Automatic Helmet Detection using EfficientDet
Muhammad Talha Ubaid, Amara Kiran, Muhammad Tayyab Raja, Umme Aliza Asim, Abdou Karim Darboe, Muhammad Asad Arshed
Abstract
Casualty rates are increasing rapidly in Asian countries due to carelessness of the motorcyclists. Helmet helps to ingest the collision, and hence it prevents head injuries which lead to death. But ensuring that riders wear helmets and monitors to detect those who’re not is an exhausting task and more prone to error due to physical limitations of human beings. An automatic system is proposed for which a custom dataset in Lahore, Pakistan is generated to detect helmets. The proposed method uses EfficientDet approach to recognize helmets. EfficientDet provides better computational accuracy and proficiency than previously proposed models. The proposed system was later compared with other models as well and achieved 95.23% accuracy for the helmet and non-helmet rider detection.