ErgoSight: Enhancing Posture and Eye Wellness (Monitor Posture and Eye Health Using Machine Learning)

Abstract

"With the advancement of technology and prevailed COVID-19 pandemic, prolong computer usage among individual’s despite of their age has significantly increased, leading to concerns related to postured and eye strain. The research problem addressed by this project is the prevalence of poor posture and eye strain resulting from prolonged computer usage, later may cause in major health concerns.

In the study, the author integrates principles from Computer Vision, Image Processing, and Machine Learning to introduce a novel approach. The eye strain detection model is trained using a well-known Facial Expression Recognition (FER) algorithm: Convolutional Neural Network (CNN). Each layer is fully connected for feature extraction and classification. Regularization techniques such as dropout, batch normalization, and data augmentation are applied to prevent overfitting and improve generalization. For posture monitoring, author has utilized a pre-trained model, Mediapipe, for landmark extraction for posture deviation calculation.

Initial test results demonstrate satisfactory outcomes, with the initial trained model achieving a 71% accuracy rate and the fine-tuned model achieving 75% accuracy in emotion detection for eye strain detection. Real-time webcam-based testing further validates the efficacy of ErgoSight in detecting eye strain in real time. Mediapipe landmark extraction yields satisfactory outcomes in posture deviation calculations. This suggests that the proposed ErgoSight system achieves significant performance by utilizing a novel approach in eye strain detection, integrating emotion and posture detection from landmark deviation calculation."