Intelligent Posture Assessment Using Machine Learning

Abstract

Poor sitting posture is a common cause of musculoskeletal disorders, especially among individuals who spend prolonged hours working at desks or using computers. To address this, we propose an intelligent, real-time posture assessment system that leverages computer vision and machine learning techniques. The system utilizes MediaPipe, a lightweight and robust pose estimation framework, to extract key body landmarks from a live webcam feed. These landmarks are processed to extract meaningful posture-related features, which are then classified using a Random Forest model to determine whether the posture is correct or incorrect. The system is capable of providing immediate visual and audio feedback to the user upon detection of poor posture. Experimental results show that the model achieves an accuracy of over 95% on a custom dataset, demonstrating its potential as a cost- effective, non-intrusive solution for posture monitoring in workplace and educational environments.