Traffic Sign Recognition System Based on Neural Network Algorithm

This paper presents the implementation of a system for automatic recognition of traffic signs using convolutional neural networks (CNN). The solution is based on the ResNet architecture with elements of recurrent layers, which increases the accuracy and versatility of image processing. Traffic sign recognition system provides user functionality to upload images of signs to the telegram bot, receive their classification, as well as useful tips on traffic rules. Development was performed using the PyTorch library, RTD dataset, and DeepPavlov language model. Test results showed high model accuracy (97.7% accuracy) and user-friendliness of the interface. The developed solution can be used both in educational institutions for driver training and as an auxiliary tool for analyzing the traffic situation.