Detecting helmets on bike riders using deep learning techniques

Abstract

The study introduces a technique for instantaneously and automatically detecting the helmets worn by bike riders. Bikers are a prevalent means of transportation for many folks in my country. Bike have become more popular than vehicles due to their reduced maintenance expenses, fewer space requirements for parking, and enhanced maneuverability and flexibility in urban environments. Although biking may be exhilarating and stimulating, it is not without of hazards. The proposed strategy seeks to provide the highest level of safety for bikers. Despite the legal requirement, a significant number of drivers continue to opt out of wearing helmets. In recent years, there has been a steady increase in the number of deaths, especially in developing nations. Installing a helmet detection system is crucial for ensuring public safety by accurately identifying drivers who are not wearing protective headgear. I use a dataset consisting of around 3202 data points in real-time for my method. In this study, I use several algorithms including Resnet50, Inception V3, EfficientNet, DenseNet201 and. These algorithms are applied to a dataset consisting of 1911 instances with helmet usage and 1291 instances without helmet usage. I achieved a remarkable 98% accuracy rate by using the EfficientNet model. The article's implementation section provides a comprehensive explanation of all the strategies used in the comparison statements. To create the most efficient model for the given circumstances, this investigation also utilizes model validation techniques.