特征融合与PSO优化SVM的交通标志分类

Traffic Sign Recognition Based on Feature Fusion and PSO Optimized SVM

摘要: 为有效提高交通标志分类的准确度，提出一种基于特征融合与PSO优化支持向量机分类器参数的交通标志分类方法。首先提取标志图像形状的局部HOG特征信息和描述图像粗略轮廓的全局Gist特征信息，采用并行的方式把这两种特征融合；再通过PSO算法优化SVM分类器参数，提高SVM分类器分类性能；最后SVM分类器进行交通标志训练与分类。试验结果表明：本文并行的特征融合与PSO优化SVM参数分类方法精确度比单一的PHOG和Gist特征的交通标志分类方法分别提高9.048%、0.952%，比串行的特征融合方法提高了8.095%。

Abstract: In order to effectively improve the accuracy of traffic sign classification, a traffic sign classification method based on feature fusion and PSO optimized support vector machine classifier parameters is proposed. First extract the local HOG feature information of the logo image shape and the global Gist feature information describing the rough outline of the image, and merge these two features in a parallel manner; then optimize the SVM classifier parameters through the PSO algorithm to improve the SVM classifier classification performance; finally the SVM classifier performs traffic sign training and classification. The experimental results show that the accuracy of the parallel feature fusion and PSO-optimized SVM parameter classification method is 9.048% and 0.952% higher than the traffic sign classification method of the single PHOG and Gist features, respectively, and 8.095% higher than the serial feature fusion method.