Automatic Detector for Bikers with no Helmet using Deep Learning
Narong Boonsirisumpun, Wichai Puarungroj, Phonratichi Wairotchanaphuttha · 2018
The success of digital image pattern recognition and feature extraction using a Convolutional Neural Network (CNN) or Deep Learning was recently acknowledged over the years. Researchers have applied these techniques to many problems including traffic offense detection in video surveillance, especially for the motorcycle riders who are not wearing a helmet. Several models of CNN were used to solve these kinds of problem but mostly required the image pre-processing step for extracting the Region of Interest (ROI) area in the image before applying CNN to classify helmet. In this paper, we proposed to apply another interesting method of deep learning called Single Shot MultiBox Detector (SSD) into helmet detection problem. This method is the state-of-the-art that is able to use only one single CNN network to detect the bounding box area of motorcycle and rider and then classify that biker is wearing or not wearing a helmet at the same time. The results of the experiment were surprisingly good. The classification accuracy of bikers not wearing a helmet was extremely high and the detection of the ROI of biker and motorcycle in the image can be done at the same time as the classification process.