A Multimodal Approach for Automatic Cricket Video Summarization

Summarizing cricket matches is a labor intensive task that demands a certain level of proficiency. The paper proposes a novel method for automatically detecting and summarizing important events in a cricket match. Our model takes into input the video recording of the entire cricket match and returns the most important clips of the game as the output. Techniques like optical character recognition, sound detection and replay detection have been used to extract important events such as boundaries, wickets and other playfield scenarios in a cricket match. These events are then clipped together to form the entire highlight for the cricket match. We performed several qualitative and quantitative experiments to evaluate our model. Our model achieves an accuracy of 89.45% to detect events like wickets, fours and sixes which indicates the usefulness of the model in real life scenarios.