Opencv 在 HoughLineP() 的输出上运行 Kmeans：向量到 Mat 的转换？

Question

我一直在尝试对 cv::HoughLinesP() 的输出运行 kmeans。我的问题来自将 HoughLinesP()、opencv_lines 的输出（它是Vector<Vec4i> 的向量）转换为矩阵。这听起来非常简单，我敢打赌，但是，我是 opencv 和数据结构的新手，并且尝试了很多东西。这是我要实现的代码。

HoughLinesP(image_after_canny, opencv_lines, 2, 0.05*CV_PI/180, lower_hough_prob_min_no_of_intersections_trackbar + hough_prob_min_no_of_intersections_trackbar, lower_hough_prob_min_no_of_points_trackbar + hough_prob_min_no_of_points_trackbar, lower_hough_prob_max_gap_bw_points_trackbar + hough_prob_max_gap_bw_points_trackbar);
//opencv_lines is an Nx4 opencv[0] = [x1, y1, x2, y2]

//Spatial Clustering on line endpoints
//For each frame the endpoints should not change much relative to location


//sanity checking
// Mat samples = Mat(opencv_lines);
// samples.convertTo(samples, CV_32F);
// cout << samples.dims << endl;
// cout << samples.depth() << endl;
// cout << samples.type() << endl;

//Convert the opencv_lines into a matrix for kmeas
//There must be an better way
Mat samples  = Mat(opencv_lines.size(), 4, CV_32FC1); //create a Nx4 matrix oftype 32F 
for(int r = 0; r<opencv_lines.size(); r++){
      Vec4i l_cur = opencv_lines[r];
      for(int c = 0; c<4; c ++){
        samples.at<float>(r,c) = opencv_lines[r][c];
      }
}



cout << "Running Kmeans" << endl;
cv::kmeans(opencv_lines,K,labels, TermCriteria( CV_TERMCRIT_ITER|CV_TERMCRIT_EPS, max_iteration, epsilone), attempts, cv::KMEANS_PP_CENTERS, centers);

我收到的错误是：

...
检测线 OpenCV 错误：kmeans 中的断言失败 (data0.dims 0)，文件 /Development/OpenCV/modules/core/src/kmeans.cpp，第 230 行 libc++abi.dylib：以 cv::Exception 类型的未捕获异常终止：/Development/OpenCV/modules/core/src/kmeans.cpp:230：错误：(-215) data0.dims 0 在函数 kmeans 中

Abort trap: 6

我尝试了多种方法来使用memecopy(samples.data, opencv_lines.data()*sizeof(float)); 和其他各种循环结构。

我什至切换到 python 实现并遇到了同样的问题！所以我正式卡住了。

PS 我知道 hough 的输出工作正常并且有线条。 更新： 我添加了一些打印语句，并且一直在使用 reshape() 函数，认为我不满足 data0.dims

这是新的编辑：

 Mat samples  = Mat(opencv_lines.size(), 4, CV_32F); //create a Nx4 matrix oftype 32F 
    for(int r = 0; r<opencv_lines.size(); r++){
          Vec4i l_cur = opencv_lines[r];
          for(int c = 0; c<4; c ++){
            samples.at<float>(r,c) = opencv_lines[r][c];
            cout << opencv_lines[r][c] << " " ;
          }
          cout << ""<< endl;
    }
    samples=samples.reshape(1,samples.rows * 2);

    samples.convertTo(samples, CV_32F);
    cout << "M = "<< endl << " "  << samples << endl << endl;

    cout << "Running Kmeans" << endl;
    cv::kmeans(opencv_lines,K,labels, TermCriteria( CV_TERMCRIT_ITER|CV_TERMCRIT_EPS, max_iteration, epsilon), attempts, cv::KMEANS_PP_CENTERS, centers);

但同样的错误：

   Fetching Video
Initializing Parameter
Begin Playback
Frame
Generating Skeleton Image
Detecting Lines
335 471 335 112 
337 443 338 105 
339 443 339 103 
355 68 382 478 
291 479 313 162 
356 66 383 478 
436 2 533 223 
293 479 316 152 
440 3 523 195 
416 0 500 427 
359 63 383 474 
416 4 499 425 
437 0 534 223 
343 479 376 153 
427 0 525 305 
427 1 579 479 
344 478 378 153 
M = 
 [335, 471;
 335, 112;
 337, 443;
 338, 105;
 339, 443;
 339, 103;
 355, 68;
 382, 478;
 291, 479;
 313, 162;
 356, 66;
 383, 478;
 436, 2;
 533, 223;
 293, 479;
 316, 152;
 440, 3;
 523, 195;
 416, 0;
 500, 427;
 359, 63;
 383, 474;
 416, 4;
 499, 425;
 437, 0;
 534, 223;
 343, 479;
 376, 153;
 427, 0;
 525, 305;
 427, 1;
 579, 479;
 344, 478;
 378, 153]

Running Kmeans
OpenCV Error: Assertion failed (data0.dims <= 2 && type == CV_32F && K > 0) in kmeans, file /Development/OpenCV/modules/core/src/kmeans.cpp, line 230
libc++abi.dylib: terminating with uncaught exception of type cv::Exception: Development/OpenCV/modules/core/src/kmeans.cpp:230: error: (-215) data0.dims <= 2 && type == CV_32F && K > 0 in function kmeans

Abort trap: 6

有人可以帮助澄清文档中描述的输入结构吗？

用于聚类的数据。需要具有浮点坐标的 N 维点数组。这个数组的例子可以是：

Mat points(count, 2, CV_32F);
Mat points(count, 1, CV_32FC2);
Mat points(1, count, CV_32FC2);
std::vector<cv::Point2f> points(sampleCount);

Answer 1

我最终更改为 python 实现并使用 sci-kit 在 hough 输出上运行 dbscan。通过这种方式，我能够通过 filterenter image description here 将不需要的线段聚集到不同的部分中。