如何使用 openCV 库在视频中提取清晰的轮廓

Posted 2023-02-17

【中文标题】如何使用 openCV 库在视频中提取清晰的轮廓

【英文标题】：How can I extract clear contours in video with openCV library

【发布时间】：2011-12-27 21:02:08

【问题描述】：

大家好，我在从视频中提取清晰轮廓时遇到了一些问题。目前，轮廓看起来像一团糟。我遵循的步骤是：

将视频作为输入 将其转换为灰度 将其阈值化为二进制 使用 cvFindContours() 查找轮廓 遍历轮廓并使用 cvDrawContours 绘制它们

代码如下：

IplImage *inFrame;
IplImage *outFrame = cvCreateImage(cvSize(width,height),IPL_DEPTH_8U,3);
CvScalar inPixel, outPixel;

//J.M ================================================//
CvMemStorage *mem = cvCreateMemStorage(0), *polymem = cvCreateMemStorage(0);
CvSeq *contours = NULL, *ptr = NULL, *poly = NULL, *convex_hull = NULL;
IplImage *cc_color = cvCreateImage(cvGetSize(outFrame), IPL_DEPTH_8U, 3);
CvScalar externalColor = CV_RGB( 255, 252, 0 );
CvScalar holeColor = CV_RGB( 255, 0, 0 );
CvFont font;
int contourNum = 0;
//=========================================================================//

frame_no = 0;
processingVideo = true;
this->GetDlgItem(IDC_STOP)->EnableWindow(true);

CString title = "Video Processing";
cvNamedWindow(title, CV_WINDOW_AUTOSIZE);

CvVideoWriter *writer = cvCreateVideoWriter(pathOutVideo,CV_FOURCC('I', 'Y', 'U', 'V'),video->getFPS(),cvSize(width,height),1);

// variables for counting time
clock_t begin, current;
long int secs, rem_secs, t1, t2;
begin = clock();

while(inFrame=cvQueryFrame(video->getCapture()))

    if(!processingVideo)
        break;

    // close opencv window?
    if(cvGetWindowHandle(title)==NULL)
        break;

    //==============================================================================================//

    /*Create gray-scale image*/
    IplImage *im_gray = cvCreateImage(cvGetSize(outFrame), IPL_DEPTH_8U, 1);
    cvCvtColor(inFrame, im_gray, CV_RGB2GRAY);
    outFrame = im_gray;

    /*Convert gray-scale image to binary*/
    IplImage* img_bw = cvCreateImage(cvGetSize(im_gray), IPL_DEPTH_8U, 1); 
    cvThreshold(im_gray, img_bw, 128, 255, CV_THRESH_BINARY | CV_THRESH_OTSU);
    outFrame = img_bw;

    contourNum = cvFindContours(outFrame, mem, &contours, sizeof(CvContour), CV_RETR_CCOMP, CV_CHAIN_APPROX_SIMPLE, cvPoint(0,0));

    /* I have tried using convex_hull but with no success
    convex_hull = cvConvexHull2( contours, polymem, CV_CLOCKWISE, 2 ); 
    poly = cvApproxPoly(convex_hull, sizeof(CvContour), polymem, CV_POLY_APPROX_DP, cvContourPerimeter(contours)*0.02, 0); 
    float size = fabs(cvContourArea( poly, CV_WHOLE_SEQ )); */

    for (ptr = contours; ptr != NULL; ptr = ptr->h_next) 
            cvDrawContours(cc_color, ptr, externalColor, holeColor, -1, CV_FILLED, 8, cvPoint(0,0));
    

    outFrame = cc_color;
    //==============================================================================================//

我还提供了各种步骤结果的图像实例。

源视频输入（http://imageshack.us/photo/my-images/823/sourcevideo.jpg/） 灰度视频 (http://imageshack.us/photo/my-images/23/grayscalei.jpg/) 二进制视频 (http://imageshack.us/photo/my-images/684/binaryi.jpg/) 最终轮廓输出 (http://imageshack.us/photo/my-images/35/finaloutput.jpg/)

如果您需要更多信息，请直说 :) 我想知道你们中的任何人是否可以就如何进行或我做错了什么给我一些建议。谢谢！！！

【参考方案1】：

您的Step: 3 Threshold it to binary 需要改进。我认为您希望围绕具有语义意义的对象绘制轮廓。在这种情况下，由于相同的灰度值是许多对象的一部分 - 结果图像不是对象边界的准确测量。

您可能需要尝试分割方法 - 例如分水岭算法或区域增长算法。 OpenCV 本身有很多实现。