SVM+HOG车牌检测含数据集

#include "opencv2/opencv.hpp" #define TRAIN_MODEL 0 void train(const std::string& posi_dir, const std::string& nega_dir, const std::string& save_pathname, cv::HOGDescriptor* hog_descriptor); #if TRAIN_MODEL int main(int argc, char** argv) { //定义正样本路径 const std::string positive_dir("./positive"); //定义负样本路径 const std::string negative_dir("./negative"); //目标检测模型保存路径和名称 const std::string save_pathname("./car_plane_detector.yaml"); //定义检测器的宽度和高度，必须是8的倍数，宽度和高度的比例大致为3：1 const int detector_width = 192; const int detector_heigt = 64; cv::Size detector_size(detector_width, detector_heigt); //定义HOG特征提取器 cv::HOGDescriptor hog_descriptor; //设置HOG特征提取器的尺寸 hog_descriptor.winSize = detector_size; //SVM模型训练 train(positive_dir, negative_dir, save_pathname, &hog_descriptor); return 0; } #else int main(int argc, char** argv) { ////搜索测试图像 //const std::string test_dir("./test"); //std::vector< cv::String > files; //cv::glob(test_dir + "/*.*", files); //std::vector<cv::Mat> list_image; //for (unsigned int i = 0; i < files.size(); i++) //{ // cv::Mat image = cv::imread(files[i], cv::IMREAD_COLOR); // cv::resize(image, image, cv::Size(1920, 1080)); // list_image.push_back(image); //} //目标检测模型保存路径和名称 const std::string save_pathname("./car_plane_detector.yaml"); //测试图像路径 const std::string test_dir("./test"); //导入目标检测模型 cv::HOGDescriptor hog_descriptor; hog_descriptor.load(save_pathname); //搜索测试图像 std::vector< cv::String > files; cv::glob(test_dir + "/*.*", files); for (unsigned int i = 0; i < files.size(); i++) { //读取测试图像 cv::Mat image = cv::imread(files[i]); //检测结果Box std::vector< cv::Rect > detections; hog_descriptor.detectMultiScale(image, detections); for (size_t j = 0; j < detections.size(); j++) { rectangle(image, detections[j], cv::Scalar(0, 255, 0), 2); printf("%d %d\n", detections[j].width, detections[j].height); } cv::imshow("检测结果", image); cv::waitKey(); } return 0; } #endif std::vector<cv::Mat> collectFeatureSet(const std::string& image_dir, cv::HOGDescriptor* hog_descriptor, const bool& is_positive) { //定义输出 std::vector<cv::Mat> feature_buffer; //搜索目录下的所有图像文件 std::vector<cv::String> image_file_list; cv::glob(image_dir + "/*.jpg", image_file_list); //定义随机变量 cv::RNG rng = cv::theRNG(); //读取图像并提取HOG特征 for (unsigned int i = 0; i < image_file_list.size(); i++) { cv::Mat image = cv::imread(image_file_list[i], cv::IMREAD_GRAYSCALE); if (is_positive) { //正样本缩放统一尺寸 cv::resize(image, image, hog_descriptor->winSize); //提取HOG特征 std::vector<float> descriptors; hog_descriptor->compute(image, descriptors); //缓存特征 feature_buffer.push_back(cv::Mat(descriptors).clone()); } else { //正负样本比例1：3 for (unsigned int j = 0; j < 3; j++) { //负样本随机裁剪 int crop_x = (image.cols - hog_descriptor->winSize.width - 1) * (float)rng; int crop_y = (image.rows - hog_descriptor->winSize.height - 1) * (float)rng; image = image(cv::Rect(crop_x, crop_y, hog_descriptor->winSize.width, hog_descriptor->winSize.height)).clone(); //提取HOG特征 std::vector<float> descriptors; hog_descriptor->compute(image, descriptors); //缓存特征 feature_buffer.push_back(cv::Mat(descriptors).clone()); } } } return feature_buffer; } void convertToML(const std::vector< cv::Mat >& train_samples, cv::Mat& train_data) { const int rows = (int)train_samples.size(); const int cols = (int)std::max(train_samples[0].cols, train_samples[0].rows); cv::Mat tmp(1, cols, CV_32FC1); train_data = cv::Mat(rows, cols, CV_32FC1); for (size_t i = 0; i < train_samples.size(); ++i) { CV_Assert(train_samples[i].cols == 1 || train_samples[i].rows == 1); if (train_samples[i].cols == 1) { transpose(train_samples[i], tmp); tmp.copyTo(train_data.row((int)i)); } else if (train_samples[i].rows == 1) { train_samples[i].copyTo(train_data.row((int)i)); } } } std::vector< float > getSvmDetector(const cv::Ptr< cv::ml::SVM >& svm) { cv::Mat sv = svm->getSupportVectors(); const int sv_total = sv.rows; cv::Mat alpha, svidx; double rho = svm->getDecisionFunction(0, alpha, svidx); CV_Assert(alpha.total() == 1 && svidx.total() == 1 && sv_total == 1); CV_Assert((alpha.type() == CV_64F && alpha.at<double>(0) == 1.) || (alpha.type() == CV_32F && alpha.at<float>(0) == 1.f)); CV_Assert(sv.type() == CV_32F); std::vector< float > hog_detector(sv.cols + 1); memcpy(&hog_detector[0], sv.ptr(), sv.cols * sizeof(hog_detector[0])); hog_detector[sv.cols] = (float)-rho; return hog_detector; } void train(const std::string& posi_dir, const std::string& nega_dir, const std::string& save_pathname, cv::HOGDescriptor* hog_descriptor) { //创建一个SVM分类器 cv::Ptr< cv::ml::SVM > svm = cv::ml::SVM::create(); svm->setTermCriteria(cv::TermCriteria(cv::TermCriteria::MAX_ITER + cv::TermCriteria::EPS, 1000, 1e-3)); //使用线性SVM svm->setKernel(cv::ml::SVM::LINEAR); svm->setType(cv::ml::SVM::C_SVC); //定义标签数组 std::vector< int > labels; //导入正样本，并进行特征提取 std::cout << "--->载入正样本..." << std::endl; std::vector<cv::Mat> feature_set = collectFeatureSet(posi_dir, hog_descriptor, true); labels.assign(feature_set.size(), +1); std::cout << "正样本数量:" << feature_set.size() << std::endl; //导入负样本，并进行特征提取 std::cout << "--->载入负样本..." << std::endl; std::vector<cv::Mat> nega_feature_buffer = collectFeatureSet(nega_dir, hog_descriptor, false); labels.insert(labels.end(), nega_feature_buffer.size(), -1); feature_set.insert(feature_set.end(), nega_feature_buffer.begin(), nega_feature_buffer.end()); std::cout << "负样本数量:" << nega_feature_buffer.size() << std::endl; cv::Mat train_data; convertToML(feature_set, train_data); cv::Ptr<cv::ml::TrainData> svm_train_auto_data = cv::ml::TrainData::create(train_data, cv::ml::SampleTypes::ROW_SAMPLE, labels); std::cout << "--->开启训练SVM..." << std::endl; svm->trainAuto(svm_train_auto_data); //保存目标检测模型 hog_descriptor->setSVMDetector(getSvmDetector(svm)); hog_descriptor->save(save_pathname); std::cout << "--->保存检测模型:" << save_pathname <<std::endl; }