基于HOG特征的实景交通标志检测

Baced on the Descriptors of HOG in Reality Images Traffic Sign Detection

针对实景道路拍摄到的交通标志大小不一、光照不均、位置不定等问题,提出了一种基于HOG特征与级联分类器的交通标志检测方法。该方法从训练集的正负样本分别提取HOG特征,完成Adaboost级联分类器的训练;然后针对实景拍摄到的道路图像,把测试样本先进行HSV空间的转化、高斯滤波和直方图均衡等预处理,最后对测试图像进行区域HOG描述子的提取并检测出图像中可能存在的交通标志。实验表明,该方法对亮度不均、大小不一的实景图像交通标志检测具有较高的检测率,对道路环境变化有较强的鲁棒性。

To deal with the problems of different scales of signs,nonuniform intensity of illumination and in-determination of the position of signs in traffic sign detection with real images,this study proposes a traffic sign detection method based on HOG descriptors and Adaboost algorithm.Firstly,this study takes the data-sets of GTSDB and divides the samples into training samples and testing samples.Secondly,the extracted HOG descriptors of positive training samples and negative training samples are combined to train the Adaboost cascade classifier.Thirdly,the real images are transformed from RGB space to HSV space for the histogram equalization of treatment in the value of luminance.Finally,the algorithm takes the HOG descriptors of ROI and scales the’traffic signs’in the image.Experiment results show that this method has a good detection accuracy and an outstanding robust performance in reality images tracffic sign detection.