Harnessing Computer Vision and Deep Learning Model for Optimal Sitting Posture Detection

Sitting is a basic action and resting position in which the body weight is supported primarily by lower parts of the body that are in contact with the ground or a horizontal surface such as a chair seat. Poor sitting posture can damage the spine health for a long period. Thus, maintaining a healthy sitting posture is crucial for a person who needs to sit for a long time. By detecting the sitting posture of a person able to repair and warn the bad sitting posture. This project proposed the use of computer vision to develop a human skeleton model using MediaPipe Pose by plotting the landmarks on the joint point throughout the body that is called the keypoint. The pipeline's posture estimation component predicts the location of all 33-person keypoints with three degrees of freedom. These keypoints will be used to calculate body angle and classify the best sitting posture. This project able to produce overall precision of 92.5% for straight sitting posture and recall result that achieve 84.091% in real-time data and image. The overall accuracy for this project is 81.739%.