Pose Recognition in Cricket using Keypoints

In the present-day time, there has been a gain in interest in video summarization and highlights generation in football, cricket, basketball, and baseball. Some pose recognition methods for recognizing the pose of an umpire in cricket have been proposed, but none of them leverage the potential of pose estimation and neural networks, which are two of the most powerful tools in Deep learning. In this paper, we work on the dataset termed SNOW, for the detection of umpire pose in the game of cricket. This dataset has been assessed as an introductory aid for pose recognition of the umpire in cricket. The umpire in cricket has the power to give decisions, and these decisions are conveyed using hand signals. On the basis of identifying the umpire's pose from the frames of a cricket video, we try to identify five such signals: NO BALL, SIX, WIDE, OUT, and NO ACTION. This paper discusses a technique for recognition of the gestures and poses of the umpire using keypoints generated using pose estimation. The experimental results show that the accuracy of our proposed technique is 87%, and the evaluation metrics of our technique are quite promising compared to existing state-of-the-art works.