Abstract
This study develops a helmet detection system using the YOLOv8 algorithm to support automatic work safety monitoring. A total of 4,924 training images were used, yielding 1,407 valid images after filtering. The system detects helmet objects with confidence levels ranging from 0.73 to 0.85 and processes images within an average of 0.65 seconds. The model achieved an F1-score of 91.1% and a precision of 0.95. This real-time detection system is expected to enhance safety compliance in workplaces and traffic environments.