基于HOG-Gabor特征融合与Softmax分类器的交通标志识别方法

发布时间：2018-03-26 04:17

  本文选题：交通信息工程　切入点：智能车　出处：《交通运输工程学报》2017年03期

【摘要】：为了提高交通标志识别的正确率和实时性,提出了一种基于HOG-Gabor特征融合与Softmax分类器的交通标志识别方法。采用Gamma矫正方法提取HOG特征,采用对比度受限的自适应直方图均衡化方法提取Gabor特征,基于线性特征融合原理,将提取的HOG和Gabor特征向量直接串联,得到刻画交通标志的融合特征向量,采用Softmax分类器对融合特征向量进行分类,采用德国交通标志识别基准(GTSRB)数据库测试了所提方法的有效性,比较了基于单特征与融合特征的交通标志识别效果。试验结果表明:在图像增强过程中,针对HOG特征,采用Gamma矫正方法的分类正确率最大,为97.11%,针对Gabor特征,采用限制对比度的直方图均衡化方法的分类正确率最大,为97.54%;采用Softmax分类器的最小分类正确率为97.11%,耗时小于2s;针对HOG-Gabor融合特征,采Softmax分类器的识别率高达97.68%,因此,基于HOG-Gabor特征融合与Softmax分类器的交通标志识别方法的识别率高,实时性强。
[Abstract]:In order to improve the accuracy and real time of traffic sign recognition, a traffic sign recognition method based on HOG-Gabor feature fusion and Softmax classifier is proposed. Gamma correction method is used to extract HOG features. The adaptive histogram equalization method with limited contrast is used to extract Gabor features. Based on the principle of linear feature fusion, the extracted HOG and Gabor feature vectors are connected in series directly, and the fused feature vectors depicting traffic signs are obtained. The fusion feature vector is classified by Softmax classifier, and the validity of the proposed method is tested by using the German Traffic sign recognition benchmark (GTSRB) database. The effect of traffic sign recognition based on single feature and fusion feature is compared. The experimental results show that in the process of image enhancement, the correct rate of Gamma correction method is 97.11 for HOG feature, and 97.11 for Gabor feature. The histogram equalization method with restricted contrast has the highest classification accuracy of 97.54; the minimum classification accuracy with Softmax classifier is 97.11 and the time consuming is less than 2 s; for the HOG-Gabor fusion feature, the recognition rate of the Softmax classifier is as high as 97.68, so the recognition rate of the Softmax classifier is as high as 97.68. The traffic sign recognition method based on HOG-Gabor feature fusion and Softmax classifier has high recognition rate and high real-time performance.
【作者单位】： 长安大学信息工程学院;广东省特种设备检测研究院珠海检测院;神龙汽车有限公司;伊利诺伊大学芝加哥分校土木与材料工程系;
【基金】：国家自然科学基金项目(61201406) 中央高校基本科研业务费专项资金项目(310824162022)
【分类号】：TP391.41;U463.6
，

本文编号：1666296