Research on Traffic Sign Image Recognition Based on Deep Learning

In this paper, a Convolutional Neural Network (CNN) model incorporating image brightness histogram equilibrium is proposed for traffic sign image recognition. With the increase of traffic flow in cities, the traffic sign recognition system is of great significance in intelligent transportation, but the problems of complex backgrounds, illumination variations, and small target recognition pose challenges to recognition. Traditional methods rely on handcrafted feature extraction with limited effect, while the development of deep learning, especially CNN, provides new ways to solve these problems. In this paper, an optimized traffic sign recognition system is constructed by image preprocessing module, feature extraction module, and classification and identification module, and the German Traffic Sign Recognition Benchmark Dataset (GTSRB) is used for experimental validation. The results show that the model performs well in accuracy, F1-SCORE and average precision mean, with accuracy as high as 97.26%, and accuracy is further improved to 98.91% using image enhancement. The experimental results show that the CNN model combined with image preprocessing has significant advantages and applied value in the traffic sign recognition task.