An Improved ANN Algorithm for Traffic Sign Recognition

Abstract

An automatic traffic sign recognition system can help drivers operate the vehicle properly. Most existing systems include a detection phase and a classification phase. In this paper, a new classification method is presented based on an improved artificial neural network (ANN) algorithm for recognizing traffic signs. When the ANN algorithm is chosen to for traffic sign recognition, the key factor is to get the right weights in neural networks. Traditionally, the weights were solved by training the neural networks with a give samples set. But in most of the cases the convergence for the training is very slow, even it becomes divergence. In this paper, an improved BP neural networks algorithm was proposed. Compared with the old algorithm, a dynamic learning rate was used to get an optimization learning rate instead of a fixed learning rate. Combined the moment features, GSC features, experiments show that the iterative times for ANN training is reduced evidently.