Vigilant Air Traffic Control: Gaze-based Recognition of Detection Failures to Visual Warnings

Abstract

Air traffic controllers sometimes fail to detect visual warnings due to limited attention resources. This challenge would even be exacerbated by the increasing complexity of visual data in future digital tower integrations. Detection failures (DF) manifest in three primary types: ordinary blindness (OB), look but fail to see (LBFTS) error, and misinterpretation (MI), each resulting from disruptions in the detection process stages and necessitating specific countermeasures. This study employs machine learning and eye-tracking in a simulated air traffic control (ATC) environment to identify and differentiate types of DF. Eye movements of 26 participants were tracked across 108 OB, 109 LBFTS, and 95 MI instances to ATC warnings. Seven machine learning models, including three basic and four advanced tree-based models, were assessed for DF recognition. Results found that the gradient boosting decision tree exhibited superior performance with 74% accuracy in four-detection-type recognition, particularly in recognizing OB and LBFTS. Additionally, correct detection and MI are more challenging but still effectively recognized, with correct detection better identified by k-Nearest Neighbour, and MI by light gradient boosting machine. These findings demonstrate the feasibility of real-time gaze-based DF recognition in ATC and offer valuable insights for ATC management in enhancing visual warning detection and aviation safety.

Keywords: Warning detection, detection failure recognition, eye-tracking, air traffic control, detection failure types

DOI: 10.54941/ahfe1005190