Classifying Motorcycle Rider Helmet on a Low Light Video Scene using Deep Learning

For safety in transportation, it is important to always monitor the use of proper motorcycle helmet, especially at night. One way to enforce transportation rules and regulations in wearing proper motorcycle helmet is to use computer vision technology. This study focusses on classifying motorcycle rider helmet at low light video conditions, like at dusk and at night, using YOLOv5 and YOLOv7 with Deep SORT. In these deep learning methods, the study tunes and optimizes hyperparameters to attain high accuracy in classifying motorcycle rider helmet at this challenging environment. To accomplish this objective, a vast and diverse dataset was employed, containing classes such as riders, different types of helmets (valid and invalid), and instances of riders not wearing helmets at all in Metro Manila, Philippines. The results show that Hyperparameter 3 consistently outperformed other settings in terms of precision (95.6%), recall (91.2%), and mean average precision (mAP) scores across multiple scales and time frames with 95.1% on mAP@0.5 and 76.3% on mAP@0.95, owing to greater epochs, quicker learning rates, and lower batch sizes.

@article{Tomas2024,
  title     = {Classifying Motorcycle Rider Helmet on a Low Light Video Scene using Deep Learning},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {15},
  number    = {2},
  year      = {2024},
  publisher = {The Science and Information Organization},
  author    = {John Paul Q. Tomas and Bonifacio T. Doma},
  doi       = {10.14569/IJACSA.2024.0150205},
  url       = {https://doi.org/10.14569/IJACSA.2024.0150205}
}