Artificial intelligence-enabled non-intrusive vigilance assessment approach to reducing traffic controller's human errors

Abstract

To be vigilant is highly required for traffic controllers in transportation fields, such as air traffic management, vessel traffic service, and railway management, as they need to monitor traffic conditions and notice any potential hazards. Hence, emerging studies have been conducted to develop an objective and non-intrusive approach to assessing vigilance levels and generate warnings if needed. This study aims to investigate the effects of impaired vigilance on human performance via non-intrusive data analysis, namely spatial and temporal gaze pattern analytics, and develop an objective model for vigilance assessment accordingly. A novel four-phase framework, including vigilance test design, non-intrusive data collection, spatial and temporal gaze pattern analytics, and a shallow neural network-based model was proposed to achieve this aim. Meanwhile, an illustrative experiment in the maritime industry was conducted to verify the proposed method. The spatial and temporal gaze patterns analytics revealed that low vigilance levels impacted comprehension time but not perception time, with longer fixations duration but stable time-to-the-nearest-fixation under a low vigilance level. It is found that even a person with impaired vigilance can quickly notice abnormal events. The effectiveness and empirical implications of this model can help traffic controllers avoid fatigue-induced vigilance reduction. In addition, it provides evidence, references, and solutions for designing human–computer interfaces to reduce human errors caused by low vigilance.