排球影片攻擊及攔網事件之偵測與分析

Abstract

由於多媒體擷取設備的蓬勃發展，產生了大量的運動影片。對一般的觀眾而言，只有比賽中的一部分精華事件能引起他們的注意。而教練及球員則希望能夠針對比賽做戰術分析，像是對手所使用的攻擊戰術和防守策略等。在此篇論文中，我們提出了一個方法，能夠在排球影片中偵測攻擊及攔網事件，並且對攔網的型態做分析。首先我們對影片做前置處理，像是利用哨音偵測將影片切成一段一段連續對打的過程。接著找出球場及網子，進一步找出攻擊及攔網所發生的位置。然後我們分析網子內移動像素的改變，並且將攔網的區域框出來。而根據攔網區域的寬度，我們將攔網的型態分成兩種類別，無人攔網或一人攔網，以及兩人或三人攔網。最後我們可以利用攻擊事件發生的時間以及攔網的位置來做戰術上的分析，例如時間差及位置差的戰術。而我們的實驗結果顯示出我們提出的方法之有效性及正確性。

With the explosive proliferation of multimedia capturing devices, a considerable amount of sports videos have been recorded. For general viewers, only portions of games related to specific topics attract their attention. Coaches and professional players are interested in the high level semantics of events in the tactic perspective, such as the attack and defense tactics executed by opponents. In this thesis, we propose an approach for spiking events detection and blocking pattern classification for real volleyball videos. First, we perform video pre-processing such as whistle detection to segment the entire videos into video clips for rallies. Second, the camera calibration is performed to find the court region. To discover the positions of spiking and blocking, we further detect the location of net. Then, the changes of moving pixels are analyzed along the net. A bounding box is made around the blocking location, and we classify the blocking patterns into two main categories based on the width of bounding box. Finally, tactics analysis is made to discriminate two important tactics in volleyball matches: delayed spiking and alternate position spiking. With the iii information of spiking events and blocking locations, we can collect the statistical data and make tactics inference easily. The experimental results on the videos recorded by NCTU school volleyball team are promising and demonstrate the effectiveness of our proposed scheme.