结构光:光条中心线提取-Steger算法(基于Hessian矩阵)

光条中心线提取-Steger算法(基于Hessian矩阵)

参考以下博客进行了修改,修改了灰度阈值以及最后标注红线的部分。https://blog.csdn.net/dangkie/article/details/78996761#comments

void StegerLine()
{
	Mat img0 = imread("D:\\结构光实验数据\\2019-11-20\\1_nl.bmp", 1);
	Mat img;
	cvtColor(img0, img0, CV_BGR2GRAY);
	img = img0.clone();

	//高斯滤波
	img.convertTo(img, CV_32FC1);
	GaussianBlur(img, img, Size(0, 0), 6, 6);

	//一阶偏导数
	Mat m1, m2;
	m1 = (Mat_<float>(1, 2) << 1, -1);  //x偏导
	m2 = (Mat_<float>(2, 1) << 1, -1);  //y偏导

	Mat dx, dy;
	filter2D(img, dx, CV_32FC1, m1);
	filter2D(img, dy, CV_32FC1, m2);

	//二阶偏导数
	Mat m3, m4, m5;
	m3 = (Mat_<float>(1, 3) << 1, -2, 1);   //二阶x偏导
	m4 = (Mat_<float>(3, 1) << 1, -2, 1);   //二阶y偏导
	m5 = (Mat_<float>(2, 2) << 1, -1, -1, 1);   //二阶xy偏导

	Mat dxx, dyy, dxy;
	filter2D(img, dxx, CV_32FC1, m3);
	filter2D(img, dyy, CV_32FC1, m4);
	filter2D(img, dxy, CV_32FC1, m5);

	//hessian矩阵
	double maxD = -1;
	int imgcol = img.cols;
	int imgrow = img.rows;
	vector<double> Pt;
	for (int i = 0; i<imgcol; i++)
	{
		for (int j = 0; j<imgrow; j++)
		{
			if (img0.at<uchar>(j, i)>50)
			{
				Mat hessian(2, 2, CV_32FC1);
				hessian.at<float>(0, 0) = dxx.at<float>(j, i);
				hessian.at<float>(0, 1) = dxy.at<float>(j, i);
				hessian.at<float>(1, 0) = dxy.at<float>(j, i);
				hessian.at<float>(1, 1) = dyy.at<float>(j, i);

				Mat eValue;
				Mat eVectors;
				eigen(hessian, eValue, eVectors);

				double nx, ny;
				double fmaxD = 0;
				if (fabs(eValue.at<float>(0, 0)) >= fabs(eValue.at<float>(1, 0)))  //求特征值最大时对应的特征向量
				{
					nx = eVectors.at<float>(0, 0);
					ny = eVectors.at<float>(0, 1);
					fmaxD = eValue.at<float>(0, 0);
				}
				else
				{
					nx = eVectors.at<float>(1, 0);
					ny = eVectors.at<float>(1, 1);
					fmaxD = eValue.at<float>(1, 0);
				}

				double t = -(nx*dx.at<float>(j, i) + ny*dy.at<float>(j, i)) / (nx*nx*dxx.at<float>(j, i) + 2 * nx*ny*dxy.at<float>(j, i) + ny*ny*dyy.at<float>(j, i));

				if (fabs(t*nx) <= 0.5 && fabs(t*ny) <= 0.5)
				{
					Pt.push_back(i);
					Pt.push_back(j);
				}
			}
		}
	}


	cvtColor(img0, img0, CV_GRAY2BGR);
	for (int k = 0; k<Pt.size() / 2; k++)
	{
		Point rpt;
		rpt.x = Pt[2 * k + 0];
		rpt.y = Pt[2 * k + 1];
		circle(img0, rpt, 1, Scalar(0, 0, 255));
	}

	namedWindow("result", CV_WINDOW_NORMAL);
	imshow("result", img0);
	imwrite("output.bmp", img0);
	
	waitKey(0);
}

修改后的接口


vector<Point> StegerLine(cv::Mat& InputImg)
{
	Mat img0 = InputImg;//此为原图,有背景
	Mat img;//此为处理图
	img0.copyTo(img);//拷贝矩阵

	//高斯滤波
	img.convertTo(img, CV_32FC1);
	GaussianBlur(img, img, Size(0, 0), 6, 6);

	//一阶偏导数
	Mat m1, m2;
	m1 = (Mat_<float>(1, 2) << 1, -1);  //x偏导
	m2 = (Mat_<float>(2, 1) << 1, -1);  //y偏导

	Mat dx, dy;
	filter2D(img, dx, CV_32FC1, m1);
	filter2D(img, dy, CV_32FC1, m2);

	//二阶偏导数
	Mat m3, m4, m5;
	m3 = (Mat_<float>(1, 3) << 1, -2, 1);   //二阶x偏导
	m4 = (Mat_<float>(3, 1) << 1, -2, 1);   //二阶y偏导
	m5 = (Mat_<float>(2, 2) << 1, -1, -1, 1);   //二阶xy偏导

	Mat dxx, dyy, dxy;
	filter2D(img, dxx, CV_32FC1, m3);
	filter2D(img, dyy, CV_32FC1, m4);
	filter2D(img, dxy, CV_32FC1, m5);

	//hessian矩阵
	double maxD = -1;
	int imgcol = img.cols;
	int imgrow = img.rows;
	vector<double> Pt;
	for (int i = 0; i<imgcol; i++)
	{
		for (int j = 0; j<imgrow; j++)
		{
			if (img0.at<uchar>(j, i)>50)
			{
				Mat hessian(2, 2, CV_32FC1);
				hessian.at<float>(0, 0) = dxx.at<float>(j, i);
				hessian.at<float>(0, 1) = dxy.at<float>(j, i);
				hessian.at<float>(1, 0) = dxy.at<float>(j, i);
				hessian.at<float>(1, 1) = dyy.at<float>(j, i);

				Mat eValue;
				Mat eVectors;
				eigen(hessian, eValue, eVectors);

				double nx, ny;
				double fmaxD = 0;
				if (fabs(eValue.at<float>(0, 0)) >= fabs(eValue.at<float>(1, 0)))  //求特征值最大时对应的特征向量
				{
					nx = eVectors.at<float>(0, 0);
					ny = eVectors.at<float>(0, 1);
					fmaxD = eValue.at<float>(0, 0);
				}
				else
				{
					nx = eVectors.at<float>(1, 0);
					ny = eVectors.at<float>(1, 1);
					fmaxD = eValue.at<float>(1, 0);
				}

				double t = -(nx*dx.at<float>(j, i) + ny*dy.at<float>(j, i)) / (nx*nx*dxx.at<float>(j, i) + 2 * nx*ny*dxy.at<float>(j, i) + ny*ny*dyy.at<float>(j, i));

				if (fabs(t*nx) <= 0.5 && fabs(t*ny) <= 0.5)
				{
					Pt.push_back(i);
					Pt.push_back(j);
				}
			}
		}
	}

	cvtColor(img0, img0, CV_GRAY2BGR);//不需要验证时可注释掉
	vector<Point> LaserLinePtSet;//输出的点集
	for (int k = 0; k<Pt.size() / 2; k++)
	{
		Point rpt;
		rpt.x = Pt[2 * k + 0];
		rpt.y = Pt[2 * k + 1];
		LaserLinePtSet.push_back(rpt);//将点加入到输出点集集合中
		circle(img0, rpt, 1, Scalar(0, 0, 255));
	}
	//以下部分为验证代码
	namedWindow("result", CV_WINDOW_NORMAL);
	imshow("result", img0);
	imwrite("output.bmp", img0);

	return LaserLinePtSet;
}

你可能感兴趣的:(算法)