我正在实现O'reilly书的opencv教科书中示例中给出的立体视觉深度映射。
实现此代码时，我的摄像机都是水平的，而且都不相同。执行代码后，我得到非常奇怪的结果。我想验证这些怪异的结果是否仅是由于照相机的方向错误或代码内是否有任何问题，请帮帮我。
#include "cv.h"
#include "cxmisc.h"
#include "highgui.h"
//#include "cvaux.h"
#include <vector>
#include <string>
#include <algorithm>
#include <stdio.h>
#include <ctype.h>
#define WIDTH 426
#define HEIGHT 320
using namespace std;
int main()
{
//---------Initial--------
int nx=7, ny=7, frame = 0, n_boards =20, N;
int count1 = 0,count2 = 0, result1=0, result2=0;
int showUndistorted = 1, successes1 = 0,successes2 = 0 ;
const int maxScale = 1;
const float squareSize = 2.f; //Set this to your actual square size
CvSize imageSize = {WIDTH,HEIGHT};
CvCapture *capture1= NULL, *capture2= NULL;
CvSize board_sz = cvSize( nx,ny );
int i, j, n = nx*ny, N1 = 0, N2 = 0;
vector<CvPoint2D32f> points[2];
vector<int> npoints;
vector<CvPoint3D32f> objectPoints;
vector<CvPoint2D32f> temp1(n);
vector<CvPoint2D32f> temp2(n);
double M1[3][3], M2[3][3], D1[5], D2[5];
double R[3][3], T[3], E[3][3], F[3][3];
double Q[4][4];
CvMat _Q = cvMat(4,4, CV_64F, Q);
CvMat _M1 = cvMat(3, 3, CV_64F, M1 );
CvMat _M2 = cvMat(3, 3, CV_64F, M2 );
CvMat _D1 = cvMat(1, 5, CV_64F, D1 );
CvMat _D2 = cvMat(1, 5, CV_64F, D2 );
CvMat _R = cvMat(3, 3, CV_64F, R );
CvMat _T = cvMat(3, 1, CV_64F, T );
CvMat _E = cvMat(3, 3, CV_64F, E );
CvMat _F = cvMat(3, 3, CV_64F, F );
//---------Starting WebCam----------
capture1= cvCaptureFromCAM(0);
assert(capture1!=NULL); cvWaitKey(0);
capture2= cvCaptureFromCAM(1);
assert(capture2!=NULL);
//assure capture size is correct...
int res=cvSetCaptureProperty(capture1,CV_CAP_PROP_FRAME_WIDTH,WIDTH);
printf("%d",res);
res=cvSetCaptureProperty(capture1,CV_CAP_PROP_FRAME_HEIGHT,HEIGHT);
printf("%d",res);
res=cvSetCaptureProperty(capture2,CV_CAP_PROP_FRAME_WIDTH,WIDTH);
printf("%d",res);
res=cvSetCaptureProperty(capture2,CV_CAP_PROP_FRAME_HEIGHT,HEIGHT);
printf("%d",res); fflush(stdout);
IplImage *frame1 = cvQueryFrame( capture1 );
IplImage* gray_fr1 = cvCreateImage( cvGetSize(frame1), 8, 1 );
IplImage *frame2 = cvQueryFrame( capture2 );
IplImage* gray_fr2 = cvCreateImage( cvGetSize(frame1), 8, 1 );
//imageSize = cvGetSize(frame1);
//Show Window
cvNamedWindow( "camera2", 1 );
cvNamedWindow( "camera1", 1 );
cvNamedWindow("corners camera1",1);
cvNamedWindow("corners camera2",1);
while((successes1<n_boards)||(successes2<n_boards))
{
//--------Find chessboard corner--------------------------------------------------
if((frame++ % 20) == 0)
{
//----------------CAM1-------------------------------------------------------------------------------------------------------
result1 = cvFindChessboardCorners( frame1, board_sz,&temp1[0], &count1,CV_CALIB_CB_ADAPTIVE_THRESH|CV_CALIB_CB_FILTER_QUADS);
cvCvtColor( frame1, gray_fr1, CV_BGR2GRAY );
//----------------CAM2--------------------------------------------------------------------------------------------------------
result2 = cvFindChessboardCorners( frame2, board_sz,&temp2[0], &count2,CV_CALIB_CB_ADAPTIVE_THRESH|CV_CALIB_CB_FILTER_QUADS);
cvCvtColor( frame2, gray_fr2, CV_BGR2GRAY );
if(count1==n&&count2==n&&result1&&result2)
{
cvFindCornerSubPix( gray_fr1, &temp1[0], count1,cvSize(11, 11), cvSize(-1,-1),cvTermCriteria(CV_TERMCRIT_ITER+CV_TERMCRIT_EPS,30, 0.01) );
cvDrawChessboardCorners( frame1, board_sz, &temp1[0], count1, result1 );
cvShowImage( "corners camera1", frame1 );
N1 = points[0].size();
points[0].resize(N1 + n, cvPoint2D32f(0,0));
copy( temp1.begin(), temp1.end(), points[0].begin() + N1 );
++successes1;
cvFindCornerSubPix( gray_fr2, &temp2[0], count2,cvSize(11, 11), cvSize(-1,-1),cvTermCriteria(CV_TERMCRIT_ITER+CV_TERMCRIT_EPS,30, 0.01) );
cvDrawChessboardCorners( frame2, board_sz, &temp2[0], count2, result2 );
cvShowImage( "corners camera2", frame2 );
N2 = points[1].size();
points[1].resize(N2 + n, cvPoint2D32f(0,0));
copy( temp2.begin(), temp2.end(), points[1].begin() + N2 );
++successes2;
putchar('$');
}
else
{ cvShowImage( "corners camera2", gray_fr2 );
cvShowImage( "corners camera1", gray_fr1 );
}
frame1 = cvQueryFrame( capture1 );
cvShowImage("camera1", frame1);
frame2 = cvQueryFrame( capture2 );
cvShowImage("camera2", frame2);
if(cvWaitKey(15)==27) break;
}
}
cvReleaseCapture( &capture1 );
cvReleaseCapture( &capture2 );
cvDestroyWindow("camera1");
cvDestroyWindow("camera2");
cvDestroyWindow("corners camera1");
cvDestroyWindow("corners camera2");
printf("\n");
//--------------Calibaration-------------------
N = n_boards*n;
objectPoints.resize(N);
for( i = 0; i < ny; i++ )
for(j = 0; j < nx; j++ ) objectPoints[i*nx + j] = cvPoint3D32f(i*squareSize, j*squareSize, 0);
for( i = 1; i < n_boards; i++ ) copy( objectPoints.begin(), objectPoints.begin() + n, objectPoints.begin() + i*n );
npoints.resize(n_boards,n);
CvMat _objectPoints = cvMat(1, N, CV_32FC3, &objectPoints[0] );
CvMat _imagePoints1 = cvMat(1, N, CV_32FC2, &points[0][0] );
CvMat _imagePoints2 = cvMat(1, N, CV_32FC2, &points[1][0] );
CvMat _npoints = cvMat(1, npoints.size(), CV_32S, &npoints[0] );
cvSetIdentity(&_M1);
cvSetIdentity(&_M2);
cvZero(&_D1);
cvZero(&_D2);
printf("Running stereo calibration ...");
fflush(stdout);
cvStereoCalibrate( &_objectPoints, &_imagePoints1, &_imagePoints2, &_npoints,&_M1, &_D1, &_M2, &_D2,imageSize, &_R, &_T, &_E, &_F,
cvTermCriteria(CV_TERMCRIT_ITER+CV_TERMCRIT_EPS, 100, 1e-5),
CV_CALIB_FIX_ASPECT_RATIO+CV_CALIB_ZERO_TANGENT_DIST + CV_CALIB_SAME_FOCAL_LENGTH );
printf("done\n");
//-------------Undistort------------------------------------------
cvUndistortPoints( &_imagePoints1, &_imagePoints1,&_M1, &_D1, 0, &_M1 );
cvUndistortPoints( &_imagePoints2, &_imagePoints2,&_M2, &_D2, 0, &_M2 );
//--------Using bouguet algorithm-------------------
CvMat* mx1 = cvCreateMat( imageSize.height,imageSize.width, CV_32F );
CvMat* my1 = cvCreateMat( imageSize.height,imageSize.width, CV_32F );
CvMat* mx2 = cvCreateMat( imageSize.height,imageSize.width, CV_32F );
CvMat* my2 = cvCreateMat( imageSize.height,imageSize.width, CV_32F );
CvMat* frame1r = cvCreateMat( imageSize.height,imageSize.width, CV_8U );
CvMat* frame2r = cvCreateMat( imageSize.height,imageSize.width, CV_8U );
CvMat* disp = cvCreateMat( imageSize.height, imageSize.width, CV_16S );
CvMat* vdisp = cvCreateMat( imageSize.height,imageSize.width, CV_8U );
CvMat* Image3D = cvCreateMat(imageSize.height, imageSize.width, CV_32FC3);
CvMat* pair;
double R1[3][3], R2[3][3], P1[3][4], P2[3][4];
CvMat _R1 = cvMat(3, 3, CV_64F, R1);
CvMat _R2 = cvMat(3, 3, CV_64F, R2);
//Calib with Bouguet algrithm
CvMat _P1 = cvMat(3, 4, CV_64F, P1);
CvMat _P2 = cvMat(3, 4, CV_64F, P2);
cvStereoRectify( &_M1, &_M2, &_D1, &_D2, imageSize,&_R, &_T,&_R1, &_R2, &_P1, &_P2, &_Q,0/*CV_CALIB_ZERO_DISPARITY*/ );
//Find matrix for cvRemap()
cvInitUndistortRectifyMap(&_M1,&_D1,&_R1,&_P1,mx1,my1);
cvInitUndistortRectifyMap(&_M2,&_D2,&_R2,&_P2,mx2,my2);
pair = cvCreateMat( imageSize.height, imageSize.width*2,CV_8UC3 );
//Paramater for stereo corrrespondences
CvStereoBMState *BMState = cvCreateStereoBMState();
assert(BMState != 0);
BMState->preFilterSize=31;
BMState->preFilterCap=31;
BMState->SADWindowSize=35;
BMState->minDisparity= 0;
BMState->numberOfDisparities=48;
BMState->textureThreshold=20; //reduce noise
BMState->uniquenessRatio=15; // uniquenessRatio > (match_val–min_match)/min_match.
/* CvStereoBMState *state = cvCreateStereoBMState(CV_STEREO_BM_BASIC);
BMState->speckleRange = 50;
BMState->textureThreshold = 400;*/
//Bat camera va hien thi
//cvNamedWindow( "camera2", 1 );
//cvNamedWindow( "camera1", 1 );
cvNamedWindow( "rectified",1 );
cvNamedWindow( "disparity",1);
cvNamedWindow("depthmap",1);
capture1= cvCaptureFromCAM(0);
assert(capture1!=NULL); cvWaitKey(10);
capture2= cvCaptureFromCAM(1);
assert(capture2!=NULL);
res=cvSetCaptureProperty(capture1,CV_CAP_PROP_FRAME_WIDTH,WIDTH);
printf("%d",res);
res=cvSetCaptureProperty(capture1,CV_CAP_PROP_FRAME_HEIGHT,HEIGHT);
printf("%d",res);
res=cvSetCaptureProperty(capture2,CV_CAP_PROP_FRAME_WIDTH,WIDTH);
printf("%d",res);
res=cvSetCaptureProperty(capture2,CV_CAP_PROP_FRAME_HEIGHT,HEIGHT);
printf("%d",res); fflush(stdout);
frame1 = cvQueryFrame( capture1 );
frame2 = cvQueryFrame( capture2 );
while(1)
{
CvMat part;
cvCvtColor( frame1, gray_fr1, CV_BGR2GRAY );
cvCvtColor( frame2, gray_fr2, CV_BGR2GRAY );
cvRemap( gray_fr1, frame1r, mx1, my1 );
cvRemap( gray_fr2, frame2r, mx2, my2 );
cvFindStereoCorrespondenceBM( frame1r, frame2r, disp, BMState);
/* cvShowImage("camera1", frame1);
cvShowImage("camera2", frame2); */
// cvConvertScale( disp, disp, 16, 0 );
cvNormalize( disp, vdisp, 0, 256, CV_MINMAX );
cvShowImage( "disparity", vdisp );
cvReprojectImageTo3D(disp, Image3D, &_Q);
cvShowImage("depthmap",Image3D);
//Hien thi anh da rectify
cvGetCols( pair, &part, 0, imageSize.width );
cvCvtColor( frame1r, &part, CV_GRAY2BGR );
cvGetCols( pair, &part, imageSize.width, imageSize.width*2 );
cvCvtColor( frame2r, &part, CV_GRAY2BGR ); //CV_GRAY2BGR
for( j = 0; j < imageSize.height; j += 16 )
cvLine( pair, cvPoint(0,j), cvPoint(imageSize.width*2,j), CV_RGB(0,255,0));
cvShowImage( "rectified", pair );
frame1 = cvQueryFrame( capture1 );
frame2 = cvQueryFrame( capture2 );
if( cvWaitKey(15) == 27 ) break;
}
while( 1 ) { if((cvWaitKey(10)&0x7f) == 27 ) break; }
cvReleaseStereoBMState(&BMState);
cvReleaseMat( &mx1 );
cvReleaseMat( &my1 );
cvReleaseMat( &mx2 );
cvReleaseMat( &my2 );
cvReleaseCapture( &capture1 );
cvReleaseCapture( &capture2 );
cvReleaseMat( &frame1r );
cvReleaseMat( &frame2r );
cvReleaseMat( &disp );
cvReleaseMat(&Image3D);
}
和差异图的照片在此link中给出
一个是网络摄像头，另一个是USB摄像头
最佳答案
拥有不同的相机会使您的生活更加困难。首先，我怀疑摄像机的焦距是否相同，因此设置的参数CV_CALIB_SAME_FOCAL_LENGTH会导致您对该对进行错误的校准(将其删除为固定参数，因为几乎肯定会有所不同)。
同样，在相关的参数说明中，由于径向变形参数k1-k6都不固定，因此可以为校准算法提供很大的灵活性。首先尝试使用最少数量的参数以获得良好的拟合度。例如，修复除k1和k2以外的所有内容(即，使用CV_CALIB_FIX_K3 + CV_CALIB_FIX_K4 + CV_CALIB_FIX_K5 + CV_CALIB_FIX_K6)。如果那不起作用，请允许k3可变等。
编辑:
另一个技巧是，考虑分别校准每个摄像机，然后将CV_CALIB_FIX_INTRINSIC与两个摄像机固有矩阵一起传递以缩小搜索空间，这有望使该算法收敛到全局而不是局部校准最小值。
希望有帮助!
关于c++ - 视差映射中的立体视觉，我们在Stack Overflow上找到一个类似的问题：https://stackoverflow.com/questions/10703702/