Predicting roadway workers' safety behaviour in short-term work zones

Towards a more detailed understanding of how roadway workers' instinctive reactions (e.g., moving their entire body, turning their head) to traffic contribute to fatal work zone crashes, there is a need for accurate models of workers' behaviour in these dangerous scenarios. While related studies address similar challenges in vertical building construction sites, this study proposes a deep learning model for worker behaviour in horizontal roadway work zones, building on a previously developed wearable sensor and virtual reality (VR) platform designed to capture safety related behavioural data (e.g., head position and orientation, field of view) on workers exposed to hazardous traffic vehicles as they experience immersive simulations of real-world roadway work zones. Using gated recurrent units (GRUs) trained with behavioural data collected from the platform, the deep learning model's accuracy (i.e., average displacement error) in predicting a worker's future position trajectory will be evaluated.