Cricket Video Summarization Using Deep Learning

Abstract

Video summarising is the process of choosing and presenting the most relevant or fascinating components from a longer video document in order to create a brief summary for potential consumers. Cricket is a highly regulated sport played for quite a longer duration than most other sports. This study presents a paradigm for recognising and clipping crucial occurrences in a cricket match that takes into account event-based attributes. Cues used to capture such moments include replays, audio intensity, player celebrations, and playfield scenarios. This project focuses on building a model which will summarise the original video into three types of summaries. Different Deep Learning models and Natural Language Processing techniques are used to build the model. Key frame extraction is a crucial task that is done to generate the cricket highlights. The project's purpose is to show how to extract a high-quality summary out of an original video using a high-performance approach. This research presents a robust video summarization model for the cricket videos that is able to generate high-quality cricket highlight videos along with textual and audio summaries based on the user's preferences. The focus is on the automation of video summarization process and provide a better quality of summary to the user. From the comparison of the generated highlights to the actual existing highlights manually, it was concluded that the methodology adopted gave a satisfactory result covering almost 80-90% of the events and captions with Bilingual evaluation understudy score-4 of 0.748953.