Deep learning techniques for motion recognition of Mae Mai Muay Thai styles

Abstract

Technological advancements and machine learning algorithms have attracted the academic community's interest in human activity detection in recent years. Most notably when discussing the most researched issue over the last decade the automated evaluation of athletic abilities. As a result of the intense rivalry inherent in most sporting events, it is crucial to keep detailed records of each competitor's actions to make fair judgments about their performances. This study aims to provide a technique for automatically recognizing Mae Mai Muay Thai (MMMT) fighting styles using still images and time-stamped boxing sequences. The identification of MMMT styles is handled with deep learning techniques such as Convolutional Neural Network (CNN), a Long Short-Term Memory (LSTM) classifier, and a Long-Term Recurrent Convolution Networks (LRCNs). The MMMT data set was used in a series of experiments with four professional boxers. In addition, we will utilize a confusion matrix to evaluate the model's performance as a whole. The accuracy, precision, recall, and the F1-score were also investigated as performance indicators.