OpenCV3.4两种立体匹配算法效果对比

以OpenCV自带的Aloe图像对为例：

   

1.BM算法（Block Matching）

参数设置如下：

    int numberOfDisparities = ((imgSize.width / ) + ) & -;
    cv::Ptr<cv::StereoBM> bm = cv::StereoBM::create(, );
    cv::Rect roi1, roi2;
    bm->setROI1(roi1);
    bm->setROI2(roi2);
    bm->setPreFilterCap();
    bm->setBlockSize();
    bm->setMinDisparity();
    bm->setNumDisparities(numberOfDisparities);
    bm->setTextureThreshold();
    bm->setUniquenessRatio();
    bm->setSpeckleWindowSize();
    bm->setSpeckleRange();
    bm->setDisp12MaxDiff();
    bm->compute(imgL, imgR, disp);

效果如下：

BM算法得到的视差图（左），空洞填充后得到的视差图（右）

  

2.SGBM（Semi-Global Block matching）算法:

参数设置如下：

enum { STEREO_BM = , STEREO_SGBM = , STEREO_HH = , STEREO_VAR = , STEREO_3WAY =  };
    int numberOfDisparities = ((imgSize.width / ) + ) & -;
    cv::Ptr<cv::StereoSGBM> sgbm = cv::StereoSGBM::create(, , );
    sgbm->setPreFilterCap();
    int SADWindowSize = ;
    int sgbmWinSize = SADWindowSize >  ? SADWindowSize : ;
    sgbm->setBlockSize(sgbmWinSize);
    int cn = imgL.channels();
    sgbm->setP1( * cn*sgbmWinSize*sgbmWinSize);
    sgbm->setP2( * cn*sgbmWinSize*sgbmWinSize);
    sgbm->setMinDisparity();
    sgbm->setNumDisparities(numberOfDisparities);
    sgbm->setUniquenessRatio();
    sgbm->setSpeckleWindowSize();
    sgbm->setSpeckleRange();
    sgbm->setDisp12MaxDiff();

    int alg = STEREO_SGBM;
    if (alg == STEREO_HH)
        sgbm->setMode(cv::StereoSGBM::MODE_HH);
    else if (alg == STEREO_SGBM)
        sgbm->setMode(cv::StereoSGBM::MODE_SGBM);
    else if (alg == STEREO_3WAY)
        sgbm->setMode(cv::StereoSGBM::MODE_SGBM_3WAY);
    sgbm->compute(imgL, imgR, disp);

效果如图：

SGBM算法得到的视差图（左），空洞填充后得到的视差图（右）

  

可见SGBM算法得到的视差图相比于BM算法来说，减少了很多不准确的匹配点，尤其是在深度不连续区域，速度上SGBM要慢于BM算法。OpenCV3.0以后没有实现GC算法，可能是出于速度考虑，以后找时间补上对比图，以及各个算法的详细原理分析。

后面我填充空洞的效果不是很好，如果有更好的方法，望不吝赐教。

---

preFilterCap（）匹配图像预处理

• 两种立体匹配算法都要先对输入图像做预处理，OpenCV源码中中调用函数 static void prefilterXSobel(const cv::Mat& src, cv::Mat& dst, int preFilterCap)，参数设置中preFilterCap在此函数中用到。函数步骤如下，作用主要有两点：对于无纹理区域，能够排除噪声干扰；对于边界区域，能够提高边界的区分性，利于后续的匹配代价计算：

1. 先利用水平Sobel算子求输入图像x方向的微分值Value；
2. 如果Value<-preFilterCap, 则Value=0;
   如果Value>preFilterCap,则Value=2*preFilterCap;
   如果Value>=-preFilterCap &&Value<=preFilterCap,则Value=Value+preFilterCap;
3. 输出处理后的图像作为下一步计算匹配代价的输入图像。

static void prefilterXSobel(const cv::Mat& src, cv::Mat& dst, int ftzero)
{
    int x, y;
    const int OFS =  * , TABSZ = OFS *  + ;
    uchar tab[TABSZ];
    cv::Size size = src.size();

    for (x = ; x < TABSZ; x++)
        tab[x] = (uchar)(x - OFS < -ftzero ?  : x - OFS > ftzero ? ftzero *  : x - OFS + ftzero);
    uchar val0 = tab[ + OFS];

    for (y = ; y < size.height - ; y += )
    {
        const uchar* srow1 = src.ptr<uchar>(y);
        const uchar* srow0 = y >  ? srow1 - src.step : size.height >  ? srow1 + src.step : srow1;
        const uchar* srow2 = y < size.height -  ? srow1 + src.step : size.height >  ? srow1 - src.step : srow1;
        const uchar* srow3 = y < size.height -  ? srow1 + src.step *  : srow1;
        uchar* dptr0 = dst.ptr<uchar>(y);
        uchar* dptr1 = dptr0 + dst.step;

        dptr0[] = dptr0[size.width - ] = dptr1[] = dptr1[size.width - ] = val0;
        x = ;
        for (; x < size.width - ; x++)
        {
            int d0 = srow0[x + ] - srow0[x - ], d1 = srow1[x + ] - srow1[x - ],
                d2 = srow2[x + ] - srow2[x - ], d3 = srow3[x + ] - srow3[x - ];
            int v0 = tab[d0 + d1 *  + d2 + OFS];
            int v1 = tab[d1 + d2 *  + d3 + OFS];
            dptr0[x] = (uchar)v0;
            dptr1[x] = (uchar)v1;
        }
    }

    for (; y < size.height; y++)
    {
        uchar* dptr = dst.ptr<uchar>(y);
        x = ;
        for (; x < size.width; x++)
            dptr[x] = val0;
    }
}

自己实现的函数如下：

void mySobelX(cv::Mat srcImg, cv::Mat dstImg, int preFilterCap)
{
    assert(srcImg.channels() == );
    int radius = ;
    int width = srcImg.cols;
    int height = srcImg.rows;
    uchar *pSrcData = srcImg.data;
    uchar *pDstData = dstImg.data;
    for (int i = ; i < height; i++)
    {
        for (int j = ; j < width; j++)
        {
            int idx = i*width + j;
            if (i >= radius && i < height - radius && j >= radius && j < width - radius)
            {
                int diff0 = pSrcData[(i - )*width + j + ] - pSrcData[(i - )*width + j - ];
                int diff1 = pSrcData[i*width + j + ] - pSrcData[i*width + j - ];
                int diff2 = pSrcData[(i + )*width + j + ] - pSrcData[(i + )*width + j - ];

                int value = diff0 +  * diff1 + diff2;
                if (value < -preFilterCap)
                {
                    pDstData[idx] = ;
                }
                else if (value >= -preFilterCap && value <= preFilterCap)
                {
                    pDstData[idx] = uchar(value + preFilterCap);
                }
                else
                {
                    pDstData[idx] = uchar( * preFilterCap);
                }

            }
            else
            {
                pDstData[idx] = ;
            }
        }
    }
}

函数输入，输出结果如图：

  

---

 filterSpeckles（）视差图后处理

• 两种立体匹配算法在算出初始视差图后会进行视差图后处理，包括中值滤波，连通域检测等。其中中值滤波能够有效去除视差图中孤立的噪点，而连通域检测能够检测出视差图中因噪声引起小团块(blob)。在BM和SGBM中都有speckleWindowSize和speckleRange这两个参数，speckleWindowSize是指设置检测出的连通域中像素点个数，也就是连通域的大小。speckleRange是指设置判断两个点是否属于同一个连通域的阈值条件。大概流程如下：

1. 判断当前像素点四邻域的邻域点与当前像素点的差值diff，如果diff<speckRange，则表示该邻域点与当前像素点是一个连通域，设置一个标记。然后再以该邻域点为中心判断其四邻域点，步骤同上。直至某一像素点四邻域的点均不满足条件，则停止。
2. 步骤1完成后，判断被标记的像素点个数count，如果像素点个数count<=speckleWindowSize，则说明该连通域是一个小团块(blob)，则将当前像素点值设置为newValue(表示错误的视差值，newValue一般设置为负数或者0值)。否则，表示该连通域是个大团块，不做处理。同时建立标记值与是否为小团块的关系表rtype[label]，rtype[label]为0，表示label值对应的像素点属于小团块，为1则不属于小团块。
3. 处理下一个像素点时，先判断其是否已经被标记：
   如果已经被标记，则根据关系表rtype[label]判断是否为小团块（blob），如果是，则直接将该像素值设置为newValue；如果不是，则不做处理。继续处理下一个像素。
   如果没有被标记，则按照步骤1处理。
4. 所有像素点处理后，满足条件的区域会被设置为newValue值，后续可以用空洞填充等方法重新估计其视差值。

OpenCV中有对应的API函数，void filterSpeckles(InputOutputArray img, double newVal, int maxSpeckleSize, double maxDiff, InputOutputArray buf=noArray() )

函数源码如下，使用时根据视差图或者深度图数据类型设置模板中的数据类型：

typedef cv::Point_<short> Point2s;
template <typename T> void filterSpecklesImpl(cv::Mat& img, int newVal, int maxSpeckleSize, int maxDiff, cv::Mat& _buf)
{
    using namespace cv;

    int width = img.cols, height = img.rows, npixels = width*height;
    size_t bufSize = npixels*(int)(sizeof(Point2s) + sizeof(int) + sizeof(uchar));
    if (!_buf.isContinuous() || _buf.empty() || _buf.cols*_buf.rows*_buf.elemSize() < bufSize)
        _buf.create(, (int)bufSize, CV_8U);

    uchar* buf = _buf.ptr();
    int i, j, dstep = (int)(img.step / sizeof(T));
    int* labels = (int*)buf;
    buf += npixels * sizeof(labels[]);
    Point2s* wbuf = (Point2s*)buf;
    buf += npixels * sizeof(wbuf[]);
    uchar* rtype = (uchar*)buf;
    int curlabel = ;

    // clear out label assignments
    memset(labels, , npixels * sizeof(labels[]));

    for (i = ; i < height; i++)
    {
        T* ds = img.ptr<T>(i);
        int* ls = labels + width*i;

        for (j = ; j < width; j++)
        {
            if (ds[j] != newVal)   // not a bad disparity
            {
                if (ls[j])     // has a label, check for bad label
                {
                    if (rtype[ls[j]]) // small region, zero out disparity
                        ds[j] = (T)newVal;
                }
                // no label, assign and propagate
                else
                {
                    Point2s* ws = wbuf; // initialize wavefront
                    Point2s p((short)j, (short)i);  // current pixel
                    curlabel++; // next label
                    int count = ;  // current region size
                    ls[j] = curlabel;

                    // wavefront propagation
                    while (ws >= wbuf) // wavefront not empty
                    {
                        count++;
                        // put neighbors onto wavefront
                        T* dpp = &img.at<T>(p.y, p.x); //current pixel value
                        T dp = *dpp;
                        int* lpp = labels + width*p.y + p.x; //current label value

                        //bot
                        if (p.y < height -  && !lpp[+width] && dpp[+dstep] != newVal && std::abs(dp - dpp[+dstep]) <= maxDiff)
                        {
                            lpp[+width] = curlabel;
                            *ws++ = Point2s(p.x, p.y + );
                        }
                        //top
                        if (p.y >  && !lpp[-width] && dpp[-dstep] != newVal && std::abs(dp - dpp[-dstep]) <= maxDiff)
                        {
                            lpp[-width] = curlabel;
                            *ws++ = Point2s(p.x, p.y - );
                        }
                        //right
                        if (p.x < width -  && !lpp[+] && dpp[+] != newVal && std::abs(dp - dpp[+]) <= maxDiff)
                        {
                            lpp[+] = curlabel;
                            *ws++ = Point2s(p.x + , p.y);
                        }
                        //left
                        if (p.x >  && !lpp[-] && dpp[-] != newVal && std::abs(dp - dpp[-]) <= maxDiff)
                        {
                            lpp[-] = curlabel;
                            *ws++ = Point2s(p.x - , p.y);
                        }

                        // pop most recent and propagate
                        // NB: could try least recent, maybe better convergence
                        p = *--ws;
                    }

                    // assign label type
                    if (count <= maxSpeckleSize)   // speckle region
                    {
                        rtype[ls[j]] = ;   // small region label
                        ds[j] = (T)newVal;
                    }
                    else
                        rtype[ls[j]] = ;   // large region label
                }
            }
        }
    }
}

或者下面博主自己整理一遍的代码：

typedef cv::Point_<short> Point2s;
template <typename T> void myFilterSpeckles(cv::Mat &img, int newVal, int maxSpeckleSize, int maxDiff)
{
    int width = img.cols;
    int height = img.rows;
    int imgSize = width*height;
    int *pLabelBuf = (int*)malloc(sizeof(int)*imgSize);//标记值buffer
    Point2s *pPointBuf = (Point2s*)malloc(sizeof(short)*imgSize);//点坐标buffer
    uchar *pTypeBuf = (uchar*)malloc(sizeof(uchar)*imgSize);//blob判断标记buffer
    //初始化Labelbuffer
    int currentLabel = ;
    memset(pLabelBuf, , sizeof(int)*imgSize);

    for (int i = ; i < height; i++)
    {
        T *pData = img.ptr<T>(i);
        int *pLabel = pLabelBuf + width*i;
        for (int j = ; j < width; j++)
        {
            if (pData[j] != newVal)
            {
                if (pLabel[j])
                {
                    if (pTypeBuf[pLabel[j]])
                    {
                        pData[j] = (T)newVal;
                    }
                }
                else
                {
                    Point2s *pWave = pPointBuf;
                    Point2s curPoint((T)j, (T)i);
                    currentLabel++;
                    int count = ;
                    pLabel[j] = currentLabel;
                    while (pWave >= pPointBuf)
                    {
                        count++;
                        T *pCurPos = &img.at<T>(curPoint.y, curPoint.x);
                        T curValue = *pCurPos;
                        int *pCurLabel = pLabelBuf + width*curPoint.y + curPoint.x;
                        //bot
                        if (curPoint.y < height -  && !pCurLabel[+width] && pCurPos[+width] != newVal  && abs(curValue - pCurPos[+width]) <= maxDiff)
                        {
                            pCurLabel[+width] = currentLabel;
                            *pWave++ = Point2s(curPoint.x, curPoint.y + );
                        }
                        //top
                        if (curPoint.y >  && !pCurLabel[-width] && pCurPos[-width] != newVal && abs(curValue - pCurPos[-width]) <= maxDiff)
                        {
                            pCurLabel[-width] = currentLabel;
                            *pWave++ = Point2s(curPoint.x, curPoint.y - );
                        }
                        //right
                        if (curPoint.x < width- && !pCurLabel[+] && pCurPos[+] != newVal  && abs(curValue - pCurPos[+]) <= maxDiff)
                        {
                            pCurLabel[+] = currentLabel;
                            *pWave++ = Point2s(curPoint.x + , curPoint.y);
                        }
                        //left
                        if (curPoint.x >  && !pCurLabel[-] && pCurPos[-] != newVal && abs(curValue - pCurPos[-]) <= maxDiff)
                        {
                            pCurLabel[-] = currentLabel;
                            *pWave++ = Point2s(curPoint.x - , curPoint.y);
                        }

                        --pWave;
                        curPoint = *pWave;
                    }

                    if (count <= maxSpeckleSize)
                    {
                        pTypeBuf[pLabel[j]] = ;
                        pData[j] = (T)newVal;
                    }
                    else
                    {
                        pTypeBuf[pLabel[j]] = ;
                    }
                }
            }
        }
    }

    free(pLabelBuf);
    free(pPointBuf);
    free(pTypeBuf);
}

如下视差图中左上角部分有7个小团块，设置speckleWindowSize和speckleRange分别为50和32，连通域检测后结果为如下图右，小团块能够全部检测出来，方便后续用周围视差填充。当然还有一个缺点就是，图像中其他地方尤其是边界区域也会被检测为小团块，后续填充可能会对边界造成平滑。