Abstract

This work presents a methodology for summarization of sports videos (Cricket)\nusing both audio and visual information. In recent years, there has been massive growth in\nrecorded audio-visual content and on an average people are spending most of their time\nwatching long sports videos. It will be useful if we can produce a glimpse of the video, by\ngenerating the summary of it. In this paper, we propose audio-video processing based\napproach for generating the summary of sports videos. The original video is divided into\ntwo processing tracks: video track and audio track. In the video track, video frames\nfeatures are extracted using the convolutional neural network (CNN) and temporal\nsequences or scores are generated that represent how important the particular frame is. The\nthreshold is applied and the sequences are learned using Long-short term memory (LSTM).\nIn the audio track, audio samples are read in the interval of the sampling rate and\nintensities are calculated. The threshold is set and an array of seconds are aggregated which\nis aligned with the output of video track. After alignment, the summary is generated. The\nCNN trained with 74 videos of cricket sports actions. Our approach achieved much\naccuracy compare to previous methods.