Ready-to-Serve Detection in Badminton Videos

Video analysis has gained significant importance as a valuable tool for coaches and players to analyze their matches. In a recorded badminton video, the service marks the commencement of a point which can lead to the segment of points won and lost in a game. This paper presents a proposed method that utilizes court details and poses to accurately identify readyto-serve action from the other actions in the video. The proposed method uses nineteen features extracted from a video frame and trained by neural networks to classify them as ready-to-serve and non-serve. In this study, we analyze the classification accuracies of a three-layered neural network and a convolutional neural network. The extracted features from twelve frames arranged as a 1-D input is given to a three-layered neural network (NN) while they are arranged in a 2-D format for a convolutional neural network (CNN). Two datasets consisting of video frames of topranked badminton players executing ready-to-serve shots and other typical actions such as jump smash, netplay, defensive stroke, and strokes from the backcourt were used in this study. The training data came from one dataset only and the remaining data was used for testing. The second dataset was used for testing only. The testing results using both datasets show high accuracies in classifying ready-to-serve and non-serve strokes by the CNN.