Workplace tripping risk identification via millimeter-wave radar-based contactless gait analysis

Introduction : Trips arising from gait instability constitute a leading cause of nonfatal injury on complex construction sites. Reliable detection of pre-trip gait perturbations is essential for mitigating tripping risk. Previous wearable sensor methods require direct contact with workers and may be inconvenient for continuous monitoring, whereas vision-based methods are sensitive to lighting conditions and raise privacy concerns. To address these limitations, this study investigates non-contact millimeter wave (mmw) radar for identifying tripping risk levels in construction. Method: Gait data were collected during controlled tripping experiments in a simulated construction environment across multiple ranges and angles. Tripping risk levels were assigned based on participants’ post-trial perceived instability, with synchronized video records used to verify the corresponding postural responses. The micro-Doppler method was adopted to process raw data and generate time–frequency (T-F) spectrograms. A Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) network was then developed to explore the nonlinear relationship between T-F spectrograms and tripping risk levels. Results: The CNN-LSTM network achieved an accuracy of 95.1% and outperformed the baseline and ablation models, indicating that velocity domain feature extraction and temporal sequence modeling jointly contribute to tripping risk identification. Moreover, mmw radar effectively identified multiple tripping risk levels at different distances and angles, with all evaluation metrics exceeding 90%. Conclusions and Practical Applications: This study facilitates construction site safety management and inspection by providing a convenient, non-contact, and privacy-preserving solution for identifying workplace tripping risk.