opencv摄像机标定（附源码已调试）

opencv摄像机标定（附源码已调试）

最近在左摄像机标定，要求反投影的误差在1个像素内，发现matlab工具箱更好，opencv比较好些因为有了现成的函数。

以下是用opencv2.1安装包只带的照片做的实验，运行的平台为vs2008+opencv2.1 ，windows xp

#include <cv.h>
#include <highgui.h>
#include <iostream>

using namespace std;
void PrintMat(CvMat*);
void FputMat(FILE *,CvMat *);
int main(int argc ,char * argv[])
{
 /*读入图像*/
 CvSize image_size;
 int n_board=9;//图像数目
 int sn_board=0;//成功找到角点的图像数目
 int board_w=6;
 int board_h=9;
 int board_n=board_h*board_w;//每幅图像的角点数
 CvSize patter_size=cvSize(board_w,board_h);//每幅图像的角点数
 CvPoint2D32f * corners=new CvPoint2D32f[board_n];//一幅图像的角点数组
 CvMat * object_points= cvCreateMat(board_n*n_board,3,CV_32FC1);
 CvMat * image_points=cvCreateMat(board_n*n_board,2,CV_32FC1);
 CvMat * point_counts=cvCreateMat(n_board,1,CV_32SC1);
 for(int i=1;i<=n_board;i++)
 {
  /*读入图像*/
  char path[100]="F:\\vs2008test\\mycalib_0\\right0";
  char num[10];
  itoa(i,num,10);
  strcat(num,".jpg");
  IplImage *SourceImg=cvLoadImage(strcat(path,num),CV_LOAD_IMAGE_COLOR);
  image_size=cvGetSize(SourceImg);//图像的大小
  IplImage *SourceImgGray=cvCreateImage(image_size,IPL_DEPTH_8U,1);

  cvCvtColor(SourceImg,SourceImgGray,CV_BGR2GRAY);
  cvNamedWindow("MyCalib",CV_WINDOW_AUTOSIZE);
  cvShowImage("MyCalib",SourceImg);
  cvWaitKey(NULL);

  /*提取角点精确到亚像素*/

  int corner_count;

  if(0==cvFindChessboardCorners(SourceImgGray,patter_size,corners,&corner_count,CV_CALIB_CB_ADAPTIVE_THRESH|CV_CALIB_CB_FILTER_QUADS))

  { cout<<"......无法找出第"<<i<<"幅图的角点"<<endl;
  cvWaitKey(NULL);
  cvReleaseImage(&SourceImgGray);
  cvReleaseImage(&SourceImg);
  continue;
  //return -1;
  }
  else
  {
   cvFindCornerSubPix(SourceImgGray,corners,corner_count,cvSize(11,11),cvSize(-1,-1),cvTermCriteria(
    CV_TERMCRIT_EPS+CV_TERMCRIT_ITER,30,0.1));//CvSize win 和matlab标定工具箱类似
   cvDrawChessboardCorners(SourceImg,patter_size,corners,corner_count,1);
   cvShowImage("MyCalib",SourceImg);
   //cvSaveImage("F:\\vs2008test\\mycalib_0\\11.jpg",SourceImg);
  } 
  cvWaitKey(0);

  for (int j=0;j<board_n;j++)
  {

   CV_MAT_ELEM(*image_points,float,sn_board*board_n+j,0)=corners[j].x;
   CV_MAT_ELEM(*image_points,float,sn_board*board_n+j,1)=corners[j].y;
   CV_MAT_ELEM(*object_points,float,sn_board*board_n+j,0)=float(j/board_w);
   CV_MAT_ELEM(*object_points,float,sn_board*board_n+j,1)=float(j%board_w);
   CV_MAT_ELEM(*object_points,float,sn_board*board_n+j,2)=0.0f;
   CV_MAT_ELEM(*point_counts,int,sn_board,0)=board_n;
  }
  sn_board++;
  cvReleaseImage(&SourceImgGray);
  cvReleaseImage(&SourceImg);
  cout<<"......成功找出第"<<i<<"幅图的角点"<<endl;
 }
 cout<<"......一共成功获得"<<sn_board<<"幅图像的角点"<<"......无法获得"<<n_board-sn_board<<"幅图像的角点"<<endl;
 //重新赋值
 CvMat * object_points0= cvCreateMat(board_n*sn_board,3,CV_32FC1);
 CvMat * image_points0=cvCreateMat(board_n*sn_board,2,CV_32FC1);
 CvMat * point_counts0=cvCreateMat(sn_board,1,CV_32SC1);
 for (int i=0;i<sn_board*board_n;i++)
 {

  CV_MAT_ELEM(*image_points0,float,i,0)=CV_MAT_ELEM(*image_points,float,i,0);
  CV_MAT_ELEM(*image_points0,float,i,1)=CV_MAT_ELEM(*image_points,float,i,1);
  CV_MAT_ELEM(*object_points0,float,i,0)=CV_MAT_ELEM(*object_points,float,i,0);
  CV_MAT_ELEM(*object_points0,float,i,1)=CV_MAT_ELEM(*object_points,float,i,1);
  CV_MAT_ELEM(*object_points0,float,i,2)=0.0f;
 }
 for (int i=0;i<sn_board;i++)
 {
  CV_MAT_ELEM(*point_counts0,int,i,0)=CV_MAT_ELEM(*point_counts,int,i,0);
 }
 cvReleaseMat(&object_points);
 cvReleaseMat(&point_counts);
 cvReleaseMat(&image_points); 
 /*摄像机标定并求得内部参数*/

 CvMat * camera_matrix=cvCreateMat(3,3,CV_32FC1);
 CvMat * distortion_coeffs=cvCreateMat(1,4,CV_32FC1);
 CvMat * rotation_vectors=cvCreateMat(sn_board,3,CV_32FC1);
 CvMat * translation_vectors=cvCreateMat(sn_board,3,CV_32FC1);
 int flags=0;

 

 cvCalibrateCamera2(object_points0,image_points0,point_counts0,image_size,camera_matrix
  //,distortion_coeffs,NULL,NULL,flags);
  ,distortion_coeffs,rotation_vectors,translation_vectors,0);
 //输出结果
 printf("/*****摄像机内部参数*****/\n");
 PrintMat(camera_matrix);
 printf("/*****畸变参数k1,k2,p1,p2*****/\n");
 PrintMat(distortion_coeffs);
 cvWaitKey(NULL);
 //保存数据
 cout<<"......保存内部参数与畸变参数"<<endl;
 cvSave("camera_matrix1111.xml",camera_matrix);
 cvSave("distortion_coeffs.xml",distortion_coeffs);

 FILE *fp=NULL;
 if ((fp=fopen("F:\\vs2008test\\mycalib_0\\mycalib_0\\camera_matrix.txt","w"))==NULL)
 {
  printf("无法打开文件！");
  exit(1);
 }
 FputMat(fp,camera_matrix);
 fclose(fp);
 if ((fp=fopen("F:\\vs2008test\\mycalib_0\\mycalib_0\\distortion_coeffs.txt","w"))==NULL)
 {
  printf("无法打开文件！");
  exit(1);
 }
 FputMat(fp,distortion_coeffs);
 fclose(fp);

 //误差分析
 CvMat * object_points2=cvCreateMat(board_n,3,CV_32FC1);
 CvMat * image_points2=cvCreateMat(board_n,2,CV_32FC1);
 CvMat * rotation_vectors2=cvCreateMat(1,3,CV_32FC1);
 CvMat * translation_vectors2=cvCreateMat(1,3,CV_32FC1);
 CvMat * Err=cvCreateMat(sn_board*board_n,2,CV_32FC1);
 for(int k=0;k<sn_board;k++)
 {
  for (int i=0;i<board_n;i++)//取一幅图的数据
  {
   CV_MAT_ELEM(*object_points2,float,i,0)=CV_MAT_ELEM(*object_points0,float,k*board_n+i,0);
   CV_MAT_ELEM(*object_points2,float,i,1)=CV_MAT_ELEM(*object_points0,float,k*board_n+i,1);
   CV_MAT_ELEM(*object_points2,float,i,2)=0.0f;
  }
  for (int i=0;i<3;i++)
  {
   CV_MAT_ELEM(*rotation_vectors2,float,0,i)=CV_MAT_ELEM(*rotation_vectors,float,k,i);
   CV_MAT_ELEM(*translation_vectors2,float,0,i)=CV_MAT_ELEM(*translation_vectors,float,k,i);
  }
  cvProjectPoints2(object_points2,rotation_vectors2,translation_vectors2,camera_matrix,
   distortion_coeffs,image_points2);
  for (int i=0;i<board_n;i++)
  {
   CV_MAT_ELEM(*Err,float,k*board_n+i,0)=CV_MAT_ELEM(*image_points0,float,k*board_n+i,0)-CV_MAT_ELEM(*image_points2,float,i,0);
   CV_MAT_ELEM(*Err,float,k*board_n+i,1)=CV_MAT_ELEM(*image_points0,float,k*board_n+i,1)-CV_MAT_ELEM(*image_points2,float,i,1);
  }
 }
 //PrintMat(Err);
 CvMat * Err_abs=cvCreateMat(board_n*sn_board,2,CV_32FC1);
 cvAbs(Err,Err_abs);
 CvScalar Dmean;
 CvScalar Ddev;
 cvAvgSdv(Err_abs,&Dmean,&Ddev);
 printf("/*****反投影误差分析*****/\n");
 cout<<"......绝对值误差矩阵的平均值："<<endl;
 cout<<Dmean.val[0]<<endl;
 cout<<"......绝对值误差矩阵的标准差："<<endl;
 cout<<Ddev.val[0]<<endl;
 cout<<"......保存误差矩阵"<<endl;
 cvSave("err.xml",Err);
 if ((fp=fopen("F:\\vs2008test\\mycalib_0\\mycalib_0\\err.txt","w"))==NULL)
 {
  printf("无法打开文件！");
  exit(1);
 }
 FputMat(fp,Err);
 fclose(fp);
 
 cvReleaseMat (&object_points2);
 cvReleaseMat (&image_points2);
 cvReleaseMat (&rotation_vectors2);
 cvReleaseMat (&translation_vectors2);
 cvReleaseMat (&Err);
 cvReleaseMat (&Err_abs);

 //释放总变量
 cvWaitKey(NULL);
 cvReleaseMat(&object_points0);
 cvReleaseMat(&point_counts0);
 cvReleaseMat(&image_points0);
 cvReleaseMat(&camera_matrix);
 cvReleaseMat(&distortion_coeffs);
 cvReleaseMat(&rotation_vectors);
 cvReleaseMat(&translation_vectors);
 delete [] corners;
 cvDestroyWindow("MyCalib");
}

void PrintMat(CvMat* arry)
{
 for (int i=0;i<arry->rows;i++)
 {
  for (int j=0;j<arry->cols;j++)
  {
   printf("%f",CV_MAT_ELEM(*arry,float,i,j));
   if(j<arry->cols-1)
    printf(",");
  }
  printf("\n");
 }
}
void FputMat(FILE * fp,CvMat * arry)
{
 for (int i=0;i<arry->rows;i++)
 {
  for (int j=0;j<arry->cols;j++)
  {
   fprintf(fp,"%f",CV_MAT_ELEM(*arry,float,i,j));
   if(j<arry->cols-1)
    fprintf(fp,",");
  }
  fprintf(fp,"\n");
 }
}

 

实验结果

9幅图像中其中一幅图像

标定过程出项的问题：

opencv的cvCalibrateCamera2  之用一幅图像也得到标定结果，这点和摄像机标定的理论有也不符，至少应在在2幅图像。

个人觉得cvFindChessboardCorners这个函数不是十分的智能，比如上例中第7、8图就无法找出，另外你把标定板倒一个，cvFindChessboardCorners也会找不出来，因为行数和列数倒个了。

相对于matlab标定工具箱来说，虽然matlab工具箱没有那么智能，但是把所有图像的角点找出来是没有问题的而却精度比opencv的还要高

原文 http://hi.baidu.com/lin65505578/item/fcb30f34031003f9e7bb7aa8