ABSTRACT
In today's post-covid culture, where everyone works from home, there is a huge possibility of serious long-term health problems. A lot of people have started taking up exercises at home and if done incorrectly, they can have major negative effects. Another one of the main contributors to these health issues is bad sitting posture, which is only exacerbated when working for hours on end. Hand gesture detection has many useful applications in elderly healthcare, automating actions and gesture-based presentations and games. To help users with these actions, our paper proposes pinpointing the points of the error to the user in real-time and in a lightweight manner for yoga posture correction. The incorrect positions shall be shown in real-time on top of the user's video feed to help them correct it properly. The user shall be told about when they are sitting in a bad position, and the overall bad posture time will also be shown for the session, which will provide the required information to the user. To further help users in a useful manner, our paper looks to augment the hand gesture detection feature with federated learning and personalization to avoid the common pitfall of privacy concerns, while still allowing users to customize their experience. The proposed library for the implementation of these tasks is the MediaPipe library. This library is one of the key components that makes the features lightweight and easy to use. The aforementioned library also looks to implement the features in real time with no lag while keeping the resource requirements as low as possible. © 2023 IEEE.