Guardrail detection on construction sites using drones and computer vision

Abstract

Using drones and computer vision enhances the safety inspection efficiency on construction sites. However, there is a gap in the integrated application of these technologies for inspecting guardrail systems.This paper aims to develop a guardrail detection system for construction sites based on drones and computer vision. The research involves defining and validating a drone image collection protocol, image collection at two projects, and training and testing four object detection models in a computer vision-based system using original and artificial images (with zoom, rotation, and shear). The automated guardrail detection showed high performance with a precision of 96.83% and 95.31% in the test stage for models trained with images with shear and rotation, respectively. This research contributed to advancing the integration of drones and computer vision for automated guardrail inspections, enabling greater efficiency and agility in collecting, processing, and analyzing results.