基於形狀與移動分析分辨網球員動作類別

Abstract

隨著傳播和儲存技術的發展，數以萬計的網球影片被製作並且上傳到網路上。辨識網球影片的動作對於進階的視覺化訓練應用提供巨大的幫助，儘管有著巨大的幫助，但是因為傳播的影片數量龐大，人工去辨識網球的所有動作是冗長且不可能的，因此，自動化偵測分析網球運動動作成為一項重要的議題。在這篇論文中，藉由分析球員的形狀和移動，我們提出一個創新的網球員動作分析方法。我們藉由過濾網球員的背景顏色和球場線來獲取球員的形狀，再藉由這個形狀來取出輪廓點的分布圖特徵。使用網球員的移動特性和形狀的特徵來訓練我們的分類器。我們的實驗顯示出我們提出的分法達到令人滿意的表現，整體的正確率表現相當高。

Along with development of the broadcast and online storage, hundred thousands of tennis videos are created and uploaded on Internet. Identifying events in video offers great benefit for advancing visual sports training applications. Despite having this advantage, it is labor-intensive for users to label events one after another. Thus, automatically detecting and classifying the player events from video database has become an important issue. In this thesis, we propose a novel method for tennis player events detection and classification by player shape recognition and movement analysis. After extract the shape of a player, we take contour points histogram as the shape features. They are then combined with the properties of the player movements to train the classifier of player event classification. The experiments demonstrate that the proposed method achieves satisfactory performance with high accuracy.