J. Dang, Y. Wang, and Z. Zhu (PRC)
Traffic signs recognition, genetic algorithm(GA), radial basis function neural network(RBFNN), feature extraction, and adaptive principal components extraction neural network(APCENN)
The paper proposes to use radial basis function neural networks (RBFNN) to recognize traffic signs. Genetic algorithm (GA) adapting to traffic sign recognition is developed to train RBFNN to obtain appropriate structures and parameters according to given fitness functions. Hence RBFNN can escape from local minima to obtain global optimal parameters. Before sign images are fed into RBFNN for recognition, their features are extracted by adaptive principal components extraction neural network (APCENN) to decrease complexity of RBFNN and to speed the sign recognition. APCENN, based on principal component analysis (PCA), can auto-adaptively and rapidly extract principal component of images to improve the real-time performance by parallel calculation. In order to improve recognition speed and accuracy, traffic signs are classified into three categories by special color and shape information. Three APCENNs are designed for the three categories, and the outputs of APCENN are the inputs of the RBFNN. The training set imitating possible sign transformations in real environment is created to train and test the nets. In the experiment, the performance of the RBFNN trained by genetic algorithm is compared with that of the RBFNN trained by K-mean. The results show the superiority of the proposed method.
Important Links:
Go Back
