實用的價值。 在本篇論文中我們提出一套足球廣播錄影視訊的自動場分析判斷和球員偵測與追蹤的法則。我們的方法包括二個階段,第一個是自動劃分錄影的場景分析工作。我們提出一套基於HSV顏色的高斯分類模型並設計了一個隨視訊內容自動更新模型的方法。第二個是球員分類階段。我們提出一個基於HSV直方圖判斷樹的無監督分類方法進行球員的分類工作,並且利用線性預估模型進行球員的追蹤定位工作。實驗結果證明我們提出的方法是簡單並且有效的。
Event mining of sports video for understanding, event retrieval and content searching is becoming more and more important in recent years. Sports items, such as soccer, basketball, football and tennis, are among the most interesting parts. One of the interesting applications is automatic sports video summarization and browsing. Semantic based analysis, such as highlight detection, tactics analysis, and player activity analysis are the goals for the applications. In these applications, the scene or shot classification and the localization, labeling and tracking of players are the most important tasks. These tasks are quite challenging due to many difficulties, such as different daylight, shadow, occlusion, similar player appearance with low discrimination, varying number of players, abrupt camera motion, player pose variance, various noises. The video source for sports analysis can generally divide into two types, one is the broadcast video, the other is the personal made video, and the broadcast video is the most interesting one due to its great amount of viewers and video contents. In this thesis, we proposed a broadcast soccer video analysis system to analyze the scene and perform the detection and tracking of the players automatically. Our method consists of two phase, the first one is the video scene analysis phase which can classify the video into different scene or shot based on hue and saturation 2-D Gaussian color model automatically. An incremental color model update method is proposed. The second one is the player analysis phase. A color histogram based method is proposed to classify the player using a decision tree and a linear prediction model is used for tracking the players. Experimental results show that the proposed method is simple yet effective.