Comparison of Precision and Accuracy of Image Processing Algorithm to monitor Safety Violations at a construction site

Abstract

Ensuring the reduction of fatalities and injuries is paramount in construction site management. Recent advancements in computer vision have proven instrumental in monitoring complex site conditions and progress through image processing surveillance. Although multiple imaging algorithms are currently employed for visualizing site progress and identifying safety violations, there exists a critical need to ascertain the most robust algorithm for this purpose.This study focuses on the efficacy of different computer vision models, namely YOLO Series, Detectron 2 and GroundNino in the detection of safety violations and the improvement of safety protocols within the construction sector. The results emphasize YOLOv8 as the preferred option as compared to its predecessors and detectron 2 owing to its remarkable efficiency, precision, adaptability, and developer-centric attributes. The real-time processing capabilities of YOLOv8, in conjunction with its high precision, render it a well-suited solution for the prompt monitoring of safety in the ever-changing settings of construction sites. This study highlights the potential of computer vision technology in enhancing safety protocols within the construction industry, hence facilitating more effective safety monitoring and accident avoidance.