Deep Learning-based Yoga Pose Recognition System using Hyperparameter Tuning
Shahina Anwarul, Manya Mohan · 2022 10th International Conference on Reliability, Infocom Technologies and Optimization (Trends and Future Directions) (ICRITO) · 2022
People have been forgetful of health concerns due to hectic schedules and busy lifestyles. As much as they wish to give priority to healthcare, they aren't able to because it's a choice. The fitness community has been booming in the last few years, with people becoming conscious about their general health. A majority of this community had resorted to Yoga, allowing them to access gateways to earn peace and spirituality. The issue with this would be the accuracy of the Asana being performed. Therefore, an automated deep learning-based yoga pose recognition system is proposed in the present research to accurately recognize the yoga pose performed by the user. The proposed system is divided into three modules: pose detection using MediaPipe, pose recognition using the proposed deep learning model, and alert generation. The intended model is optimized by using the concept of hyperparameter tuning. All the experiments are conducted on a self-created dataset of 5 poses and achieved a 99.6% recognition rate providing competent accuracy with other existing state-of-the-art techniques.