Answers are grounded in this paper and cite their sources.
Demonstrates enhanced safety sign effectiveness in construction, suggesting innovative communication methods.
Although construction safety signs have long been used on the jobsite to prevent accidents, their limitations (e.g., habituation) have been continuously reported. Prior studies have suggested the potential of leveraging the optical illusion to enhance safety communication between safety measures and humans due to its perceptual advantages. However, the current practice of developing optical illusion designs has heavily relied on professionals’ expertise and manual work. A promising alternative is to adopt a text-to-image approach, which allows for generating diverse and quality imagery outputs based on input prompts. Unfortunately, however, minimal research efforts have been made on proposing a framework that can assist in automatically developing optical illusion-based construction safety signs, using text-to-image. In addition, how the graphical components (e.g., shape and pictogram) of signs can be integrated with the prompt engineering and demonstrating the effectiveness of the optical illusion safety signs versus conventional ones, shape critical research gaps. To this end, this study aims to introduce a text-to-image-assisted framework for developing optical illusion construction safety signs and prove its effectiveness. The results demonstrate that the optical illusion safety signs associated with three critical hazard types (fall, struck-by, and caught-in/-between), created by component-level prompting strategies, keyword analysis, and users’ evaluation, outperformed the traditional signs in five aspects (e.g., readability, comprehensibility, and attention). The research outcome contributes to the generative artificial intelligence domain by presenting a feasible technical roadmap for the applications of text-to-image, as well as the construction safety field, where effective recognition of safety information is critical.
No takes yet. Share an insight, caveat, or question.
Kang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: