Real-time posture monitoring prediction for mitigating sedentary health risks using deep learning techniques

Abstract

Sedentary behavior has become a pressing global public health issue. This study introduces an innovative method for monitoring and addressing posture changes during inactivity, offering real-time feedback to individuals. Unlike our prior research, which focused on post-analysis, this approach emphasizes real-time monitoring of upper body posture, including hands, shoulders, and head positioning. Image capture techniques document sedentary postures, followed by preprocessing with bandpass filters and morphological operations such as dilation, erosion, and opening to enhance image quality. Texture feature extraction is employed for comprehensive analysis, and deep neural networks (DNN) are used for precise predictions. A key innovation is a feedback system that alerts individuals through an alarm, enabling immediate posture adjustments. Implemented in MATLAB, the method achieved accuracy, sensitivity, and specificity rates of 98.2%, 90.7%, and 99.2%, respectively. Comparative analysis with established methods, including support vector machine (SVM), random forest, and K-nearest neighbors (KNN), demonstrate the superiority of our approach in accuracy and performance metrics. This real-time intervention strategy has the potential to mitigate the adverse effects of sedentary behavior, reducing risks associated with cardiovascular and musculoskeletal diseases. By providing immediate corrective feedback, the proposed system addresses a critical gap in sedentary behavior research and offers a practical solution for improving public health outcomes.