Helmet Detection Using Faster Region-Based Convolutional Neural Networks and Single-Shot MultiBox Detector

In a country like India, with excessive population density in all big cities, motorcycles have become dominant modes of transport. It is observed that most motorcyclists avoid wearing helmets despite it being an indispensable safety equipment, whose use can significantly reduce the risk of severe head and brain injuries during accidents. Due to violations of most of the traffic and safety rules, motorcycle accidents have been skyrocketing in the recent years. Hence, it’s the need of the hour to build an effective and scalable system capable of automatic helmet detection by analyzing the surveillance camera’s traffic videos. Although several theoretical deep learning-based models have been proposed to detect helmets for the traffic surveillance aspect, an optimal solution for the industry application is less discussed. This paper demonstrates a novel implementation of the Faster R-CNN and SSD framework for accurate helmet detection in real-time low-quality surveillance videos. The experimental results claim that there is a trade-off between accuracy and execution speed. We also present a comprehensive comparative analysis of the two algorithms and determine the best real-time use case scenarios for each of them.