Risk Behavior-Based Trajectory Prediction for Construction Site Safety Monitoring

Abstract

Construction sites are often described as some of the most hazardous work environments due to the unscripted nature of tasks which puts workers and equipment in close proximity, potentially resulting in near-miss situations or life-threatening contact collisions. Previous research has investigated location-aware methods to improve construction safety but has mostly fallen short in exploring the extent to which prediction techniques can be used to model and formulate the role and attributes of individual workers in addition to the physical characteristics of the jobsite that may lead to safety incidents. This paper studies the feasibility of a preemptive proximity-based safety framework by investigating two trajectory prediction models, namely polynomial regression (PR) and hidden Markov model (HMM). The HMM prediction is further calibrated by factoring in a worker’s risk profile, which is a measure of his or her affinity for or aversion to risky behavior near hazards. The method is tested in a series of field experiments involving trajectories of different shapes and complexity. Results demonstrate that the developed methodology can reliably detect unsafe movements and impending collision events.