Coupling risk attitude and motion data mining in a preemtive construction safety framework

Abstract

Construction sites comprise constantly moving heterogeneous resources that operate in close proximity of each other. The sporadic nature of field tasks creates an accident prone physical space surrounding workers. Despite efforts to improve site safety using location-aware proximity sensing techniques, major scientific gaps still remain in reliably forecasting impending hazardous scenarios before they occur. In the research presented in this paper, spatiotemporal data of workers and site hazards is fused with a quantifiable model of an individual's attitude toward risk to generate proximity-based safety alerts in real time. In particular, a worker's risk index is formulated and coupled with robust hidden Markov model (HMM)-based trajectory prediction to approximate his/her future position, and detect imminent contact collisions. The designed methodology is explained and assessed using several experiments emulating interactions between site workers and hazards. Preliminary results demonstrate the effectiveness of the designed methods in robustly predicting potential collision events.