Real-Time Proximity Monitoring between Heavy Vehicles and Workers for Struck-By Accident Prevention on Construction Site

Accidents at construction sites are a major concern in society, often causing injuries, disability, and even death. Struck-by accidents have caused injuries in various industries, 75% of which are caused by heavy vehicles. Hence, several studies on preventing accidents caused by construction vehicles have been conducted. However, they proposed prevention methods involving installation of sensors on construction objects. To address this limitation, we proposed a real-time proximity-monitoring system that utilized a pre-installed camera. The system consisted of three steps: (i) construction-object detection with YOLOv7, (ii) depth estimation by comparing the heights of heavy equipment and workers in two-dimensional (2D) images, and (iii) proximity monitoring using Euclidean distance and depth estimation. First, we compared YOLOv7 and YOLOv7-X and concluded that YOLOv7 was the most appropriate model for object detection in continuously changing construction fields. Then, we compared the heights of the bounding boxes of the equipment and worker safety vests, and calculated the dangerous area range around the construction equipment to be in the range 2809≤R≤14782.5. Proximity monitoring was conducted with an accuracy of 80.85% and the recall value was 0.97. This system is expected to be utilized at construction sites because it does not require additional equipment or an image transformation procedure to reduce the distortion of 2D images.