Fast Traffic Sign Recognition Algorithm Based on Multi-scale Convolutional Neural Network

In order to solve the problems of low accuracy and slow recognition speed of traffic signs, a multi-layer and multi-scale convolutional neural network recognition algorithm is proposed in this paper. Under the premise that the algorithm has high recognition accuracy, the traffic sign recognition model can be quickly established. This paper firstly improves the feature extraction method in the single-scale convolutional neural network to extract the global features and local features of the traffic sign image. Meanwhile, the features generated by different levels are fused into multi-scale features and transmitted to the full- connection layer classifier to improve the accuracy of traffic sign recognition. Then, the output data of the convolutional layer is processed by batch normalization (Batch Normalization) method, by normalizing the mean value and variance of each hidden layer, the occurrence of gradient explosion or disappearance is reduced, the training convergence speed is increased, also the training time is reduced. Finally, the validity of the proposed algorithm is tested in German traffic signs benchmark database (GTSRB). Experimental results show that the proposed algorithm can not only maintains high accuracy, but also performs well in training time and recognition time. Also, the fast recognition algorithm of multi-layer and multi-scale convolutional neural network has good generalization ability and real-time performance, providing important technical support for the reliability and safety of intelligent driving.