OpenCV 应用边缘检测与霍夫线变换实现旋转角度检测

Myth丶恋晨 2022-09-30 06:10 202阅读 0赞

首先我们需要画出一个这样的图片，此时兴趣区域的角度为0度

![这里写图片描述][b3fbd84e2fe95eceac07dd8e26371dee.png]

使用opencv的图片旋转功能，把它旋转一个角度，旋转方法参考：  
http://blog.csdn.net/chaipp0607/article/details/63263347

逆时针旋转3.3度后，作为待检测图片。

![这里写图片描述][045b69fb07389b4e13a83086bad13e27.png]

由于我的原始图片很大2592\*2048，前面做了很多预处理工作，流程图：

![这里写图片描述][7bdc399448917e399b055ed1e50fec63.png]

double t = (double)getTickCount();
    	Mat SrcImage;
    	Mat ResizeImage;
    	Mat  ResizeroiImage;
    	Mat SrcroiImage;
    	Mat grayImage;
    	Mat thresholdImage;
    	Mat  roiimage;
    	Mat CannyImage;
    	Mat ThinImage;
    	double Srccutpointx,Srccutpointy,time;
    		int  anglenum = 0;
    		double angleall = 0;
    		SrcImage = imread("3_000_3.3.bmp");
    		cvtColor(SrcImage,grayImage,CV_BGR2GRAY);
    		Size dsize = Size(grayImage.cols*0.1,grayImage.rows*0.1);
    		resize(grayImage, ResizeImage,dsize);
    		ResizeroiImage = ResizeImage(Rect(ResizeImage.cols/3,ResizeImage.rows/3,ResizeImage.cols/3,ResizeImage.rows/3));
    		threshold(ResizeroiImage,thresholdImage, 0, 255, CV_THRESH_OTSU+CV_THRESH_BINARY);
    		vector<vector<Point> > contours;
    		vector<Vec4i>hierarchy;
    		findContours(thresholdImage,contours,hierarchy,CV_RETR_TREE,CV_CHAIN_APPROX_NONE);  
    		for(int k = 0; k < (int)contours.size(); k++)    //查找轮廓
    		{
    			if (contours.at(k).size()>100)
    			{
    				//把原图的rio区域涂成蓝色
    				// drawContours(SrcroiImage, contours, k, Scalar(255,0,0), CV_FILLED);
    				Rect position = boundingRect(contours.at(k));
    				//在缩小的图上画出芯片位置
    				// rectangle(
    				//ResizeImage,
    				//Point(position.tl().x+ResizeImage.cols/3, position.tl().y+ResizeImage.rows/3),
    				//Point(position.br().x+ResizeImage.cols/3, position.br().y+ResizeImage.rows/3),
    				//Scalar(0,0,255));
    
    				//在原图上画出芯片位置
    				// rectangle(
    				//SrcImage,
    				//Point(10*(position.tl().x+ResizeImage.cols/3)-30, 10*(position.tl().y+ResizeImage.rows/3)-30),
    				//Point(10*(position.br().x+ResizeImage.cols/3)+30, 10*(position.br().y+ResizeImage.rows/3)+30),
    				//Scalar(0,0,255));
    
    				//在原图上截取出芯片位置		
    				SrcroiImage =SrcImage(Rect(
    					10*(position.tl().x+ResizeImage.cols/3)-30,
    					10*(position.tl().y+ResizeImage.rows/3)-30,
    					10*position.width+60,
    					10*position.height+60));
    
    				Srccutpointx =	10*(position.tl().x+ResizeImage.cols/3)-30;
    				Srccutpointy =	10*(position.tl().y+ResizeImage.rows/3)-30;
    				//  roiimage = SrcImage(juxing);
    				//	  mu[k] = moments( contours[k], false );   
    				//		  mc[k] = Point2d( mu[k].m10/mu[k].m00 , mu[k].m01/mu[k].m00 ); 
    				//		  Point center = Point(mc[k].x,mc[k].y);  
    				//		  int r = 1;  
    				//		  circle(SrcImage,center,r,Scalar(0,0,255),-1);  
    				//		  printf("x=%f,y=%f\n",mc[k].x,mc[k].y);
    			}
    		}
    		Mat DelSrcroiImage;
    		Mat ThreholdsrcroiImage;
    		Mat graysrcroiImage;
    		SrcroiImage.copyTo(DelSrcroiImage);
    		cvtColor(SrcroiImage,graysrcroiImage,CV_BGR2GRAY);
    		blur( graysrcroiImage, graysrcroiImage, Size(7,7) );
    		threshold(graysrcroiImage,ThreholdsrcroiImage, 0, 255, CV_THRESH_OTSU+CV_THRESH_BINARY);
    		Canny(ThreholdsrcroiImage, CannyImage, 50, 200, 3);
    		vector<Vec4i> mylines;
    		HoughLinesP(CannyImage, mylines, 1, CV_PI/180, 100, 200, 10 );
    
    		for( size_t j = 0; j < mylines.size(); j++ )  
    		{
    			Vec4d l = mylines[j]; 
    
    			if (l[0]!=l[2]&&l[1]!=l[3])
    			{
    				anglenum ++;
    				Point start = Point (l[0], l[1]);  
    				Point end =  Point (l[2], l[3]);  	
    				line(DelSrcroiImage,start,end,Scalar(255));
    				double  duijixoabian = l[1]-l[3];
    				double  linbian = l[2]-l[0];
    				double  radian = atan(duijixoabian/linbian);
    				double  angle =(radian*180)/CV_PI;
    				angleall = angleall+angle;
    				cout<<"霍夫线检测角度="<<angle<<endl;
    			}
    		}
    		double angleaverage =angleall/anglenum;
    		cout<<"角度平个数="<<anglenum<<endl;
    		cout<<"角度平均值="<<angleaverage<<endl;
    
    	    t = (double)getTickCount() - t;
       	    printf("execution time = %lfms\n", t*1000. / getTickFrequency());
    	    waitKey(0);
    	    getchar();
    	    return 0;

结果：

![这里写图片描述][c0f285cae165c509e7b13b85b1e12bdf.png]

可以看到，结果的误差在0.3度左右，霍夫直线检测作为角度测量应用中的效果一般。

[b3fbd84e2fe95eceac07dd8e26371dee.png]: /images/20220708/3fab39d36ec646d58166141aa75fd64a.png
[045b69fb07389b4e13a83086bad13e27.png]: /images/20220708/42dfc35d421a47e6ab7d54c94bc80d4b.png
[7bdc399448917e399b055ed1e50fec63.png]: /images/20220708/6e58dea13e184b239d6d23447f645f2d.png
[c0f285cae165c509e7b13b85b1e12bdf.png]: /images/20220708/c1800b6f51634dbc82883add40069147.png