Automatic helmet detection in real-time and surveillance video

Abstract

In current world, number of vehicles growing day by day results in higher accidents. So, “helmet” is the one key element for ensuring safety of bike riders. Knowing the fact, people tend to avoid wearing helmet. Government had imposed various rules making it compulsory to wear helmet and fine being levied on offenders. But it is not possible to track each rider in current manual tracking and video surveillance system. A model for detection and classification of bike riders who are wearing the helmet as well as those who are not wearing it is proposed in this paper. The proposed model trained on COCO dataset uses only one neural network per image and that is quicker than R-CNN and Fast R-CNN as they use multiple neural networks.