An automated system for monitoring the use of personal protective equipment in the construction industry

We present a novel computer vision system which generates automated indicators of proper use of personal protective equipment (PPE) of great importance in the construction industry, specifically the use of safety helmet and high visibility vest. The system is built on a neural network architecture that works on digital images. First, the OpenPose network is used for the detection of anthropometric points of the visualized workers. These points are used next to automatically segment regions of interest (ROI) located about a worker's head and trunk. On these ROIs, a neuronal classifier estimates the presence or absence of each PPE of interest. Obtained results in moving videos from drones or smartphones show that our system is fully capable of carrying out a complete evaluation of usage indicators of these two PPEs without human intervention, with the main purpose of preventing potentially dangerous incidents in the workplace.