opencv--使用opencv实现旋转角度的模板匹配

下面的例子是简单的使用opencv 实现的模板匹配流程，其中时间性能和精确度还需要调整，如果直接使用会出问题，所以这个只是例子，根据代码原理可以实现尺度变化的模板匹配和旋转尺度变化同时，具体根据实现的旋转代码进一步实现，但是就结果而言和halcon的模板匹配差距较大，性能更不行，因此仅供参考，同时本人整理了两个使用各种加速的高性能版本，有兴趣的可以下载看看

struct MatchResult
{std::vector<cv::Point> points;double angle;double score;MatchResult(std::vector<cv::Point> points, double angle, double score) :points(points), angle(angle), score(score) {};
};//旋转图像
cv::Mat ImageRotate(cv::Mat image, double angle)
{cv::Mat newImg;cv::Point2f pt = cv::Point2f((float)image.cols / 2, (float)image.rows / 2);cv::Mat M = cv::getRotationMatrix2D(pt, angle, 1.0);cv::warpAffine(image, newImg, M, image.size());return newImg;
}
std::vector<cv::Point> GetRotatePoints(cv::Size size, double angle) {// 定义模板图像的中心点cv::Point2f center(size.width / 2.0, size.height / 2.0);// 计算旋转矩阵cv::Mat rotationMatrix = cv::getRotationMatrix2D(center, angle, 1.0);// 定义模板图像的四个顶点std::vector<cv::Point2f> srcPoints = {cv::Point2f(0, 0),cv::Point2f(size.width, 0),cv::Point2f(size.width, size.height),cv::Point2f(0, size.height)};// 存储旋转后的四个顶点std::vector<cv::Point2f> dstPoints(4);// 进行仿射变换cv::transform(srcPoints, dstPoints, rotationMatrix);// 将结果转换为cv::Point类型并返回std::vector<cv::Point> resultPoints(4);for (int i = 0; i < 4; ++i) {resultPoints[i] = cv::Point(cvRound(dstPoints[i].x), cvRound(dstPoints[i].y));}return resultPoints;
}/*
旋转模板匹配函数（通过图像金字塔、增大旋转步长来提升匹配速度）
Mat src：原图像
Mat model：模板图
double startAngle：旋转的最小角
double endAngle：旋转的最大角
double firstStep：角度旋转时的最大步长
double secondStep：角度旋转时的最小步长
int numLevels = 0：图像金字塔缩放次数
*/
MatchResult rotateMatch(cv::Mat src, cv::Mat model, double startAngle, double endAngle, double firstStep, double secondStep, int numLevels = 0) {//对模板图像和待检测图像分别进行图像金字塔下采样for (int i = 0; i < numLevels; i++) {cv::pyrDown(src, src, cv::Size(src.cols / 2, src.rows / 2));cv::pyrDown(model, model, cv::Size(model.cols / 2, model.rows / 2));}cv::Mat rotatedImg, result;double score = -1;cv::Point location;double angle;bool isSecond = false;while (true) {for (double curAngle = startAngle; curAngle <= endAngle; curAngle += firstStep) {rotatedImg = ImageRotate(model, curAngle);//imshow("rotated", rotatedImg);//imshow("src-pyrDown", src);//waitKey();matchTemplate(src, rotatedImg, result, cv::TM_CCOEFF_NORMED);double minval, maxval;cv::Point minloc, maxloc;cv::minMaxLoc(result, &minval, &maxval, &minloc, &maxloc);if (maxval > score){location = maxloc;score = maxval;angle = curAngle;}}if (isSecond && firstStep<= secondStep) break;startAngle = angle - firstStep;endAngle = angle + firstStep;if ((endAngle - startAngle) / 5 > secondStep) {firstStep = (endAngle - startAngle) / 5;}else {firstStep = secondStep;isSecond = true;}}cv::Point finalPoint = cv::Point(location.x * pow(2, numLevels), location.y * pow(2, numLevels));std::vector<cv::Point> points = GetRotatePoints(cv::Size(model.cols * pow(2, numLevels), model.rows * pow(2, numLevels)), angle);for (int j = 0; j < points.size(); j++){points[j].x += finalPoint.x;points[j].y += finalPoint.y;}return MatchResult(points, angle, score);
}int main() {//读取所有图像std::vector<cv::Mat> imgs;std::string imageName;std::string path = "E:\\prj\\shape_based_matching-master\\test\\board\\test";std::vector<std::string> img_paths;cv::glob(path, img_paths);for (auto& p : img_paths){cv::Mat img = cv::imread(p);imgs.push_back(img);}cv::Mat templateImg = cv::imread("E:\\prj\\shape_based_matching-master\\test\\board\\train.png");cv::Rect box(cv::Point(135, 120), cv::Point(470, 365));//cv::rectangle(drawFrame, box, cv::Scalar(0, 255, 0), 2);templateImg = templateImg(box).clone();int i = 0;for (cv::Mat img : imgs){i += 1;MatchResult matchResult = rotateMatch(img, templateImg, 0, 360, 30, 0.1, 0);std::vector<cv::Point> points = matchResult.points;std::cout << i << "- 角度：" << matchResult.angle << std::endl;std::cout << i << "- 得分：" << matchResult.score << std::endl;cv::line(img, points[0], points[1], cv::Scalar(255, 0, 0), 2);cv::line(img, points[1], points[2], cv::Scalar(255, 0, 0), 2);cv::line(img, points[2], points[3], cv::Scalar(255, 0, 0), 2);cv::line(img, points[3], points[0], cv::Scalar(255, 0, 0), 2);cv::Point pt1 = cv::Point((points[0].x + points[3].x) / 2, (points[0].y + points[3].y) / 2);cv::Point pt2 = cv::Point((points[1].x + points[2].x) / 2, (points[1].y + points[2].y) / 2);cv::arrowedLine(img, pt2, pt1, cv::Scalar(0, 0, 255), 2);cv::imshow("img_" + std::to_string(i), img);cv::waitKey(0);}return 0;
}

结果如下：