Comprehensive CNN-Based Approach for Helmet Use Detection of Tracked Motor Cycles
B Ravikrishna, K. Sai Priya, J. Harika, M. Krishna Pranathi, N L Apoorva · 2021
Precautionary measures for safety on road and traffic have become key roles played by the new Traffic Regulations and Management, especially for two-wheeler riders. It has been made mandatory for the two-wheeler vehicle riders to wear helmets, which is, unfortunately not being followed by a few riders. In such a scenario, several aspects have been taken into account, resulting in the current project, that consists of the technique where, any two-wheeler rider without wearing helmet, can be detected in places it must be worn compulsorily. In this project, helmet detection and recognition have been deployed mainly using algorithms such as YOLOv2 (You Only Look Once version 2) and LeNet. Any kind of helmet can be detected and recognised under several conditions and cases. Whenever a rider is detected without a helmet from the source file in the format of either image or video, the instance of such malpractice is considered and the output is built by analysing the data.. The trained model gives an accurate result at the rate of95% under multiple cases and marks to 1 FPS using CPU.