Deep Learning Models for Recognition of Hand Gestures in Basketball Sports
Rinki Gupta, Varun Kavanal, Anya Joshi · 2024
In the domain of sports applications, the recognition of hand motions holds significant importance. Leveraging the capabilities of deep learning, this paper is dedicated to devising robust solutions for identifying hand movements of referees in basketball sport. The dataset under consideration encapsulates multiple inputs, consisting of data from surface electromyography (sEMG) sensors and accelerometers. These sensors play a pivotal role in capturing muscle activity and three-dimensional motion information of hand gestures during sports activities. This paper presents a trifecta of Convolutional Neural Network (CNN) and long short-term memory (LSTM) models, each meticulously designed to accommodate specific input data categories. The proposed CNN-LSTM models optimized with just accelerometer, just EMG, as well as multi-channel EMG and accelerometer data yield accuracies of 94.4%, 89.8%, and 94.7%, respectively. Our results unequivocally underline the efficacy of the combined model, which adeptly fuses information from EMG and accelerometer data. This implementation of deep learning methods augments hand motion recognition in the context of basketball refereeing, signifying a significant stride towards streamlined human-motion interaction and more proficient officiating in basketball games.