An integrated vision-based dynamic collision risk assessment framework of workers and mobile machinery on construction site

The concepts of environmental impact index and spatial conflict degree have gained prominence in enhancing the controllability of collision accidents and mitigating the likelihood of collisions during construction processes. Nevertheless, prior studies predominantly focused on exploring collisions in terms of proximity and congestion between pairs of entities, thereby overlooking a comprehensive consideration of workers, construction machinery, environmental factors, and the spatial interaction of specific activities.To this end, this study aims to propose an integrated vision-based dynamic collision risk assessment framework by setting workers and mobile machinery as targeted research objectives and embedding a comprehensive risk assessment model in the proposed framework, thereby comprehensively assessing four types of risk factors (i.e., proximity, congestion, environmental impact index, and spatial conflict), and visualize the hierarchy of risk warnings. Firstly, a comprehensive risk assessment model was developed by using the fuzzy comprehensive evaluation method. This is followed by developing a dynamic risk assessment framework to extract the spatial information of the monitored objects by using computer vision as the underlying data of the risk factors. Finally, the proposed integrated framework was validated by an experimental study. This experiment’s safety risk assessment results are consistent with expectations, which largely illustrates the effectiveness of the evaluation model constructed in this paper. For the vision module, the accuracy of classification and monitoring is more than 95%, and the speed of the object detection algorithm to process the video is about 10 frames per second, which shows the feasibility of this study.