Algorithms, Experimentation.pdf

A Robust Scene Recognition System for Baseball Broadcast Using Data-Driven Approach Ryoichi Ando, Koichi Shinoda, Sadaoki Furui Department of Computer Science Tokyo Institute of Technology 2-12-1 Ookayama, Meguro-ku Tokyo 152-8552 Japan ando@ks.cs.titech.ac.jp {shinoda, furui}@cs.titech.ac.jp Takahiro Mochizuki NHK Science Technical Research Laboratories 1-10-11 Kinuta, Setagaya-ku Tokyo 157-8510 Japan mochizuki.t-fm@nhk.or.jp ABSTRACT We propose a robust scene recognition system for baseball broad- cast videos. This system is based on the data-driven approach which has been successful in continuous speech recognition. It uses a multi-stream hidden Markov model to model each scene and an unsupervised adaptation method to achieve robustness against differences in environmental conditions among games. It also em- ploys an n-gram language model to represent the contexts among scenes, and a model for scene length information. The proposed system was evaluated in scene recognition experiments with 16 scene types acquired from video data of 25 baseball games. The system reduced errors in scene recognition by 6.3 % absolute. Categories and Subject Descriptors I.2.10 [Vision and Scene Understanding]: Video analysis; I.4.8 [Scene Analysis]: Time-varying imagery; H.2.4 [System]: Multi- media databases General Terms Algorithms, Experimentation Keywords CBVIR, sports video, indexing, HMM, n-gram model, scene con- text, adaptation 1. INTRODUCTION Recent advances in computer technology, particularly in storage technology, have resulted in significant increases in the number and quality of video databases. As it has become difficult for ordi- nary people to browse the entire content of each video database, database indexing is strongly required for searching and summa- rization. The construction of such indexes is mostly carried out by Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies ar