Klasifikasi Postur Duduk Menggunakan Pre-Trained Residual Network 50 V2 (Resnet50-V2) pada Pengguna Komputer/Laptop

Abstract

Most computer/laptop users spend hours, even up to entire days, sitting in front of their screens, whether for work, gaming, completing tasks, browsing, or more. Prolonged sitting in front of a computer/laptop leads individuals to focus their vision and thoughts on their screen activities, often unintentionally adopting poor posture. Poor sitting posture over extended periods can lead to bone and joint health issues, especially Musculoskeletal Disorders (MSD). Problems affecting bones and joints can disrupt daily activities and have potentially serious consequences. Given the current Computer Vision technology, measuring risk variables is challenging, thus, this research employs images of users’ sitting postures to assess these risks. This study develops a system capable of classifying sitting postures based on neck and upper body position categories, implemented in an application for real-time monitoring of users' sitting postures. The system identifies four posture categories: Leaning Right, Leaning Left, Normal Sitting, and Risky Middle Sitting Posture. The research utilized a dataset of 2660 instances, with 2128 data points for Training, 264 for Validation, and 268 for Testing. Following testing procedures, the study achieved an accuracy of 94.77%.