opencv图像角点提取

全栈程序员站长

发布于 2022-09-27 09:44:54

5300

发布于 2022-09-27 09:44:54

opencv角点检测(二)

改进的Harris角点检测算法

<span style="font-size:18px;">harris角点检测算法的结果一定程度上取决于系数k，有人对Harris的角点检测算法进行了改进，直接利用像素点协方差矩阵的特征值提取角点。而且不在进行非极大值抑制，而是采用一种容忍距离的形式，在角点的一定范围内只有一个角点。</span>

<span style="font-size:18px;">具体原理：首先计算图像每个像素点的协方差矩阵，并求取对应的特征值，将最小的特征值最大的那个像素点作为第一个角点（具体来说，就是求出每个像素点的协方差矩阵对应的特征值，找出最小的那个，所有最小的特征值中哪个最大，就将哪个所对应的像素点作为角点max(min(e1,e2)),e1、e2为像素协方差矩阵的特征值）。然后依次按照最大最小特征值的顺序寻找角点，并保证在容忍距离内只有一个角点。</span>

<span style="font-size:18px;">opencv测试代码：</span>

<span style="font-size:18px;"></span><pre name="code" class="cpp">#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp><img src="https://img-blog.csdn.net/20150322224206153?watermark/2/text/aHR0cDovL2Jsb2cuY3Nkbi5uZXQvdTAxNDI2MDg5Mg==/font/5a6L5L2T/fontsize/400/fill/I0JBQkFCMA==/dissolve/70/gravity/Center" alt="" /><img src="https://img-blog.csdn.net/20150322224206153?watermark/2/text/aHR0cDovL2Jsb2cuY3Nkbi5uZXQvdTAxNDI2MDg5Mg==/font/5a6L5L2T/fontsize/400/fill/I0JBQkFCMA==/dissolve/70/gravity/Center" alt="" />
#include <opencv2/imgproc/imgproc.hpp>

#include <iostream>
using namespace std;
using namespace cv;

Mat src,src_gray;
int cornernum = 20;
int maxnum = 200;
char* windowname = "Imagecorners";

void goodFeaturesdemo(int,void*);
int main(int argc,char* argv[])
{
	src = imread("road.jpg");
	cvtColor(src,src_gray,CV_BGR2GRAY);//将图像转化为灰度图；

	namedWindow(windowname,CV_WINDOW_AUTOSIZE);
	createTrackbar("CornersNum:",windowname,&cornernum,maxnum,goodFeaturesdemo);//创建控制条，与cornernum变量相        //关联

	imshow(windowname,src);
	goodFeaturesdemo(0,0);

	waitKey(0);
	return 0;
}

void goodFeaturesdemo(int,void*)
{
	if(cornernum < 1){ cornernum = 1;}

	vector<Point2f> points;
	double qualityLevel = 0.02;
	double minDistance = 10;
	int blockSize = 3;
	bool useHarrisDetector = false;
	double k = 0.04;

	Mat copy;
	copy = src.clone();
        //进行角点检测
	goodFeaturesToTrack(src_gray,              //要进行检测的图像
		                points,            //存储检测到的角点坐标，Point2f类型
		                cornernum,         //检测到的角点的最大数目
				qualityLevel,      //角点的阈值条件，即角点的质量等级：qualityLevel*max(min(e1,e2))
				minDistance,       //容忍距离，单位像素
				Mat(),
				blockSize,
				useHarrisDetector,
				k);
	cout<<"Number of corners detected: "<<points.size()<<endl;
	int r = 4;
	for(int i = 0; i < points.size(); i++)
	{
		circle(copy,points[i],r,Scalar(0,0,255),1,8);
	}

	//namedWindow(windowname,CV_WINDOW_AUTOSIZE);
	imshow(windowname,copy);
}

检测结果如下：

<img src="https://img-blog.csdn.net/20150322224206153?watermark/2/text/aHR0cDovL2Jsb2cuY3Nkbi5uZXQvdTAxNDI2MDg5Mg==/font/5a6L5L2T/fontsize/400/fill/I0JBQkFCMA==/dissolve/70/gravity/Center" alt="" />

<span style="font-size:18px;">可以看出检测到的角点比较均匀，不会在一个小区域产生角点聚集的情况。</span>

<span style="font-size:18px;">
</span>

<span style="font-size:18px;"></span><pre name="code" class="cpp">定制自己的角点检测算法：

opencv提供了求取特征值和特征向量的函数，可以实现自己设计的角点提取算法，主要包括下面两个函数：

cornerEigenCalsAndVecs：计算像素对应的特征值和特征向量；

cornerMinEigenVal：求取像素点最小的特征值；

下面程序实现了一个定制化的Harris角点检测算法，和类似的Shi-Tomsi算法：

#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/imgproc/imgproc.hpp>

#include <iostream>
using namespace std;
using namespace cv;

Mat src,src_gray;
Mat myHarris_dst,myHarris_copy,Mc;
Mat myShiTomsi_dst,myShiTomsi_copy;

int myST_qualityLevel = 50;
int myH_qualityLevel = 50;
int max_qualityLevel =100;

double myHarris_minVal,myHarris_maxVal;
double myST_minVal,myST_maxVal;

RNG rng(2444);

char* myHarris = "myHarris Corner:";
char* myST = "myST Corner:";

void myHarris_function(int,void*);
void myST_function(int,void*);

int main(int argc,char* argv[])
{
	src = imread("road.jpg");

	cvtColor(src,src_gray,CV_BGR2GRAY);

	int blocksize = 3;
	int aperture = 3;
	myHarris_dst = Mat::zeros(src_gray.size(),CV_32FC(6));
	Mc =  Mat::zeros(src_gray.size(),CV_32FC1);

	cornerEigenValsAndVecs(src_gray,myHarris_dst,blocksize,aperture,BORDER_DEFAULT);//求取像素点特征值和特征向         
        //量，具体参数意义见用户手册；论坛上应该有。
	for( int i = 0; i < src_gray.rows; i++)
	{
		for(int j = 0; j < src_gray.cols; j++)
		{
			float lambda_1 = myHarris_dst.at<Vec6f>(i,j)[0];//取出两个特征值，后几位是特征向量；
			float lambda_2 = myHarris_dst.at<Vec6f>(i,j)[1];
			Mc.at<float>(i,j) = lambda_1*lambda_2 - 0.04*pow((lambda_1+lambda_2),2);//harris角点检测法
		}
	}

	minMaxLoc(Mc,&myHarris_minVal,&myHarris_maxVal);//求出最大值和最小值，用来筛选角点；

	namedWindow(myHarris,CV_WINDOW_AUTOSIZE);
	createTrackbar("Quality Level",myHarris,&myH_qualityLevel,max_qualityLevel,myHarris_function);
	myHarris_function(0,0);

	myShiTomsi_dst = Mat::zeros(src_gray.size(),CV_32FC1);
	cornerMinEigenVal(src_gray,myShiTomsi_dst,blocksize,aperture,4);

	minMaxLoc(myShiTomsi_dst,&myST_minVal,&myST_maxVal);
	namedWindow(myST,CV_WINDOW_AUTOSIZE);
	createTrackbar("Quality Level",myST,&myST_qualityLevel,max_qualityLevel,myST_function);
	myST_function(0,0);

	waitKey();
	return 0;
}

void myHarris_function(int,void*)
{
	myHarris_copy = src.clone();

	if(myST_qualityLevel < 1)
	{ myST_qualityLevel = 1;}

	for(int j = 0; j < src_gray.rows ; j++)
	{
		for(int i = 0; i < src_gray.cols ; i++)
		{
		    if( Mc.at<float>(j,i) > myHarris_minVal + (float)(myHarris_maxVal - myHarris_minVal)*
			myH_qualityLevel/max_qualityLevel)
			     circle(myHarris_copy,Point(i,j),3,Scalar(rng.uniform(0,255),rng.uniform(0,255),rng.uniform(0,255)),-1,8,0);//在检测到的焦点位                             //置画圆
		}
	}
	imshow(myHarris,myHarris_copy);
}

void myST_function(int,void*)
{
	myShiTomsi_copy = src.clone();

	for(int i = 0; i < src.rows; i++)
	{
		for(int j = 0; j < src.cols ; j++)
		{
			if(myShiTomsi_dst.at<float>(i,j) > myST_minVal + (myST_maxVal - myST_minVal)*
				myH_qualityLevel/max_qualityLevel)
				circle(myShiTomsi_copy,Point(j,i),3,Scalar(rng.uniform(0,255),rng.uniform(0,255),rng.uniform(0,255)),-1,8,0);
		}
	}
	imshow(myST,myShiTomsi_copy);
}

定制化的Harris焦点检测结果：

<span style="font-size:18px;"><img src="https://img-blog.csdn.net/20150322224743120?watermark/2/text/aHR0cDovL2Jsb2cuY3Nkbi5uZXQvdTAxNDI2MDg5Mg==/font/5a6L5L2T/fontsize/400/fill/I0JBQkFCMA==/dissolve/70/gravity/Center" alt="" />
</span>

<span style="font-size:18px;">定制化的Shi-Tomsi算法检测结果：</span>

<span style="font-size:18px;"><img src="https://img-blog.csdn.net/20150322224845286?watermark/2/text/aHR0cDovL2Jsb2cuY3Nkbi5uZXQvdTAxNDI2MDg5Mg==/font/5a6L5L2T/fontsize/400/fill/I0JBQkFCMA==/dissolve/70/gravity/Center" alt="" />
</span>

<span style="font-size:18px;">
</span>

<span style="font-size:18px;">亚像素角点检测：</span>

<span style="font-size:18px;"></span><pre name="code" class="cpp">opencv提供函数cornerSubPixel函数进行亚像素精度的角点检测。具体代码如下：

<pre name="code" class="cpp">#include <iostream>
using namespace std;
using namespace cv;

Mat src,src_gray;
int cornersnum = 20;
int maxnum = 50;
RNG rng(342);

char* imgwindow = "Image";

void subPixelCorner(int,void*);
int main(int argc,char* argv[])
{
	src = imread("cat.jpg");

	cvtColor(src,src_gray,CV_BGR2GRAY);
	namedWindow(imgwindow,CV_WINDOW_AUTOSIZE);
	createTrackbar("Corners NUM:",imgwindow,&cornersnum,maxnum,subPixelCorner);
	subPixelCorner(0,0);

	//imshow(imgwindow,src);

	waitKey(0);
	return 0;
}

void subPixelCorner(int,void*)
{
	if(cornersnum < 1) { cornersnum = 1;}

	Mat copy = src.clone();

	vector<Point2f> points;
	double qualityLevel = 0.03;
	double minDistance = 10;
	int blocksize = 3;
	bool useHarrisDetector = false;
	double k = 0.045;

	goodFeaturesToTrack(src_gray,
		            points,
		            cornersnum,
			    qualityLevel,
			    minDistance,
			    Mat(),
			    blocksize,
			    useHarrisDetector,
			    k);

	cout<<"Number of detected corners:"<<points.size()<<endl;
	int r = 4;
	for( int i = 0; i < points.size(); i++)
	{
		circle(copy,points[i],r,Scalar(rng.uniform(0,255),rng.uniform(0,255),rng.uniform(0,255)),-1,8,0);
	}

	imshow(imgwindow,copy);

	Size size = Size(5,5);
	Size zerosize = Size(-1,-1);

	TermCriteria criteria = TermCriteria(CV_TERMCRIT_EPS + CV_TERMCRIT_ITER,30,0.01);//设置迭代结束的条件
	cornerSubPix(src_gray,points,size,zerosize,criteria);//亚像素精度角点检测

	for(int j = 0; j < points.size(); j++)
	{ cout<<"Refined Corner ["<<j<<"] : ("<<points[j].x<<","<<points[j].y<<")"<<endl;}
}

cornernum阈值为36时的运行结果：

重新定位的亚像素精度的角点：

<img src="https://img-blog.csdn.net/20150322225130178?watermark/2/text/aHR0cDovL2Jsb2cuY3Nkbi5uZXQvdTAxNDI2MDg5Mg==/font/5a6L5L2T/fontsize/400/fill/I0JBQkFCMA==/dissolve/70/gravity/Center" alt="" />

至此，利用opencv检测角点的算法总结完毕，本文的程序是在VC2010+opencv2.4.9+win7下可直接运行；下篇总结不同的特征检测算子。敬请期待！

<span style="font-size:18px;">
</span>

发布者：全栈程序员栈长，转载请注明出处：https://javaforall.cn/179735.html原文链接：https://javaforall.cn

本文参与腾讯云自媒体同步曝光计划，分享自作者个人站点/博客。

如有侵权请联系 cloudcommunity@tencent.com 删除

java

https

网络安全

本文分享自作者个人站点/博客前往查看

如有侵权，请联系 cloudcommunity@tencent.com 删除。

本文参与腾讯云自媒体同步曝光计划，欢迎热爱写作的你一起参与！

java

https

网络安全

登录后参与评论

0 条评论

热度

opencv图像角点提取

opencv图像角点提取

社区

活动

资源

关于

腾讯云开发者

热门产品

热门推荐

更多推荐