A Robust Development of Human Posture Recognition Model by using Artificial Intelligence based Learning Scheme

Fall risk prediction is a critical aspect of healthcare, aiming to identify individuals at risk and prevent injuries through early intervention. This study introduces a robust human posture recognition framework utilizing a novel deep learning architecture, PoseGaitNet, for fall risk prediction. Using the MMU Fall Risk Prediction (MMU- FRiP) dataset, the proposed model leverages markerless motion capture and Human Posture Estimation (HPE) to analyze gait features. Advanced preprocessing techniques, including temporal scaling and Gaussian noise addition, enhance data variability, while a custom feature selection mechanism refines the input for optimal model performance. Comparative evaluations with nine baseline models reveal the superior accuracy of the proposed model (98.50%), outperforming prominent architectures such as SE-ResNet (97.30%) and Xception (96.10%). In addition to achieving state-of-the-art accuracy, PoseGaitNet demonstrates reduced computation time, making it suitable for real-time applications. This work provides a significant contribution to fall risk assessment, combining high predictive accuracy with computational efficiency.