Helmet Usage Detection on Motorcyclist Using Deep Residual Learning

In Indonesia and other developing countries, motorcycles are popular means of transportation. With the growing number of motorcyclists, it becomes harder for law enforcement to monitor motorcyclists who do not use a helmet. In this research, we propose a new system to detect motorcyclist that does not wear a helmet from a dashboard camera footage. We use a convolutional neural network to detect motorcyclists from the footage and two different residual network types to count the number of passengers and number of helmets, respectively. Our best motorcyclist detection model can achieve 100% accuracy, while our passenger and helmet detection models can achieve F1 Scores of 0.99 and 0.97, respectively. Our system can achieve 0,93 in video testing accuracy.