Helmet Violation Processing Using Deep Learning

Abstract

Today, road accidents are one of the major causes of human deaths. Among the different type of road accidents, motorcycle accidents are common in Asia and cause severe injuries. The helmet is the main protection device for motorcyclists. Most countries have rules for use of helmets but many people fail to follow these rules for various reasons. One solution is a robust system that can find the motorcyclists who are violating helmet rules and record the evidence necessary for legal action. This thesis paper presents the use of deep convolutional neural networks (DCNNs) for finding motorcyclists who are violating helmet rules. I describe incremental development of a DCNN and an evaluation in terms of accuracy and speed for finding motorcyclists who are violating the helmet use rules. The license plate character DCNN provides highly accurate and fast character recognition of Thailand license plates.