Event Detection in Cricket

Abstract

Cricket is the second most popular game in the world and this paper aims to automate the process of generating text-based commentary in live cricket matches using video analysis. For improving the amount of detail in the generated text-based commentary to enable the reader to visualize the sequence of events that occur during each delivery as good as observing the match, we propose a solution that successfully detects multiple critical events in a single delivery, such as ball release, bounce point, ball contact by the batter, and the direction of the ball following the hit. To achieve this, we apply various deep learning and computer vision techniques, including object detection, image classification, transfer learning, and optical flow. Our solution leverages neural network architectures such as Convolutional Neural Networks (CNN), and You-Only-Look-Once (YOLO) to train multiple deep learning models, including ball detector, batter detector, bowler detector, batter orientation classifier, hit detector, and scene detector. Our models are trained on a custom dataset, resulting in accurate and efficient event detection during live cricket matches.