Automatic video genre categorization and event detection techniques on large-scale sports data

  • Authors:

  • Ning Zhang;Ling-Yu Duan;Qingming Huang;Lingfang Li;Wen Gao;Ling Guan

  • Affiliations:

  • Ryerson University, Toronto, ON, Canada;Peking University, Beijing, P.R. China;Chinese Academy of Sciences, Beijing, China and Graduate University of Chinese Academy of Sciences, Beijing, P.R. China;Chinese Academy of Sciences, Beijing, China and Graduate University of Chinese Academy of Sciences, Beijing, P.R. China;Peking University, Beijing, P.R. China;Ryerson University, Toronto, ON, Canada

  • Venue:
  • Proceedings of the 2010 Conference of the Center for Advanced Studies on Collaborative Research
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper presents an efficient and robust automatic process for large-scale sports video analysis. The proposed system firstly identifies the genre of the query video, and then accomplishes the interesting event detection task. The significance of this framework is its automatic characteristic in testing with minimum human involvement in training, as well as the scalability and expansibility in dealing with a large-scale dataset. Domain-knowledge independent local features are extracted from an input video sequence and a histogram based distribution representation is created using the bag-of-visual-words (BoW) model. In genre categorization, k-nearest neighbor (k-NN) classifiers with various dissimilarity measures are assessed and evaluated analytically. For the event detection, a hidden conditional random field (HCRF) structured prediction model is utilized. Overall, this framework demonstrates the efficiency and accuracy in processing voluminous data from sports collection and achieves various tasks in video analysis. It also demonstrates a potential technology transformation from the "laboratory bench" to commercial applications.