Content-Aware Summarization of Broadcast Sports Videos: An Audio–Visual Feature Extraction Approach

A large number of videos available on the internet belong to the category of sports. Generally, a sports video has a long duration and consists of only a few exciting moments. Sports enthusiasts keep themselves updated on the current happenings, in less time, by means of a summarized version of the sports video known as highlights. For the past few years, sports video summarization is regaining attention among the research community. Automatic generation of highlights form a sports video is a challenging task as different sports games have different rules and situations. In this paper, we propose a method for automatically generating highlights from broadcast sports videos. The proposed method generates highlights by extracting audio and visual features from a sports video. Our method automatically learns the scorebox template from a broadcast sports video using SIFT features, and then locates and extracts the template from a video stream. The extracted template is further analyzed to find out all the possible text regions. Afterward, the information is extracted from all the text regions by means of deep neural network. Based on user preferences, the most relevant information is extracted and converted to a keyframe representation which helps to generate highlights. Extensive experiments were performed to evaluate the effectiveness of the proposed method. Results of the experiments reveal the effectiveness and superiority of the proposed method.