Traffic Violation Detection System on Two-Wheel Vehicles Using Convolutional Neural Network Method

Abstract

The number of vehicles is increasing every year, and along with it, the number of traffic violations is also rising.. Traffic violations are one of the causes of traffic accidents. Currently, traffic violation detection still uses conventional methods, involving the police to take action. Preliminary research on traffic violation detection by several researchers mostly uses the Yolo Method. The study aims to design a traffic violation detection system for two-wheeled vehicles using the Convolutional Neural Network (CNN). In this research, the CNN method was used with the Faster RCNN architecture. Faster R-CNN is composed of convolution layers, Relu, and pooling layers which are used to extract features from images. An image in the size of 3264 x 1836 pixels, with the type of marking violation and helmet use was used as a sample. The number of images used was 660 images with 600 images for training and 60 images for testing. The system will detect traffic violations on two-wheeled vehicles, namely helmet use violations and road marking violations. This traffic violation detection system for two-wheeled vehicles produces the highest accuracy, namely 85% with a maxpooling kernel size value of 1x1, stride 1 and a learning rate of 0.003. This research has the potential to be applied to areas that are less accessible to the police, because the system will record and analyze violations.