摘要
在道路交通标志的检测中,针对自然实景情况中拍摄到的图像存在的交通标志大小和位置不确定等困难问题,本文提出一种基于实景图像的多特征融合的道路交通标志检测方法。论文把样本分为了训练和测试样本,首先对训练样本图像进行盲复原处理;其次对复原处理后的图像进行自适应性的形状区域裁剪,提取裁剪区域图像的颜色、纹理和形状特征;再次分别对颜色、纹理和形状特征进行SVM分类检测,从而获得颜色、纹理和形状三个分类模型;最后对模型的权值进行自适应性计算,得到加权的特征融合模型。通过测试样本对模型的检测,结果表明特征融合识别方法有很高的准确度,另外对比实验得到的数据显示融合模型提高了道路交通检测的准确度和鲁棒性。
In the detection of road traffic signs,this study proposes a road traffic signs feature detection methods based on many featured fusion in image,which carefully analysis the characteristics and problems of road traffic signs,for example,images in real traffic road traffic signs detecting are always distorted and the size of signs as well as position is uncertain.The samples of study are divided into training samples and testing samples.Firstly,the study make a blind restoration process with the images of training sample,Secondly, we cut the recovered images according to their own color,texture and shape features.Thirdly,color,texture and shape features are respectively detected with SVM classification,so we can get these classification models respectively.At last,the study get a weighted feature fusion model which the weighted value of color,texture and shape features for the adaptive weighted feature fusion model. After the model is tested by these test samples,the results show that the feature fusion recognition method has a very high accuracy.In addition,the contrast data of the comparative experiments show that fused features can effectively improve the accuracy and robustness of traffic detecting.
出处
《信号处理》
CSCD
北大核心
2011年第10期1616-1620,共5页
Journal of Signal Processing
