Enhanced Helmet Detection with YOLOv8: A Deep Learning Model for Boosting Road Safety

Abstract

Road traffic accidents involving motorcycles or scooters are usually very fatal and non-wearing of helmets. Although laws exist, a lot of riders put themselves at great risk of grave injury or death as they do not comply with the basic rule of wearing helmets. In this study, the authors seek to improve an existing system for helmet use detection by using the latest object detection model, YOLOv8, for real time helmet detection on motorcyclists. The system is developed from a single dataset containing images of riders both wearing and not wearing helmets and utilizing sophisticated data augmentation and image preprocessing methods. The primary reason why YOLOv8 was chosen was because of accuracy, speed, obstructions, and variable lighting conditions. The results of the model were outstanding, achieving high accuracy and mAP, and therefore enabling automatic helmet law enforcement. This approach provides support for traffic enforcement officials, thereby lessening the need for active supervision and aiding in greater overall safety and compliance through smart monitoring systems.