OpenCv中实现了三种立体匹配算法：

OpenCv中实现了三种立体匹配算法：


BM算法


SGBM算法 Stereo Processing by Semiglobal Matching and Mutual Information


GC算法 算法文献：Realistic CG Stereo Image Dataset with Ground Truth Disparity Maps


参考：http://blog.csdn.net/wqvbjhc/article/details/6260844


BM算法：速度很快，效果一般


复制代码
void BM()
{
　　IplImage * img1 = cvLoadImage("left.png",0);
    IplImage * img2 = cvLoadImage("right.png",0);
    CvStereoBMState* BMState=cvCreateStereoBMState();
    assert(BMState);
    BMState->preFilterSize=9;
    BMState->preFilterCap=31;
    BMState->SADWindowSize=15;
    BMState->minDisparity=0;
    BMState->numberOfDisparities=64;
    BMState->textureThreshold=10;
    BMState->uniquenessRatio=15;
    BMState->speckleWindowSize=100;
    BMState->speckleRange=32;
    BMState->disp12MaxDiff=1;


    CvMat* disp=cvCreateMat(img1->height,img1->width,CV_16S);
    CvMat* vdisp=cvCreateMat(img1->height,img1->width,CV_8U);
    int64 t=getTickCount();
    cvFindStereoCorrespondenceBM(img1,img2,disp,BMState);
    t=getTickCount()-t;
    cout<<"Time elapsed:"<<t*1000/getTickFrequency()<<endl;
    cvSave("disp.xml",disp);
    cvNormalize(disp,vdisp,0,255,CV_MINMAX);
    cvNamedWindow("BM_disparity",0);
    cvShowImage("BM_disparity",vdisp);
    cvWaitKey(0);
    //cvSaveImage("cones\\BM_disparity.png",vdisp);
    cvReleaseMat(&disp);
    cvReleaseMat(&vdisp);
    cvDestroyWindow("BM_disparity");
}
复制代码
   


　　left.png 　　　　　　　　　　　　right.png                                    disparity.jpg


SGBM算法，作为一种全局匹配算法，立体匹配的效果明显好于局部匹配算法，但是同时复杂度上也要远远大于局部匹配算法。算法主要是参考Stereo Processing by Semiglobal Matching and Mutual Information。


opencv中实现的SGBM算法计算匹配代价没有按照原始论文的互信息作为代价，而是按照块匹配的代价。


参考：http://www.opencv.org.cn/forum.php?mod=viewthread&tid=23854


复制代码
#include <highgui.h>
#include <cv.h>
#include <cxcore.h>
#include <iostream>
using namespace std;
using namespace cv;
int main()
{


    IplImage * img1 = cvLoadImage("left.png",0);
    IplImage * img2 = cvLoadImage("right.png",0);
    cv::StereoSGBM sgbm;
    int SADWindowSize = 9;
    sgbm.preFilterCap = 63;
    sgbm.SADWindowSize = SADWindowSize > 0 ? SADWindowSize : 3;
    int cn = img1->nChannels;
    int numberOfDisparities=64;
    sgbm.P1 = 8*cn*sgbm.SADWindowSize*sgbm.SADWindowSize;
    sgbm.P2 = 32*cn*sgbm.SADWindowSize*sgbm.SADWindowSize;
    sgbm.minDisparity = 0;
    sgbm.numberOfDisparities = numberOfDisparities;
    sgbm.uniquenessRatio = 10;
    sgbm.speckleWindowSize = 100;
    sgbm.speckleRange = 32;
    sgbm.disp12MaxDiff = 1;
    Mat disp, disp8;
    int64 t = getTickCount();
    sgbm((Mat)img1, (Mat)img2, disp);
    t = getTickCount() - t;
    cout<<"Time elapsed:"<<t*1000/getTickFrequency()<<endl;
    disp.convertTo(disp8, CV_8U, 255/(numberOfDisparities*16.));


    namedWindow("left", 1);
    cvShowImage("left", img1);
    namedWindow("right", 1);
    cvShowImage("right", img2);
    namedWindow("disparity", 1);
    imshow("disparity", disp8);
    waitKey();
    imwrite("sgbm_disparity.png", disp8);   
    cvDestroyAllWindows();
    return 0;
}
复制代码
  


　　　　　　　　left.png 　　　　　　right.png                                    disparity.jpg　　　　　　　　　


GC算法 效果最好，速度最慢


复制代码
void GC()
{
    IplImage * img1 = cvLoadImage("left.png",0);
    IplImage * img2 = cvLoadImage("right.png",0);
    CvStereoGCState* GCState=cvCreateStereoGCState(64,3);
    assert(GCState);
    cout<<"start matching using GC"<<endl;
    CvMat* gcdispleft=cvCreateMat(img1->height,img1->width,CV_16S);
    CvMat* gcdispright=cvCreateMat(img2->height,img2->width,CV_16S);
    CvMat* gcvdisp=cvCreateMat(img1->height,img1->width,CV_8U);
    int64 t=getTickCount();
    cvFindStereoCorrespondenceGC(img1,img2,gcdispleft,gcdispright,GCState);
    t=getTickCount()-t;
    cout<<"Time elapsed:"<<t*1000/getTickFrequency()<<endl;
    //cvNormalize(gcdispleft,gcvdisp,0,255,CV_MINMAX);
    //cvSaveImage("GC_left_disparity.png",gcvdisp);
    cvNormalize(gcdispright,gcvdisp,0,255,CV_MINMAX);
    cvSaveImage("GC_right_disparity.png",gcvdisp);




    cvNamedWindow("GC_disparity",0);
    cvShowImage("GC_disparity",gcvdisp);
    cvWaitKey(0);
    cvReleaseMat(&gcdispleft);
    cvReleaseMat(&gcdispright);
    cvReleaseMat(&gcvdisp);
}
复制代码
 


             


　　　　　left.png 　　　　　　right.png                disparity.jpg　       


如何设置BM、SGBM和GC算法的状态参数？


参看：http://blog.csdn.net/chenyusiyuan/article/details/5967291