Abstract: In order to improve the accuracy of traffic sign recognition under complex conditions (such as occlusion, perspective distortion, etc.), the paper presents an effective traffic sign recognition method based on convolutional neural network and ensemble learning. The proposed method firstly splits out the traffic signs by incorporating color segmentation, morphology processing and shape detection, and then identifies them using SVM and Softmax classifier based on the features extracted by CNN, respectively. Finally, the two kinds of classification results are integrated under the ensemble learning framework. Experimental results show that the proposed method can effectively improve the accuracy of traffic sign recognition under complex conditions, and has high overall performance.

Key words: traffic sign recognition, ensemble learning, support vector machine, convolutional neural network, principal component analysis