The Federated Fielder: CNNs and the Future of Cricket Umpire Analysis

• Published in: 2024 Asia Pacific Conference on Innovation in Technology (APCIT) pp. 1 - 6
• Main Authors: Suryavanshi, Ankita, Mehta, Shiva, Chaudhary, Preeti, Joshi, Kireet, Jain, Vishal
• Format: Conference Proceeding
• Language: English
• Published: IEEE 26.07.2024
• Subjects: (CNN)_(FL); Accuracy; Analytical models; Augmentation image; Cricket umpire Action; Data privacy; data security; Federated learning; Recognition; Solid modeling; Solids; Technological innovation; Umpire
• DOI: 10.1109/APCIT62007.2024.10673692

As we move into a landscape where sports analysis is becoming more critical, the demand for action recognition systems that are real-time and high-accuracy is also increasing. It is vital in complex cricket games where umpire signals have taken center stage in decision-making. This research suggests a new way of discovering the invisible actions of cricket umpires. The method of FLAN associated with CNNs for the confident detection of two types of umpire signals will enable strides in automating umpiring in tennis. The very point of our solution is that five different clients use it. Each client gives us information about its climate characteristics that can be used in a compact universal model that does not need to be updated after every client response. The application of this option preserves the confidentiality of data that must be privately protected. We give a detailed breakdown of the aggregated outcomes, using macro, Micro, and weighted averages as an example, by collecting local data insights and improving the overall model. In the presence of five direct customers, the on-site trial resulted in macro averages varying between 83.99% and 90.31%. The same study also produced micro averages ranging from 83.99% to 90.28%, and weighted averages tended to match macro averages. This connotes the reliability and dependability of our system. Throughout the development, federated averaging became a solid method for enhancing a global dataset through local discrepancies of input data for the improvement of model accuracy altogether with secure privacy.