Comparative Evaluation of YOLOv8 and YOLOv12 for Motorcycle Helmet Detection and Automatic Number Plate Recognition Using Mobile Surveillance

This paper presented a rigorous comparative evaluation of YOLOv8m and YOLOv12m for motorcycle helmet compliance detection and Automatic Number Plate Recognition in Indian urban dashcam imagery. The study employed a reproducible 1,000-image dataset sourced from the publicly available Indian Helmet Detection Dataset on Kaggle [8], partitioned into 600 training, 100 validation, and 200 test images, annotated via Roboflow [11], and evaluated on Google Colaboratory T4 GPU infrastructure. YOLOv12m demonstrated statistically significant and operationally meaningful improvements across all evaluation metrics: mAP@50 of 91.20% vs. 58.70% for YOLOv8m (Δ = +32.5 pp), recall of 89.80% vs. 53.10% (Δ = +36.7 pp), and F1 of 91.08% vs. 65.50% (Δ = +25.6 pp). All three detection classes surpassed the 90% mAP@50 threshold for YOLOv12m. These results validate the transformative impact of the R-ELAN backbone and Area Attention Mechanism on domain-specific traffic enforcement detection tasks, and confirm that high-quality public Kaggle datasets provide a reproducible and effective foundation for domain-specific model development without requiring proprietary data collection infrastructure

[1] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, "You only look once: Unified, real-time object detection," in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Las Vegas, NV, USA, Jun. 2016, pp. 779–788. DOI: 10.1109/CVPR.2016.91 [2] G. Jocher, A. Chaurasia, and J. Qiu, "Ultralytics YOLOv8," GitHub Repository, Ultralytics Inc., 2023. [Online]. Available: https://github.com/ultralytics/ultralytics [3] A. Wang et al., "YOLOv12: Attention-centric real-time object detectors," arXiv preprint arXiv:2502.12524, Feb. 2025. [Online]. Available: https://arxiv.org/abs/2502.12524 [4] R. Saini, S. Kumar, V. Bansal, and P. Rawat, "Helmet detection on motorcyclists using image descriptors and classifiers," Transp. Res. C: Emerg. Technol., vol. 115, p. 102616, Jun. 2020. DOI: 10.1016/j.trc.2020.102616 [5] V. Mandal and Y. Adu-Gyamfi, "Object detection and tracking algorithms for vehicle counting: A comparative analysis," J. Big Data, vol. 9, no. 1, p. 31, Mar. 2022. DOI: 10.1186/s40537-022-00580-7 [6] A. Kumar, R. Sharma, and M. Patel, "Domain-specific fine-tuning of YOLO models for traffic violation detection," in Proc. IEEE Int. Conf. Intell. Transp. Syst. (ITSC), Bilbao, Spain, Sep. 2023. [7] C. Henry, S. Y. Ahn, and S. Lee, "Road marking and license plate detection using convolutional neural networks," Sensors, vol. 21, no. 2, p. 364, Jan. 2021. DOI: 10.3390/s21020364 [8] A. Vaid, "Indian Helmet Detection Dataset," Kaggle, 2023. [Online]. Available: https://www.kaggle.com/datasets/aryanvaid13/indian-helmet-detection-dataset [9] M. R. Arun, P. Krishnamoorthy, and R. Sundaram, "Deep learning based object detection in surveillance video," in Proc. 4th Int. Conf. Smart Electron. Commun. (ICOSEC), Trichy, India, 2023, pp. 1794–1801. DOI: 10.1109/ICOSEC58147.2023.10276262 [10] R. Laroca et al., "A robust real-time automatic license plate recognition based on the YOLO detector," in Proc. Int. Joint Conf. Neural Netw. (IJCNN), Rio de Janeiro, Brazil, Jul. 2018, pp. 1–9.