Traffic sign recognition based on improved convolutional networks

Real-time and accurate traffic sign detection and identification is a huge challenge under real vehicle driving conditions due to background diversity, illumination intensity, shooting position, lens pixel value and other factors. In this paper, an improved convolutional network based on LeNet-5 is proposed for traffic sign recognition. The inception module is introduced to enhance the performance of feature extraction. The size of convolution kernels is changed to 3×3 and 1×1. In addition, a method of image standardised pre-processing is introduced for batch processing of samples in order to improve the generalisation performance of recognition. Furthermore, the dropout layer is utilised to prevent overfitting. The experimental results show that the improved neural network has good robustness and the network recognition accuracy reaches more than 99%. Compared with the traditional Lenet-5 model, the method has more outstanding performance in the identification of multiple classification problems and has certain advancement for traffic sign recognition. Copyright © 2021 Inderscience Enterprises Ltd.