查看文章

Purpose: Pedestrian-vehicle interface risks in manufacturing work zones result in occupational injuries and fatalities. Risk assessment is currently limited to safety audits and heuristic analysis. A data-driven alternative is desirable since it will capture all risky events and result in more effective countermeasures.

Design/methodology/approach: The camera-based system uses PCA-based shape descriptor as a feature vector for pedestrians and vehicles in the work area. Supervised classification using Support Vector Machines (SVMs) is employed to detect entities and track movements. All interfaces are automatically observed, leading to an estimate of Risk Prioritization Number (RPN). RPN is the industry-standard risk assessment metric.

Findings: Experimental data was collected using a sample work area layout and scale versions of vehicles. Entity interfaces and movements were physically simulated to train and test the machine learning model. High detection accuracy, precision, and recall were observed. The automated estimates were in close agreement with human annotation.