Lawn Tennis Video Summarization based on Audiovisual and Text Feature Analysis

In this paper, a new video summarization approach for lawn tennis video is presented. The proposed method uses frame color histogram to classify video into play field color shots (PFCS) and non play field color shots (NPFCS). Play field color shots are the segments of interest and used to recognize the tournament class. A dominant colored frame from every PFCS is extracted as a salient frame. Our approach also employs dominant value of PFCS volume and energy. For the each PFCS dominant audio energy value is computed and corresponding key frame is extracted. On-screen text presented in play field color shots provides important semantic information so the last frame of the shot is extracted as key caption. Three key frames based on audio, visual and text information conveyed in every PFCS are extracted and summary of lawn tennis video is created. These key frames of every PFCS serve as accurate localization of the events on the lawn tennis field. Experiments are performed on lawn tennis videos to confirm the efficiency of the proposed method. General Terms Shot boundary detection, video summarization