A driver safety support system which recognize traffic signs

Designing vehicles with improved driving safety that aim to minimize the number of traffic accidents caused by driver errors on highways is one of the most important issues of today's automotive technology. For this purpose, the number of active and passive safety systems in vehicles is increasing day by day. Traffic sign recognition systems, which are one of the active safety systems, are the systems that recognize the traffic signs through the front looking cameras and inform the drivers and they have started to take their place in the new generation vehicles. These systems, which operate in real time, have not yet achieved the desired performance, especially in complex road conditions, and remain an important research topic. In this study, a driver safety support system (DSSS) based on convolutional neural networks (CNN), which aims to recognize traffic signs, was proposed. The German Traffic Sign Recognition Benchmarks (GTSRB) dataset was used as the traffic sign data set. Due to the numerical disproportion of the images in the training set, the number of images in the training set was increased by using data augmentation methods. The layer structure and the most appropriate training parameters of the CNN models using LeNet-5, AlexNet, GoogleNet and ResNet CNN architectures were determined experimentally. As a result of the experiments, it was shown that the proposed DSSS using ResNet architecture has 98.10% classification accuracy. Also, DSSS performance results were presented on a group of traffic signs which are taken in Turkey highways.