Research Article
Helmet Detection Using YOLOv8 and Detection Transformer
@INPROCEEDINGS{10.4108/eai.28-4-2025.2357978, author={E. Bhanu and Kamaluru Mahammad and P. Jaagruthi and Sugali Harijawahar Naik and Yellanur Amruth and Gummadi Gnanesh}, title={Helmet Detection Using YOLOv8 and Detection Transformer}, proceedings={Proceedings of the 4th International Conference on Information Technology, Civil Innovation, Science, and Management, ICITSM 2025, 28-29 April 2025, Tiruchengode, Tamil Nadu, India, Part II}, publisher={EAI}, proceedings_a={ICITSM PART II}, year={2025}, month={10}, keywords={computer vision yolov8 detr deep learning object detection helmet detection road safety traffic surveillance}, doi={10.4108/eai.28-4-2025.2357978} }- E. Bhanu
Kamaluru Mahammad
P. Jaagruthi
Sugali Harijawahar Naik
Yellanur Amruth
Gummadi Gnanesh
Year: 2025
Helmet Detection Using YOLOv8 and Detection Transformer
ICITSM PART II
EAI
DOI: 10.4108/eai.28-4-2025.2357978
Abstract
Today's traffic systems including motorcycle safety is taken as the most important issue in reducing the number of deaths and injuries by wearing a helmet while in an accident. In this paper, a deep learning based smart helmet detection system is proposed with two state of the art object detection networks YOLOv8 and the Detection Transformer (DETR). The system is real-time, and it serves for law enforcement and traffic control departments. We train the models based on a large dataset of traffic surveillance photos in diverse environmental and situational conditions, such as different light, rider orientation, upper body occlusions. YOLOv8 is used due to its high quality and fast detection times, making it suitable for real-time operation. On the other hand we choose DETR to use instead for its strong scene context model which delivers superior detection accuracy in hard contexts. The evaluation results illustrate the trade-off between the models: DETR offers more accurate detection in challenging visual conditions, while YOLOv8 provides more efficient processing, making it suitable for dynamic surveillance. The results showed that both models could be embedded in intelligent transportation systems to enhance helmet compliance and safe driving behavior.
