AutoPose: Pose Estimation for Prevention of Musculoskeletal Disorders Using LSTM

Office work has become the most prevalent occupation in contemporary society, necessitating long hours of sedentary behavior that can lead to mental and physical fatigue, including the risk of developing musculoskeletal disorders (MSDs). To address this issue, we have proposed an innovative system that utilizes the NAO robot for posture alerts and camera for image capture, YoloV7 for landmark extraction, and an LSTM recurrent network for posture prediction. Although our model has shown promise, further improvements can be made, particularly by enhancing the dataset’s robustness. With a more comprehensive and diverse dataset, we anticipate a significant enhancement in the model’s performance. In our evaluation, the model achieved an accuracy of 67%, precision of 44%, recall of 67%, and an F1 score of 53%. These metrics provide valuable insights into the system’s effectiveness and highlight the areas where further refinements can be implemented. By refining the model and leveraging a more extensive dataset, we aim to enhance the accuracy and precision of bad posture detection, thereby empowering office workers to adopt healthier postural habits and reduce the risk of developing MSDs.