Correction and Estimation of Workout Postures with Pose Estimation using AI
Tanmay Hande, Bhargavi Kakirwar, A. V. Bharadwaja, Pravin Ramdas Kshirsagar, Aaditya Gupta, P. Vijayakumar · 2023
Strength workouts are effective and popular ways to achieve health benefits, but they can cause injury for newcomers if performed incorrectly without prior knowledge. The purpose of this research is to examine how the recent developments in pose recognition, estimation and correction can be used to estimate workout postures and provide valuable feedback on workout techniques to detect specific technique issues associated with a high risk of injury for common exercises. To provide a user with feedback, action recognition will be responsible for collecting, labeling, and organizing the data, as well as training and integrating with real-time data. Using a validation dataset of 218 workout images from the Penn Action dataset as a validation set, our best model scored 97.25 % accurate. The results of this research proves that the pose recognition, estimation and correction algorithm is accurate and can yield useful feedbacks when it comes to estimate workout techniques.