Cricket Event Recognition and Classification from Umpire Action Gestures using Convolutional Neural Network

Abstract

The advancement of hardware and deep learning technologies has made it possible to apply these technologies to a variety of fields. A deep learning architecture, the Convolutional Neural Network (CNN), revolutionized the field of computer vision. One of the most popular applications of computer vision is in sports. There are different types of events in cricket, which makes it a complex game. This task introduces a new dataset called SNWOLF for detecting Umpire postures and categorizing events in cricket match. The proposed dataset will be a preliminary help, it was assessed in system development for the automatic generation of highlights from cricket sport. When it comes to cricket, the umpire has the authority to make crucial decisions about on-field incidents. The referee signals important incidents with hand signals and gestures that are one-of-a-kind. Based on detecting the referee's stance from the cricket video referee action frame, it identifies most frequently used events classification: SIX, NO BALL, WIDE, OUT, LEG BYE, and FOUR. The proposed method utilizes Convolutional Neural Networks (CNNs) architecture to extract features and classify identified frames into Umpire postures of six event classes. Here created a completely new dataset of 1040 images of Umpire Action Images containing these six events. Our method train CNNs classifier on 80% images of SNWOLF dataset and tested on 20% of remaining images. Our approach achieves an average overall accuracy of 98.20% and converges on very low cross-entropy losses. The proposed system is a influential answer for generation of cricket sport highlights