基于卷积神经网络的交通标志识别方法研究

Research on Traffic Sign Recognition Method Based on CNN

现如今智能驾驶成为发展潮流,而交通标志识别作为智能驾驶中不可或缺的一部分有着重要的研究意义。为了提高交通标志识别的分类准确度,论文提出了RI-Model模型,该模型利用经典神经网络ResNet-50与Inception-V3进行特征提取,结合特征融合的思想来提高交通标志的识别率。采用比利时交通标志数据集(BelgiumTS),在对数据集进行预处理后利用RI-Model测试其识别正确率。结果表明:该方法能够在较短的训练时间内达到更好的收敛性能且具有很好的鲁棒性,RI-Model在该数据集上的识别准确率达到98.86%,相较于直接使用ResNet-50与Inception-V3算法提升了1.5%左右。

Nowadays,intelligent driving has become a development trend,and traffic sign recognition as an integral part of in-telligent driving has important research significance.In order to improve the classification accuracy of traffic sign recognition,this paper proposes the RI-Model model,which uses the classical neural network ResNet-50 and Inception-V3 for feature extraction,combined with the idea of feature fusion to improve the recognition rate of traffic signs.The Belgian traffic sign dataset(BelgiumTS)is used,and the correct recognition rate is tested using RI-Model after pre-processing the dataset.The results show that the method can achieve better convergence performance and good robustness in a shorter training time,and the recognition accuracy of RI-Mod-el on this dataset reaches 98.86%,which is about 1.5%higher than that of the direct use of ResNet-50 and Inception-V3.