Authentication control system for the efficient detection of hard-hats using deep learning algorithms

Abstract

Hard-hats are the most essential safety gear on construction sites. Accidents happen due to the ignorance to wear a hard hat properly. To prevent construction accidents due to the non-usage of hard hats, automatic non-hardhat usage detection techniques have been observed to be more efficient. This paper analyses the usage of deep learning algorithms in detecting construction workers who fail to wear their hard hats properly. The accuracy of the 3 deep learning algorithms, i.e. YOLO v4, v5 and YOLACT++ have been tested using a set of data of Hard-hats collected from the crowdsourcing and web mining. These vision-based deep learning methods and algorithms have been applied and compared in terms of accuracy, scalability, and robustness. The research evaluated the different frames of construction site workers and classified image frames into different categories. The frames were input into the deep learning algorithm as per their visual categories. The objective of the paper is to analyze the present deep learning approaches for the real-time detection of the hardhat on the construction site in a way that the hardhat could be recognized efficiently without any failure and would be able to distinguish the workers who are not wearing their Hard-hats suitably and introduce a new access control system using deep learning methods to solve the problem of efficiently identifying Hard-hats for site workers. The experimental results demonstrate that the high score of precision, recall, and fast speed of the techniques, which would effectively detect non-usage of Hard-hats by site workers at different construction site conditions. This will help improve safety and supervision on construction sites.