<![CDATA[Helmet use is a critical safety measure for motorcycle riders, yet non-compliance remains high in Indonesia. This study introduces a real-time helmet detection system using the YOLOv8 architecture, deployed on Android devices with the Kotlin programming language. A dataset of 1,197 digital images was collected and annotated using Roboflow Annotate, containing two classes: helmet users (True) and non-users (False). To improve model generalization, data augmentation techniques such as rotation and shear were applied. The model was trained using the pretrained yolov8n.pt weights and evaluated based on mAP and Intersection over Union (IoU). During training, the model achieved a mAP50 of 98% and a mAP50–95 of 59.6%. In testing, the mAP50 reached 98.3% and mAP50–95 reached 61%, with an average IoU of 0.73. The trained model was then converted into TensorFlow Lite format and integrated into an Android application. Real-time testing showed a detection accuracy of 93.3%. These results demonstrate that YOLOv8 is effective for mobile-based real-time helmet detection and has strong potential to support traffic law enforcement systems, especially in urban environments where manual monitoring is inefficient. The system contributes to enhancing public safety through smart technology integration.]]>
Copyrights © 2025
