A Study Of Visual Cognitive Strategies During Construction Hazard Recognition

Hazard recognition is the primary prerequisite for preventing construction safety accidents.Traditional manual safety inspection is time-consuming and inefficient.Therefore,researchers attempted to develop hazard recognition techniques such as computer vision to assist and replace manual recognition.However,it is difficult for existing techniques to accurately recognize hazards in complex scenarios such as construction sites,essentially because current research rarely reveals the visual cognitive mechanism and strategy of human perceiving hazards,which cannot provide theoretical support for the advancement of hazard recognition techniques.Although some studies have attempted to explore the visual behavior of hazard recognition by cross-sectional data or eye-movement-related statistical indicators,which lack consideration of temporality or appropriate methods for analyzing visual sequences,yielding limited information to interpret visual patterns and cognitive processes.This research aims to investigate the visual cognitive processes and strategies of construction workers when recognizing hazards.First,this study designed and conducted an eye-tracking experiment with 85 construction workers as subjects and 120 images of the construction site as materials.The subjects were required to sequentially recognize hazard,while their visual behavioral data were recorded.Second,after the experiment,the visual data were programmed for cleaning and processing,and the visual behaviors were coded according to the visual sequence generation rules.Thirdly,the visual path sequences during hazard recognition were analyzed by the temporal qualitative comparative analysis(TQCA)method to summarize the visual strategies.Finally,the cognitive process of hazard recognition was synthesized and deciphered by cross-referencing and triangulated the subjects’ self-reported recognition logic in the post-experimental interviews.The result show that:(1)there are two different visual strategies presented in subjects’ hazard recognition process,i.e.,experience-driven "scenario-association" and knowledge-oriented "norm-interpretation",corresponding to bottom-up and top-down cognitive models,respectively.(2)The knowledge oriented strategy may not conform to the experience-driven cognitive habit,so the knowledge strategy should refer to the experience strategy appropriately in practice,such as improving the safety training design.(3)By referring to human experience strategies,hazard recognition techniques such as computer vision are expected to achieve progress from scene recognition to cognitive recognition,such as improving the ability to focus on and recognize the interactions between scene components.Overall,this study contributes to the understanding of visual cognitive processes and strategies in construction hazard recognition,complements TQCA as a methodological tool for visual behavior analysis,extends the application of visual cognitive theories such as the gestalt model in construction hazard recognition,and feasible practical guide for safety trainings and theoretical foundations of computer vision techniques for construction hazard recognition.