基于融合特征和BP网络的交通标志识别方法

摘要

Actual road traffic signs in the environment where the background is very complex often show varying degrees of geometric distortion and deformation.All of these factors influence the identification process.To solve the problem and improve the recognition rate,a fusion based on wavelet moment invariant and Gabor filter characteristics combined with BP neural network traffic sign recognition method was designed.A rotating translational invariance of the wavelet moment invariant shape feature extraction was used,and principal component analysis (PCA) was used to extract main features.Gabor filter was used to extract image texture feature.The two kinds of characteristics were fused and inputted to BP neural network training for identifying the target image.Experimental results show that the method possesses higher accuracy compared with the single feature recognition.%由于实际道路环境中的交通标志处于非常复杂的背景中,且往往会出现不同程度的几何失真和形变,对识别过程造成很大影响,为解决该问题,提高识别率,设计一种基于小波不变矩和Gabor滤波的融合特征与BP神经网络相结合的交通标志识别方法.采用具有旋转以及平移不变性的小波不变矩提取形状特征,用PCA主成分分析法提取主要特征;对图像进行Gabor滤波,对滤波后的输出图像提取纹理特征;将两种特征融合,送入BP神经网络训练测试,对目标图像进行识别分类.实验结果表明,该方法相比单特征识别具有更高的准确率.