Sedentary Posture Recognition and Correction Using a Convolutional Neural Network (CNN) and the You Only Look Once Version 8 (YOLOv8) Pose Estimation Model

Abstract

Poor posture is a leading contributor to musculoskeletal disorders, significantly affecting quality of life and productivity. This project introduces a deep learning framework to identify anatomical keypoints and offers a system that classifies seated posture as good, fair, or bad while providing user posture-related feedback. Initially, a custom Convolutional Neural Network (CNN) was developed with 47.3% accuracy, but due to practical constraints, the system was integrated with the You Only Look Once Version 8 (YOLOv8) pose mode with 84.9% accuracy. This system operates through a phone camera connected to a main device, achieving a posture detection accuracy of 92.3% at 30 Frames per Second (FPS). With broad applications, such as workplace ergonomics, remote learning, and online physical therapy, this project proposes a non-invasive solution for proactive posture correction.