Classifying helmeted and non-helmeted motorcyclists

Abstract

Riding a motorcycle without a helmet can cause serious injury. Although a crackdown on traffic violations is a good way to stop this unsafe practice, it is not realistic to manually find and arrest riders who do not wear helmets where there are numerous motorcycle riders, as in Vietnam. In consideration of this situation, we developed an automatic detection system for riders who are not wearing a helmet using deep learning. The proposed method’s accuracy, precision, recall, and F-measure in classifying motorcyclists into helmeted and non-helmeted are 0.966, 0.957, 0.936, and 0.946, respectively. The quality of the classification was higher than in previous work which did not use deep learning. As with other image-processing systems using deep learning, our system achieved state-of-the-art performance. This system will reduce not only the number of motorcycle riders not wearing a helmet, but also the manual work of arresting illegal riders.