Unmanned Aerial Systems and Deep Learning for Safety and Health Activity Monitoring on Construction Sites

• Published in: Sensors (Basel, Switzerland) Vol. 23; no. 15; p. 6690
• Main Authors: Akinsemoyin, Aliu, Awolusi, Ibukun, Chakraborty, Debaditya, Al-Bayati, Ahmed Jalil, Akanmu, Abiola
• Format: Journal Article
• Language: English
• Published: Switzerland MDPI AG 26-07-2023; MDPI
• Subjects: Accident prevention; activity monitoring; Algorithms; Analysis; Artificial intelligence; Automation; Behavior; Cameras; Case studies; Computer vision; Construction accidents & safety; Construction industry; construction site; Construction workers; Data analysis; Data collection; Deep learning; Fatalities; Health aspects; Health hazards; Illnesses; Internet of Things; Machine vision; safety and health; Sensors; Strategic planning (Business); UAS; Work environment; Workers; activity monitoring; deep learning; construction site; safety and health; UAS
• ISSN: 1424-8220; 1424-8220
• DOI: 10.3390/s23156690

Construction is a highly hazardous industry typified by several complex features in dynamic work environments that have the possibility of causing harm or ill health to construction workers. The constant monitoring of workers' unsafe behaviors and work conditions is considered not only a proactive but also an active method of removing safety and health hazards and preventing potential accidents on construction sites. The integration of sensor technologies and artificial intelligence for computer vision can be used to create a robust management strategy and enhance the analysis of safety and health data needed to generate insights and take action to protect workers on construction sites. This study presents the development and validation of a framework that implements the use of unmanned aerial systems (UASs) and deep learning (DL) for the collection and analysis of safety activity metrics for improving construction safety performance. The developed framework was validated using a pilot case study. Digital images of construction safety activities were collected on active construction sites using a UAS, and the performance of two different object detection deep-learning algorithms/models (Faster R-CNN and YOLOv3) for safety hardhat detection were compared. The dataset included 7041 preprocessed and augmented images with a 75/25 training and testing split. From the case study results, Faster R-CNN showed a higher precision of 93.1% than YOLOv3 (89.8%). The findings of this study show the impact and potential benefits of using UASs and DL in computer vision applications for managing safety and health on construction sites.