Utilizing computer vision and fuzzy inference to evaluate level of collision safety for workers and equipment in a dynamic environment

The construction industry is facing unique problems in accident prevention. The existing management method for detecting workers' unsafe behaviors and unsafe states of objects relies primarily on manual monitoring, which does not only consume large amounts of time and money but also cannot cover all workers in the entire construction site. Meanwhile, the workers' perception of being at risk of injury decreases when they are concentrated in a crowded and noisy environment. In this case, it is difficult for them to take essential measures to protect themselves in the face of danger. In view of the aforementioned issues, this study proposes a method of evaluating the collision safety level of construction workers based on computer vision and fuzzy inference. Specifically, the proposed model works via two modules: vision extraction and safety assessment. The vision extraction module identifies construction workers and equipment through computer vision; centroid pixel coordinates and crowdedness are then extracted from a detection box. Afterward, the spatial relationship between moving devices and workers is calculated by a pixel calibration process. In the safety assessment module, the collected status information is analyzed by evaluating the safety level of each worker and conducting accident prevention through a fuzzy inference system. The safety level, which indicates the comprehensive risk of collision between workers and equipment in a particular dynamic environment, will be displayed numerically, breaking through the limitations of conventional qualitative evaluation. Field experiments validate the feasibility of the proposed method of informing workers about potential danger situations in an objective way. Moreover, by setting a safety-level threshold, the onsite safety management personnel can take corresponding measures to avoid collision accidents when the worker's safety level is lower than the threshold.