Real-Time Umpire Signal Detection in Cricket: A Hybrid Deep Learning Solution

Abstract

The proposed research aims to classify cricket umpire signals based on a four-class machine learning model, with these four key classes: Leg Bye, Four, Wide Ball, and No Ball. Performance assessment was conducted by Precision, Recall, and F1-Score. Excellent classification results have been found in all the classes. The highest precision values were of the "Four" class, which is at 97.68%, while the strongest recall was that of "Leg Bye" at 97.56%. This signifies that the model has an exceptionally good capability of picking out certain signals. F1 scores are very well-balanced in all classes. The maximum score for "Four" was seen at 97.04%. The data used in this research consists of 11,900 images, and the network results in an accuracy of 96.97% in general. Based on the Support and Support Proportion, the accuracy metric has been relatively stable at 98% across all classes; therefore, it indicates that the network has performed trustworthy classification with no visible bias. Thus, these results have established the strong ability of the model in terms of providing the accurate classification of umpire signals and also showed its strength for real-time cricket decision-making purposes. The results suggest that this model can be used to integrate with automated systems in support of umpire decision-making, thus offering greater objectivity and efficiency. Precise classification of umpire signals would improve the accuracy of match decisions, thus benefiting the players, officials, and viewers