Schedule of Events | Search Abstracts | Invited Symposia | Symposia | Poster Sessions | Data Blitz

Neural and Behavioural Time Courses of Rapid Danger Detection

Poster Session C - Sunday, March 8, 2026, 5:00 – 7:00 pm PDT, Fairview/Kitsilano Ballrooms

Kathleen Botha¹ (kathleen.a.botha@gmail.com), Freya Chaytor¹, Evelyn Young¹, Dominika Prihodova², Adrian Bartolome¹, Hee Yeon Im^1,3; ¹University of British Columbia, ²University of Edinburgh, ³BC Children's Hospital Research Institute

To ensure safety, we must identify danger instantaneously. Prior research on gist perception demonstrates that humans can recognize basic scene properties within tens of milliseconds. This study examines the time course of danger detection to determine whether danger can be extracted as semantic gist from complex workplace scenes to rapidly engage the brain’s attention and action systems. Using ChatGPT-4.0, we generated 186 dangerous images depicting real-world occupational accident scenarios and an additional 186 nondangerous images. Low-level visual features, scene backgrounds, and human body poses were matched across dangerous and non-dangerous image pairs to minimize the influence of salience-based visual processing. Images were presented for 14, 42, or 126 msec, followed by a mask stimulus. Behavioural responses and neural activity (EEG) were recorded from 18 participants aged 19–65. Participants performed above chance at 42 msec, with higher accuracy at 126 msec. At 14 msec, participants exhibited below-chance sensitivity and a highly conservative bias against reporting danger. Event-related potentials mirrored behavioural patterns, with neural differences between dangerous and non-dangerous images evident only in the 42-msec and 126-msec conditions. Prominent peak differences were observed across posterior, central, and frontal channels 220–300 msec after the stimulus onset. Frontal positivity was greater for dangerous than non-dangerous images, whereas central negativity and posterior positivity were greater for non-dangerous images. These results highlight the temporal dynamics of rapid danger detection via semantic gist perception, demonstrating that humans can discriminate dangerous from nondangerous scenes within 42 msec, thereby enabling efficient evaluation of dangerous environments.

Topic Area: PERCEPTION & ACTION: Vision