Preliminary Assessment of a Low-Sampling-Rate Wearable Head-Mounted Inertial Sensor System for Human Activity Recognition

Human Activity Recognition (HAR) plays a crucial role in occupational safety and worker monitoring, particularly in high-risk environments where real-time activity classification can support risk assessment and injury prevention. Traditionally, HAR systems rely on high-frequency data acquisition, requiring substantial power and computational resources that can limit their integration into wearable safety equipment. This study explores the feasibility of HAR using a lowsampling rate head-mounted inertial sensor system, evaluating its effectiveness in continuous worker activity monitoring without introducing additional burdens or interfering with personal protective equipment (PPE). A wearable system was designed and tested under controlled laboratory conditions with five volunteer workers performing four standard activities: walking, standing, climbing stairs, and running. The accelerometer data were processed using feature extraction in both time and frequency domains, followed by classification with a Random Forest (RF) model. The system achieved an overall accuracy of 95.42% and an F 1-score of 0.95, demonstrating that reliable HAR can still be achieved.