以人工智慧策略偵測網球節目精彩片段

Using Artificial Intelligent Strategies to Detect Tennis Sport Highlights

邱慶麟(Ching-Lin Chiou)

指導教授 ： 黃有評

大同大學/電機資訊學院/資訊工程學系所/碩士(2007年)

摘要

近年來，由於運動比賽節目越來越普遍，一般大眾或運動專業工作者隨時可利用各式各樣的數位錄影工具將其錄製下來，並擷取有用的資訊或是精彩的片段，以便閒暇再觀看，使得運動影片分析及萃取之研究開始引人注意。其中又以探討精彩片段分析及萃取方法之研究為主，主因係運動影片結構較其他類型的影片具有一定規則，另一原因為精彩片段可協助群眾快速掌握重要運動資訊，如棒球比賽的全壘打、足球比賽的進球，以及網球比賽中雙方對壘等場景萃取，更是學界及業界積極投入研究的方向。
其中，與網球運動節目相關的研究雖已有學者投入，但為支援如此快速且大量資料的產生，以及現有影像索引技術仍待開發，並且在缺乏任何標準的情況之下，本研究試著從中找出網球節目的主要規則性，萃取出其重要事件發生位置，進而定位出精彩片段可能位置，希能為後續從事此方面研究者提供可用方法作為研究參考依據。
本研究將以網球運動節目為主，透過人工智慧方法結合聲音及影像的特徵，進行各項不同方法萃取的實驗，試著建構出可以精確擷取精彩片段的模型，並將網球運動節目中多餘的片段剔除，利用聲音的強度及選手對打的時間長度分別將精彩片段作排序，此將有效提供未來在網球訓練、新聞剪輯、運動預告片段等運用。
經實驗結果顯示，本研究所建構的模型的精確度及召回率平均皆超過89%，證明此模型在萃取網球節目的精彩片段上，具有不錯的成效，此亦證明我們所提出的方法是可行的。

關鍵字

精確率 ； 精彩片段 ； 網球運動節目 ； 人工智慧 ； 召回率

並列摘要

In recent years, due to the popularity of sport activities, general masses or sport professionals can utilize various video recording tools to record, produce, pick and fetch useful information or interesting programs. Sport video analyses have attracted attention gradually. Sport video has an inherent structure as defined in rules of the game and field production. In addition, some highlights in the sports game such as homerun in the baseball, shot in the football and ace in the tennis make the sport video more suitable for further investigation than other types of video such as the films and news.
Tennis sport video is selected as the primary domain due to its content richness and popularity. With an ongoing rapid growth of sport video information, there is an emerging demand for a sophisticated tennis content-based video indexing system. However, current video indexing solutions are still immature and lack of any standard.
Artificial intelligent strategies are proposed to combine characteristics in the audio and image domain and knowledge to find the highlight. Moreover, broadcasted sports videos generally last several hours with many redundant advertisements and the key segments are not easy to find. The proposed model can be extended to different application domains, such as tennis acrobatics training, exciting news editing, and program preview in the future.
Based on the experimental results, both the average of precision and recall rates are higher than 89%. This verifies that the proposed model is effective in extracting the highlights of tennis sport.

並列關鍵字

Precision ； Highlight ； Sport video ； Artificial Intelligence ； Recall

參考文獻

[1]N. Babaguchi, Y. Kawai, T. Ougura, and T. Kitahashi, “Personalized abstraction of broadcasted American football video by highlight selection,” IEEE Transactions on Multimedia, vol. 6, no. 4, pp.575-586, August 2004.

[2]C.L. Huang, H.C. Shih, and C.Y. Chao, “Semantic analysis of soccer video programs using Dynamic Bayesian,” IEEE Transactions on Multimedia, vol. 8, no. 4, pp.739-760, August 2006.

[3]D. Tjondronegoro and Y.P.P. Chen, “Integrating highlights for more complete sports video summarization,” IEEE Transactions on Multimedia, vol. 11, pp.22-37, October-December 2004.

[5]F. Coldefy, P. Bouthemy, M. Betser, and G. Gravier, “Tennis video abstraction from audio and visual cues,” IEEE 6th Workshop on Multimedia Signal Processing, Siena, Italy, pp.163-166, September-October 2004.

[7]L. Rabiner and R. Schafer, Digital Processing of Speech Signals, Prentice-Hall, Englewood Cliffs, NJ, USA, 1978.