Helmet detection in riders for law enforcement

Abstract

In these days motorbikes are the most popular and least expensive modes of transport as compared to other vehicles. However due to less protection there is risk involved of accidents and injuries. As a result, to this, it is highly recommended for bike-riders to use helmet while driving. Considering the value of helmet, Government of Pakistan has made it punishable offense to ride a motor-bike without helmet and have endorsed manual strategies to catch the violators who does not wear helmet while riding. The system is trained by using surveillance traffic signals videos. System can detects the bike rider who is not wearing the helmet. System have two models object detection trained by using Faster-RCNN and object classification trained by deep learning technique CNN. system detects violator using convolution neural network. It detects bike rider from video frame by subtraction of background and object segmentation, then it detennines bike rider is wearing helmet or not using classifier. If violator is detected, system will generate alert that helps the officer to identify violator and report to traffic of-ficer. System crop the violator image and save image to identify offender accuratly. In order to evaluate the system performance, extracting random 100 frames that are detected by system and counted bike manually and compared it with system detected bike and calculated the accuracy of system in which were 76.79% of bike detection, 85% of bike with helmet and 87% of violators who were driving without helmet