Baseball Pitch Recognition for Broadcast Television

• Main Authors: Chia-chang Li, 李家昶
• Other Authors: Chia-Wen Lin
• Format: Others
• Language: en_US
• Online Access: http://ndltd.ncl.edu.tw/handle/31426415780361850397

碩士 === 國立中正大學 === 資訊工程所 === 95 === Advances in the multimedia and entertainment industries, including streaming audio and digital TV, present new challenges for managing and accessing large audio-visual collections. Sport broadcasts constitute a major percentage in the total of public and commercial television broadcasts. The growing demands of the viewers require advances in video capturing, storage, delivery and video processing ability. This paper aims to recognize the baseball pitches from the baseball game video captured from broadcast television. We proposed a novel baseball pitch recognition approach which analyzes the trajectories in their spatial and temporal domain. According to our observations, each pitch has its own characteristics in speed, trajectory, acceleration and shape. We model the temporal information for each pitch by using Hidden Markov Models, and use the fundamental statistical approach by using the features acceleration and the shape of the trajectory to improve the precision. The experimental results on 195 trajectories in 6 baseball games confirm that the proposed framework is very effective and reliable and the precision of each pitch is more than 0.8.