Applied Deep Learning for Safety in Construction Industry

Abstract

The construction industry in India is the second-largest employer and contributor, but it accounts for 24.2% of occupational deaths. On average, 38 people die in accidents every day. Making it a matter of concern, this paper explains Convolutional Neural Networks, a deep learning approach used for detecting workers on-site and recognizing the safety equipment worn by handling images and videos with a dataset size of approximately 2000 annotated images tailored into over approximately 9000 classified images. Which will be utilized to categorize and recognize workers in the construction industry wearing their relevant safety equipment. In addition to providing a comparison of the output, with the accuracy gained after applying a pre-trained model (i.e. VGG16) to perform classification and model prediction improvisation, along with the classification of the workers who are wearing their equipment this paper also supports future scalability to represent it in an analytical format. In the first approach, the model implements the CNN model and, in the second approach, implements the VGG16 model, eventually comparing the accuracy of the result for further using the best model to deploy on real-time videos/images. Therefore, the Safety Manager will find it more convenient to manage and instruct workers on the site.