Realtime Safety Analysis System using Deep Learning for Fire Related Activities in Construction Sites
Uttam Kumar Dwivedi, Chayut Wiwatcharakoses, Yoshihide Sekimoto
Abstract
The era of digital transformation focuses on the integration of digital and AI based technology in construction industry for sustainable economic growth and high quality of life. This paper aims to provide a real-time detection and tracking of various construction activities and provide immediate practical safety guidelines and alert for probable accidental scenarios to ensure the safety of construction site and workers by using deep learning algorithms with vision-based edge devices and smartphone. Proposed paper develops a hybrid algorithm using scene classification first, and dependent object detection and tracking second to analyze vast category of fire related activities from video and images in real-time using computationally challenging devices. To cover the ever-changing construction location, easy to move smartphone-based applications were developed with AI as an API solution. The review of the results confirms superior real-time performance in successfully identifying and providing clear safety guidelines for indoor and outdoor fire related activities such as welding work and fire safety equipment and workers safety gear such as hardhat helmet. The study validated the practicality of IoT and deep learning-based solutions for construction jobsites with indoor and outdoor locations.