決策樹應用於網球視訊事件偵測

Event Detection for Broadcasting Tennis Sport Video Using Classification And Regression Tree

許鈞翔(Chun-Hsiang Hsu)

指導教授 ： 楊乃中

義守大學/電機資訊學院/資訊工程學系/碩士(2014年)

https://doi.org/10.6343/ISU.2014.00130

摘要

本研究的目的是利用分類與迴歸樹來對網球運動的視訊進行事件的分類。首先本研究應用適應性二值化的方式偵測出場線資訊，進而運用該資訊取得其交點座標並建立場地模型，接下來我們計算投影轉換參數用於定義球員追蹤的範圍。然後我們追蹤球員的位置，找到其移動的方向與速度，由於上場地的球員資訊不夠完整，我們提出以卡曼濾波器作為球員追蹤的主要依據。除此之外，在場地投影轉換的過程中可能會衍伸出上場球員位置上的誤差，在實驗的過程中對該誤差進行平滑化修正以降低事件誤判的機率。
在偵測上我們以賽事中常見的事件作為基準，並以發球作為事件為開端，並根據事件發生的時間長短、球員的位置、球員的移動距離等特徵進行決策樹的訓練。透過上述 以此作為判斷的基礎，完成更為精準的事件偵測系統，進而達成自動化的網球視訊分析系統。

關鍵字

運動視訊分析 ； 物件追蹤 ； 事件偵測 ； 卡曼濾波

並列摘要

In this thesis, we purpose a method to detect the events from tennis sport video based on classification and regression tree. First, we detect the court lines from videos. We can get intersection points from the line equations, and then build the court model. Second, we calculate perspective transform parameters and determine the player field for tracking. Next, we track their location to find the players’ direction and speed. Since the top player occupies very small region, it is tracked by Kalman-Filter algorithm. In building the object model in real scene, the motion trajectory of the top player is smoothed to prevent error judgment for event detection.
We define six events in tennis games, in which all events start with a player serving a ball. The decision-tree training is based on time, position of the player, and moving distance. Finally, we construct an accurate event detection system that recognizes the events from games based on the motion information.

並列關鍵字

Sport Video ； Tracking, ； Event Detection ； Kalman-Filter algorithm

參考文獻

[1] D.Zhong, S.-F. Chang, “Real-time view recognition and event detection for sports video,” in Journal of Visual Communication and Image Representation, vol. 15, pp. 330-347, 3, Sept., 2004.

[3] J. Han, P. H. N. de With “A unified and Efficient framework for Court-Net Sports Videos Analysis Using 3-D Camera Modeling,” in SPIE Electronic Imaging, vol. 1, p. 6506-15, January 2007, San Jose, USA.

[5] N. Rea, R. Dahyot, A. Kokaram, “Classification and Representation of Semantic Content in Broadcast Tennis Videos,” in IEEE International Conference on Image Processing, vo3, pp. III-1204-7, 11-14, Sept., 2005.

[8] N. Rea, R. Dahyot and A. Kokaram, “Classification and Representation of Semantic Content in Broadcast Tennis Videos,” in IEEE Int. Conference on Image Process., vo3, pp. III–1204–7, 11–14, September 2005.

[9] H. Miyamori and S. I. Iisaku, “Video Annotation for Content-based Retrieval using Human Behavior Analysis and Domain Knowledge,” Proc. of the Fourth IEEE Int. Conference on Automat. Face and Gesture Recognition, pp. 320–325, 2000.