9-1-2車牌訓練辨識SVM

9-1-2車牌訓練辨識SVM

程式說明:

運行結果:

   這邊我們將上述切割出來的圖片，手動進行車牌或非車牌的辨識，
   先利用SVM進行訓練，並且放兩張圖來做測試
   可以看到結果p的部分為1，成功辨識出來。

範例程式:

#include <iostream>
#include <opencv2/opencv.hp>
#include <opencv2/highgui/highgui.hpp>
#include <fstream>

int main(int argc, char** argv)
{
    int imgHeight = 33;
    int imgWidth = 110;

    std::string imgFileNameBatch = "fileName.txt";

    cv::Mat trainingImages;
    std::vector trainingLabels;

    std::fstream fileBatch(imgFileNameBatch, std::ios::in);

    if (!fileBatch)
       return -1;

    while (!fileBatch.eof())
    {
       std::string fileName;
       fileBatch >> fileName;
       std::cout << "fileName: " << fileName << std::endl;

       cv::Mat imgTmp = cv::imread(fileName, 0);
       if (imgTmp.empty())
          return -1;
       int label = -1;
       fileBatch >> label;
       std::cout << "label: " << label << std::endl;

       imgTmp = imgTmp.reshape(1, 1);
       trainingImages.push_back(imgTmp);
       trainingLabels.push_back(label);
    }

    fileBatch.close();

    cv::Mat trainingData;
    cv::Mat classes;

    cv::Mat(trainingImages).copyTo(trainingData);
    trainingData.convertTo(trainingData, CV_32FC1);
    cv::Mat(trainingLabels).copyTo(classes);

    cv::FileStorage fso("SVM.xml", cv::FileStorage::WRITE);
    fso << "TrainingData" << trainingData;
    fso << "classes" << classes;
    fso.release();

    cv::FileStorage fs;
    fs.open("SVM.xml", cv::FileStorage::READ);
    cv::Mat SVM_TrainingData;
    cv::Mat SVM_Classes;
    fs["TrainingData"] >> SVM_TrainingData;
    fs["classes"] >> SVM_Classes;

    std::cout << "svmClassifier1: " << std::endl;

    //Set SVM params
    CvSVMParams SVM_params;
    SVM_params.svm_type = CvSVM::C_SVC;
    SVM_params.kernel_type = CvSVM::LINEAR; //CvSVM::LINEAR;
    SVM_params.degree = 0;
    SVM_params.gamma = 1;
    SVM_params.coef0 = 0;
    SVM_params.C = 1;
    SVM_params.nu = 0;
    SVM_params.p = 0;
    SVM_params.term_crit = cvTermCriteria(CV_TERMCRIT_ITER, 1000, 0.01);
    //Train SVM
    CvSVM svmClassifier(SVM_TrainingData, SVM_Classes, cv::Mat(), cv::Mat(), SVM_params);

    std::cout << "svmClassifier2: " << std::endl;

    cv::Mat imgReCheckPlate = cv::imread("plate.bmp", 0);
    cv::Mat imgReCheckNoPlate = cv::imread("noplate.bmp", 0);
    cv::Mat p = imgReCheckPlate.reshape(1, 1);
    cv::Mat pN = imgReCheckNoPlate.reshape(1, 1);
    p.convertTo(p, CV_32FC1);
    pN.convertTo(pN, CV_32FC1);

    int responsep = (int)svmClassifier.predict(p);
    int responsepN = (int)svmClassifier.predict(pN);

    std::cout << "p result is : " << responsep << std::endl;
    std::cout << "pN result is : " << responsepN << std::endl;
    return 0;
}