基于感兴趣区域和HOG-MBLBP特征的交通标识检测

Traffic Sign Detection Based on Regions of Interest and HOG-MBLBP Features

交通标识检测中样本类别间的不平衡常常导致分类器的检测性能弱化,为了克服这一问题,该文提出一种基于感兴趣区域和HOG-MBLBP融合特征的交通标识检测方法。首先采用颜色增强技术分割提取出自然背景中交通标识所在的感兴趣区域;然后对标识样本库提取HOG-MBLBP融合特征,并用遗传算法对SVM交叉验证进行参数的优化选取,以此来训练和提升SVM分类器性能;最后将提取的感兴趣区域图像的HOG-MBLBP特征送入训练好的SVM多分类器,进行进一步的精确检测和定位,剔除误检区域。在自建的中国交通标识样本库上进行了实验,结果表明所提方法能达到99.2%的分类准确度,混淆矩阵结果也表明了该方法的优越性。

The imbalance between sample categories in traffic sign detection often results in the weakening of classification detection performance. To overcome this problem, a traffic sign detection method is proposed based on regions of interest and Histogram of Oriented Gradient and Multi-radius Block Local Binary Pattern（HOG-MBLBP） features. First, the color enhancement technology is used to segment and extract the regions of interest of the traffic signs captured in the natural background. Then HOG-MBLBP fusion features are extracted from traffic signs sample library. Moreover, genetic algorithm is used to optimize the parameters of Support Vector Machine（SVM） through cross-validation so as to train and promote SVM classifier performance. Finally, extracted HOG-MBLBP features of interest region images are put into the trained SVM multi-classifiers for further accurate detection and localization. By this method, the paper achieves the purpose of excluding false positives area. The experiments are carried out on the self-built Chinese traffic sign sample library, experimental results show that the proposed method can achieve 99.2% of classification accuracy, and the confusion matrix results also show the superiority of the proposed method.