Development of an AI Dataset for Object Detection at Construction Sites

This paper introduces a dataset for object detection training, which includes 12 types of construction machinery commonly operated at civil engineering sites. To collect this dataset, a housing development construction site in South Korea was selected, and a video collection system was operated over a six-month period. Frames were extracted from the collected video footage on a daily basis, resulting in a total of 87,766 images in the full training dataset. Using the COCO Annotator tool, labels and bounding box annotations were processed, generating a total of 856,485 objects.