Automated helmet and plates number detection using deep learning / Muhammad Zakwan Arif Muhammad Tajuddin, Noraini Hassan and Raihah Aminuddin

The increasing use of motorcycles and the corresponding rise in related accidents, particularly among riders not wearing helmets, necessitates an efficient and cost-effective solution for helmet detection. Despite the existence of laws mandating helmet use, enforcement remains a challenge due to the need for manual human assistance in monitoring compliance. This study aims to address this issue by 1) designing a system capable of detecting motorcyclists not wearing helmets and identifying their license plate numbers, 2) developing an automatic detection system using ResNet50 and EasyOCR, and 3) testing the functionality and accuracy of the developed system. The system was trained using a dataset of 500 images sourced from Kaggle, featuring riders both with and without helmets. The application of ResNet50 and EasyOCR demonstrated significant performance in recognizing helmets and license plates across various scenarios. The results indicate that the helmet detection model using ResNet50 has achieved a significant performance with a 90% accuracy rate in recognizing helmets and license plates. Despite certain limitations, this project opens avenues for future research to refine further and advance detection systems, ultimately benefiting users in monitoring and enhancing their safety on the road.