OpenCV3的接口變化挺大的，原來OpenCV2.4.X版本的SVM不能用了。
OpenCV3中使用SVM方法如下：
1, 注意其中訓練和自動訓練的接口，還有l(wèi)abelMat一定要用CV_32SC1的類型。
Ptr<SVM> svm = SVM::create(); svm->setType(SVM::C_SVC); svm->setKernel(SVM::RBF); TermCriteria criteria = cvTermCriteria(CV_TERMCRIT_ITER + CV_TERMCRIT_EPS, 1000, FLT_EPSILON); svm->setTermCriteria(criteria); Mat labelMat1(labelMat.rows, labelMat.cols, CV_32SC1); for (int i = 0; i < labelMat.rows; i++){ for (int j = 0; j < labelMat.cols; j++){ labelMat1.at<int>(i, j) = labelMat.at<float>(i, j); } } //svm->train(trainMat, ROW_SAMPLE, labelMat); Ptr<TrainData> traindata = ml::TrainData::create(trainMat, ROW_SAMPLE, labelMat1); svm->trainAuto(traindata, 10); svm->save("svm.xml");1234567891011121314151617181234567891011121314151617181，注意load模型文件的時候用法。
#include <iostream>#include <fstream>#include <opencv2/core/core.hpp>#include <opencv2/highgui/highgui.hpp>#include <opencv2/imgPRoc/imgproc.hpp>#include <opencv2/objdetect/objdetect.hpp>#include <opencv2/ml/ml.hpp>using namespace std;using namespace cv;class MySVM : public ml::SVM{public: //獲得SVM的決策函數(shù)中的alpha數(shù)組 double get_svm_rho() { return this->getDecisionFunction(0, svm_alpha, svm_svidx); } //獲得SVM的決策函數(shù)中的rho參數(shù),即偏移量 vector<float> svm_alpha; vector<float> svm_svidx; float svm_rho;};int main(){ namedWindow("src", 0); //檢測窗口(64,128),塊尺寸(16,16),塊步長(8,8),cell尺寸(8,8),直方圖bin個數(shù)9 //HOGDescriptor hog(Size(64, 128), Size(16, 16), Size(8, 8), Size(8, 8), 9);//HOG檢測器，用來計算HOG描述子的 int DescriptorDim;//HOG描述子的維數(shù)，由圖片大小、檢測窗口大小、塊大小、細胞單元中直方圖bin個數(shù)決定 //Ptr svm = ml::SVM::create(); Ptr<ml::SVM>svm = ml::SVM::load<ml::SVM>("svm.xml"); DescriptorDim = svm->getVarCount();//特征向量的維數(shù)，即HOG描述子的維數(shù) Mat supportVector = svm->getSupportVectors();//支持向量的個數(shù) int supportVectorNum = supportVector.rows; cout << "支持向量個數(shù)：" << supportVectorNum << endl; //------------------------------------------------- vector<float> svm_alpha; vector<float> svm_svidx; float svm_rho; svm_rho = svm->getDecisionFunction(0, svm_alpha, svm_svidx); //------------------------------------------------- Mat alphaMat = Mat::zeros(1, supportVectorNum, CV_32FC1);//alpha向量，長度等于支持向量個數(shù) Mat supportVectorMat = Mat::zeros(supportVectorNum, DescriptorDim, CV_32FC1);//支持向量矩陣 Mat resultMat = Mat::zeros(1, DescriptorDim, CV_32FC1);//alpha向量乘以支持向量矩陣的結(jié)果 supportVectorMat = supportVector; ////將alpha向量的數(shù)據(jù)復制到alphaMat中 //double * pAlphaData = svm.get_alpha_vector();//返回SVM的決策函數(shù)中的alpha向量 for (int i = 0; i < supportVectorNum; i++) { alphaMat.at<float>(0, i) = svm_alpha[i]; } //計算-(alphaMat * supportVectorMat),結(jié)果放到resultMat中 //gemm(alphaMat, supportVectorMat, -1, 0, 1, resultMat);//不知道為什么加負號？ resultMat = -1 * alphaMat * supportVectorMat; //得到最終的setSVMDetector(const vector& detector)參數(shù)中可用的檢測子 vector<float> myDetector; //將resultMat中的數(shù)據(jù)復制到數(shù)組myDetector中 for (int i = 0; i < DescriptorDim; i++) { myDetector.push_back(resultMat.at<float>(0, i)); } //最后添加偏移量rho，得到檢測子 myDetector.push_back(svm_rho); cout << "檢測子維數(shù)：" << myDetector.size() << endl; //設(shè)置HOGDescriptor的檢測子 HOGDescriptor myHOG; //myHOG.setSVMDetector(HOGDescriptor::getDefaultPeopleDetector()); myHOG.setSVMDetector(myDetector); //myHOG.setSVMDetector(HOGDescriptor::getDefaultPeopleDetector()); /**************讀入圖片進行HOG行人檢測******************/ //Mat src = imread("00000.jpg"); //Mat src = imread("2007_000423.jpg"); Size s1(128, 128); Size s2(64, 64); myHOG.winSize = s1; myHOG.blockSize = s1; myHOG.blockStride = s1; myHOG.cellSize = s2; myHOG.nbins = 9; Mat frame; while (true) { Mat src = imread("2.jpg"); vector<Rect> found, found_filtered;//矩形框數(shù)組 //cout << "進行多尺度HOG人體檢測" << endl; myHOG.detectMultiScale(src, found, 0, Size(32, 32), Size(32, 32), 1.05, 2);//對圖片進行多尺度行人檢測 //cout << "找到的矩形框個數(shù)：" << found.size() << endl; //找出所有沒有嵌套的矩形框r,并放入found_filtered中,如果有嵌套的話,則取外面最大的那個矩形框放入found_filtered中 for (int i = 0; i < found.size(); i++) { Rect r = found[i]; int j = 0; for (; j < found.size(); j++) if (j != i && (r & found[j]) == r) break; if (j == found.size()) found_filtered.push_back(r); } //畫矩形框，因為hog檢測出的矩形框比實際人體框要稍微大些,所以這里需要做一些調(diào)整 for (int i = 0; i < found_filtered.size(); i++) { Rect r = found_filtered[i]; r.x += cvRound(r.width*0.1); r.width = cvRound(r.width*0.8); r.y += cvRound(r.height*0.07); r.height = cvRound(r.height*0.8); rectangle(src, r.tl(), r.br(), Scalar(255, 255, 255), 3); } imshow("src", src); waitKey(0);//注意：imshow之后必須加waitKey，否則無法顯示圖像 }}轉(zhuǎn)自：http://blog.csdn.net/heroacool/article/details/50579955新聞熱點
疑難解答
