影像式排球影片內容分析之探討

Abstract

在本篇論文中，我們提出一套排球比賽影片的智慧型分析系統，此系統透過兩項影片內容的分析，以自動截取出含有精彩攻防的活球片段。首先我們計算連續影像中較有變動的像素數目總和（稱之為活動變化量），藉由在活球開始前後，活動變化量會由一極小值迅速增多之趨勢，來辨別活球片段之起點。此外，藉由偵測排球之位置也可以輔助判別活球之狀態。在偵測排球部份，此系統採用連續影像相減的方法來切割出移動物體，並利用灰階值、形狀和飽滿度等排球特徵來偵測移動中的排球。另一方面，我們對球場上部份球員之軌跡追蹤也取得初步的成果。對於多場排球比賽的影片進行分析的結果顯示，本系統對於活球狀態之分析有高度的準確性。

In this thesis, we propose an intelligent system for the analysis of volleyball game video which can automatically extract key video segments, e.g., a ball-in-play state (BIPS), of a game using two types of image content. First, the number of significantly changed pixels between two consecutive image frames is computed as an activity measure. With the observation that, such a measure has a trend to increase dramatically from a local minimum as a BIPS begins, it is used to identify the BIPS. In addition, the ball position can also help game state analysis. The system first uses temporal image difference to segment moving objects. All candidates regions of the volleyball can then be filtered out using based on their intensity, shape and fullness. On the other hand, we also get some preliminary tracking results for partial players in a court. Experimental results of several volleyball game videos show that each beginning of BIPS can be found by the proposed system with high accuracy.