Ask Your Question

[OPENCV GPU] How can I convert GpuMat and Vector<Point2f>

asked 2015-02-02 09:39:21 -0500


I am trying to implement a tracking solution in OpenCV 2.4.10 with GPU support.

So first I am extracting GFTT points :

gpu::GoodFeaturesToTrackDetector_GPU detector(MAX_COUNT, 0.01, 10);
        detector(d_frame, d_points);

Then I want to perform some computation on d_points before putting them in the PyrLKOpticalFlow tracker. So there is two solutions: Download in a vector<points2f> or directly work with d_points.

1) Download in a vector<point2f> I can easily download data in a vector of Points2f. For that I use the method:

static void download(const GpuMat& d_mat, vector<Point2f>& vec) {
Mat mat(1, d_mat.cols, CV_32FC2, (void*)&vec[0]);; }

but I am not able to re-upload the modified data in d_points with the same structure (a GpuMat of 1 row, N columns, and Points2f data type). At best, I can obtain a structure like : GpuMat, 2 rows, N columns, and float data type; which is not readable by the PyrLKOpticalFlow tracker. I tried several things :

        //Mat mat(points, false);
        //Mat mat = Mat(points, CV_32FC2);
        //Mat mat(1, d_points.cols, CV_32FC2, (void*)&points);
        //GpuMat d_mat = GpuMat(mat); d_mat.copyTo(d_points);
        //Mat mat = Mat(points, CV_32FC2).reshape(1,points.size());
        //d_points.create(1, points.size(),CV_32FC2);
        //d_points.upload(Mat(points, false));

2) Directly modify the GpuMat d_points I found several way to access d_points element, but every time I try to cout them I have a segmentation fault.

I know that d_points.cols = N and d_points.rows = 1. I tried this way :

Point2f* row_y = (Point2f*)((char*) + n * step);
            Point2f val = row_y[0];

But if I cout I get a segmentation fault.

And I also tried :

cv::Point2f* pixel_ptr = d_points.ptr<cv::Point2f >(n);

But when I want to access pixel_ptr, I also get a segmentation fault.

Is there a way to modify d_points (GpuMat with a special structure) or to access the data, modify it and re-upload it ?

Thank you for help !

edit retag flag offensive close merge delete


Any luck with this issue? I am also trying to using PyrLKOpticalFlow and I can not pass the gpuMat that i get from FAST_GPU to sparse call for PyrLKOpticalFlow.

kjoshi gravatar imagekjoshi ( 2018-03-20 07:22:22 -0500 )edit

I'm trying to do the exactly same thing as you, @kjoshi. Have you solved it?

bendikiv gravatar imagebendikiv ( 2018-04-14 08:33:41 -0500 )edit

1 answer

Sort by ยป oldest newest most voted

answered 2019-06-20 09:48:43 -0500

riger gravatar image

My solution(opencv 3.0+):

inline void upload(const vector<cv::Point2f>& vec, cv::cuda::GpuMat& d_mat)
    cv::Mat mat(1, vec.size(), CV_32FC2, (void*)&vec[0]);

Hope this helps.

edit flag offensive delete link more
Login/Signup to Answer

Question Tools



Asked: 2015-02-02 09:37:50 -0500

Seen: 1,215 times

Last updated: Jun 20 '19