Detecting loss of balance in construction workers using millimeter wave sensing

Slip, trip, and fall (STF) events remain the primary cause of reported injuries in the construction industry, highlighting the crucial necessity of efficient monitoring systems. Conventional monitoring approaches, whether based on wearable devices or vision-based methods, present significant privacy concerns and also suffer from complex deployment requirements and illumination-dependent constraints. This research presents a novel noninvasive approach utilizing millimeter-wave (mmWave) radar technology to identify and assess loss of balance (LOB) events among construction workers. The proposed approach employs micro-Doppler signatures obtained by mmWave sensing, producing time-frequency spectrograms that are investigated using a customized ResNet-based deep learning model. Thorough experimental validation was performed in simulated construction settings, encompassing data collected across multiple distances and angles. The results demonstrated robust performance metrics, achieving 94.58% accuracy in distinguishing STF from non-STF events, with both precision and recall rates exceeding 95% for STF incident detection. It achieved an overall F1 score of 90.27% in identifying specific LOB event categories, indicating consistent and adequate performance across various scenarios. This approach provides a practical, weather-resistant, and privacy-preserving solution for early hazard intervention in construction safety management.