基于opencv的SVM车牌号码识别模型训练（C++）QT

条件：opencv-4.5，QT

一、准备数据集

下载车牌相关字符样本用于训练和测试，本文使用14个汉字样本和34个数字跟字母样本，每个字符样本数为40，样本尺寸为28*28。

二、计算样本HOG特征

方向梯度直方图(Histogram of Oriented Gradient, HOG)特征是一种在计算机视觉和图像处理中用来进行物体检测的特征描述子。它通过计算和统计图像局部区域的梯度方向直方图来构成特征。具体内容详见以下链接：

HOG

OpenCV中的HOGDescriptor类可用于计算HOG特征。在计算HOG特征前需确定HOGDescriptor实例初始化参数，如下图，为HOGDescriptor类hog实例初始化设置参数。HOG特征描述子参数之间必须满足一定的关系，确保能够遍历整个特征提取窗口。

（1）blockSize的宽度和高度要分别能被cellSize的宽度和高度整除。

（ 2 ） winSize 减去 blockSize 的宽度和高度后，要能被blockStride对应的宽度和高度整除。

由于本例所使用样本为28*28，为满足以上要求，各项参数设置如下：

HOGDescriptor hog(Size(28,28),Size(4,4),Size(8,8),Size(4,4),9,1,-1);

调用HOGDescriptor类的compute方法计算HOG特征，得到特征描述子descriptors（列向量）

hog.compute(image,descriptors);//image为读取的单个样本，灰度图片

将descriptors写入push_back到temp(Mat) ,再将temp转为行向量，然后temp压入trainData，最后形成（行列数为样本数*特征数的矩阵），这里trainData大小为1920*144。

temp.push_back(descriptors);
trainData.push_back(temp.reshape(0,1));

然后需要生成标签向量，opencv SVM 接收类型为int的向量作为responses,

vector<int> trainLabel;
trainLabel.insert(trainLabel.begin()+ int(40*i),40,int(i));//i为对应的标签数据

三、设置SVM参数

由于是做分类任务，因此类型选择C_SVC，然后选择高斯核（RBF）作为kernel，因而需要设置参数Gamma

    cv::Ptr<cv::ml::SVM> SVMmodel = cv::ml::SVM::create();SVMmodel->setType(SVM::C_SVC);SVMmodel->setKernel(SVM::RBF);SVMmodel->setC(48);SVMmodel->setGamma(0.5);

设好参数后开始训练模型

    SVMmodel->train(traindata,cv::ml::ROW_SAMPLE,trainlabel);SVMmodel->save("SVM_Recognization.xml");//保存模型

在模型训练后，本文通过测试集测试其识别的准确率为92%。在实际车牌识别中也可以准确识别（车牌提取，分割处理后）

效果如下：

QString MainWindow::svmTrain(cv::Mat &src)
{using namespace std;using namespace cv;using namespace cv::ml;string province[14] = {"云","皖","苏","辽","闽","黑","京","川","沪","浙","湘","粤","陕","鲁"};string numcha[34] = {"0","1","2","3","4","5","6","7","8","9","A","B","C","D","E","F","G","H",
"J","K","L","M","N","P","Q","R","S","T","U","V","W","X","Y","Z"};Mat traindata,testData,testSample;std::vector<int> trainlabel;src.copyTo(testData);HogGenerate(traindata,trainlabel,testData,testSample);cv::Ptr<cv::ml::SVM> SVMmodel = cv::ml::SVM::create();SVMmodel->setType(SVM::C_SVC);SVMmodel->setKernel(SVM::RBF);SVMmodel->setC(48);SVMmodel->setGamma(0.5);cout << "svm !!! " << endl;SVMmodel->train(traindata,cv::ml::ROW_SAMPLE,trainlabel);SVMmodel->save("SVM_Recognization.xml");//保存模型文件到项目的目录下Mat result;cout << "testSample Row:" << testSample.rows << " cols: " << testSample.cols << endl;SVMmodel->predict(testSample,result);int preIndex = int(result.at<float>(0));if(preIndex<=13){return QString::fromStdString(province[preIndex]);}else{return QString::fromStdString(numcha[preIndex-14]);}
}

void MainWindow::HogGenerate(cv::Mat &trainData, std::vector<int> &trainLabel,cv::Mat &testData,cv::Mat &testSample)
{using namespace cv;using namespace std;vector<string> PlateChar;string province[14] = {"云","皖","苏","辽","闽","黑","京","川","沪","浙","湘","粤","陕","鲁"};string numcha[34] = {"0","1","2","3","4","5","6","7","8","9","A","B","C","D","E","F","G","H",
"J","K","L","M","N","P","Q","R","S","T","U","V","W","X","Y","Z"};size_t samples = 48;string Chapath = "/home/ghoson-x/Desktop/qt/CarImage/data/train_28_28/";string Chipath = "/home/ghoson-x/Desktop/qt/CarImage/data2/train_28_28/";vector<float> descriptors;HOGDescriptor hog(Size(28,28),Size(4,4),Size(8,8),Size(4,4),9,1,-1);string imagePath,imageName;for(size_t i=0;i<samples;++i){if(i<=13){imageName = province[i];imagePath = Chipath;}else{imageName = numcha[i-14];imagePath = Chapath;}imagePath.append(imageName).append("/").append(imageName).append("_");for(int j=1;j<=40;++j){string * ImagePath = new string;ImagePath->append(imagePath);ImagePath->append((QString::number(j)).toStdString()).append(".jpg");Mat tmp = imread(ImagePath->data(),IMREAD_GRAYSCALE);delete ImagePath;if(tmp.empty()){break;}Mat temp;hog.compute(tmp,descriptors);temp.push_back(descriptors);trainData.push_back(temp.reshape(0,1));temp.release();tmp.release();}trainLabel.insert(trainLabel.begin()+ int(40*i),40,int(i));}Mat test;hog.compute(testData,descriptors);test.push_back(descriptors);testSample.push_back(test.reshape(0,1));
}

基于opencv的SVM车牌号码识别模型训练（C++）QT相关推荐

python的快车智能车牌号码识别系统的开发
It is not always possible to prove some crimes. To achieve this, modern tools may be needed which mu ...
java毕设项目车牌号码识别系统开源了，很好玩，建议尝试
1.介绍这是一个基于spring boot + maven + opencv 实现的Demo教程项目贯穿样本处理.模型训练.图像处理.对象检测.对象识别等技术点,以学习交流为目的,代码注释超多,文档 ...
基于python+OpenCV的车牌号码识别
基于python+OpenCV的车牌号码识别车牌识别行业已具备一定的市场规模,在电子警察.公路卡口.停车场.商业管理.汽修服务等领域已取得了部分应用.一个典型的车辆牌照识别系统一般包括以下4个部分: ...
python基础教程：python+OpenCV实现车牌号码识别
这篇文章主要介绍了python+OpenCV实现车牌号码识别,文中示例代码介绍的非常详细,具有一定的参考价值,感兴趣的小伙伴们可以参考一下基于python+OpenCV的车牌号码识别,供大家参考,具 ...
基于OpenCV的 SVM算法实现数字识别(四)---代码实现
三.基于SVM算法实现手写数字识别作为一个工科生,而非数学专业的学生,我们研究一个算法,是要将它用于实际领域的.下面给出基于OpenCV3.0的SVM算法手写数字识别程序源码(参考http://bl ...
matlab的车牌照识别,基于MatLab车牌号码识别算法研究与
<基于MatLab车牌号码识别算法研究与>由会员分享,可在线阅读,更多相关<基于MatLab车牌号码识别算法研究与(15页珍藏版)>请在人人文库网上搜索. 1.基于,MatLa ...
基于LDA+SVM实现人脸识别模型
基于LDA+SVM实现人脸识别模型描述人脸识别(图像识别)是机器学习领域十经典的应用,在本质上,人脸识别属于监督学习中的分类问题.前面章节中我们已经学习了支持向量机(SVM),该算法在图像分类领域 ...
《熟练掌握OpenCV----实用计算机视觉工程案例》第5章车牌号码识别
人工智能系列经典图书翻译原名:<Mastering OpenCV with Practical Computer Vision Projects> 作者:Shervin Em ...
python读取视频流做人脸识别_基于OpenCV和Keras实现人脸识别系列——二、使用OpenCV通过摄像头捕获实时视频并探测人脸、准备人脸数据...
基于OpenCV和Keras实现人脸识别系列手记: 项目完整代码参见Github仓库. 本篇是上面这一系列手记的第二篇. 在Opencv初接触,图片的基本操作这篇手记中,我介绍了一些图片的基本操作,而 ...

基于opencv的SVM车牌号码识别模型训练（C++）QT

基于opencv的SVM车牌号码识别模型训练（C++）QT相关推荐

最新文章

热门文章