A motorcyclist helmet detection system through a two-stage CNN approach
Ricardo Alfonso, Christian Daher, Mario Arzamendia, Kevin Cikel, Derlis O. Gregor, Daniel Gutierrez, Sergio Luis Toral, Marcos Villagra · 2021 IEEE CHILEAN Conference on Electrical, Electronics Engineering, Information and Communication Technologies (CHILECON) · 2021
Motorcyclists are the road users more prone to suffer from traffic accidents and serious injuries. The effects of accidents can be alleviated by a suitable use of helmets that helps to protect a motorcyclist’s head. This paper presents a tool for facilitating the recognition of motorcyclists who circulate in public roads without wearing helmets. This tool can help local authorities to quantify the compliance levels of motorcyclists and prevent irreversible damage to them. The proposed system of this work was developed using image processing and convolutional neural networks (CNN), which combined a pre-trained model of SSD Mobilenet V2 for the detection of motorcycles followed by a custom made model for helmet detection. The automated tool was implemented using an NVIDIA Jetson TX2 board. Results show a precision of 95% for the detection of motorcyclists and 93% for the classification of motorcyclists with and without a protective helmet.