ThomCastel's profile - activity

Did you get the streaming to work?

Yes, I got it to work with an IP camera. I guess I'll have to modify the code, even to only resize the input stream images. Thanks.

Hi, I'm using the code bellow for stabilization (OpenCV 3.0, Ubuntu).

It's working fine but I want to be able to change the type of video source (webcam, IP stream...) and resize the input frames, how should I do that?

    std::string videoFile = getMyFile();

    MotionModel model = cv::videostab::MM_AFFINE;
    bool use_gpu = true;

    cv::Ptr<VideoFileSource> video = cv::makePtr<VideoFileSource>(videoFile,true); 

    cv::Ptr<OnePassStabilizer> stabilizer = cv::makePtr<OnePassStabilizer>();

    cv::Ptr<MotionEstimatorBase> MotionEstimator = cv::makePtr<MotionEstimatorRansacL2>(model);

    cv::Ptr<ImageMotionEstimatorBase> ImageMotionEstimator;
    if (use_gpu)
    {
        ImageMotionEstimator = cv::makePtr<KeypointBasedMotionEstimatorGpu>(MotionEstimator);
    }
    else
    {
        ImageMotionEstimator = cv::makePtr<KeypointBasedMotionEstimator>(MotionEstimator);
    }

    stabilizer->setFrameSource(video);
    stabilizer->setMotionEstimator(ImageMotionEstimator);
    stabilizer->setLog(cv::makePtr<cv::videostab::NullLog>());
    stabilizer->setBorderMode(BORDER_REFLECT_101);
    stabilizer->setRadius(10);

    std::string windowTitle = "Stabilized Video";
    cv::namedWindow(windowTitle, cv::WINDOW_AUTOSIZE);

    while(true)
    {
        double procT = (double)getTickCount();
        cv::Mat frame = stabilizer->nextFrame();

        if(frame.empty())
        {
            break;
        }

        procT = ((double)getTickCount() - procT)/getTickFrequency();
        cout << 1/procT << " FPS" << endl;

        cv::imshow(windowTitle,frame);
        cv::waitKey(1);
    }

Thanks for your answer.

It seems to be fine, I successfully tested a bunch of other function like calcOpticalFlowFarneback. But the same code does work fine using the OpenCV 2.4.9 I compiled a while back...

Any other idea?

Hi, I'm having trouble with goodFeatureToTrack using OpenCV 3.0 in C++ with Qt and MSVC2010.

When I run the following code, the size of corners is 4,289,272,441 (image is ok and displays).

#include <iostream>
#include "opencv.hpp"
#include "world.hpp"
#include <vector>

using namespace cv;
using namespace std;

int main()
{
    Mat frame = imread("lena_color.tif");
    std::vector<cv::Point2f> corners;
    //Mat corners;
    cvtColor(frame,frame,COLOR_RGB2GRAY);

    goodFeaturesToTrack(frame,corners,10,0.001,10);
    cout << corners.size() << endl;

    imshow("Img",frame);
    waitKey();

    return 0;
}

I know corners is not a Mat but when I use it as such I get a size of [1x10], which makes sense.

Has anyone run into this behavior and knows how to explain and fix it?

If your images are all like the ones you showed us, meaning no scale or illumination change and nice background, then you could either use:

1. Template matching using matchTemplate and select the sample that gives the best result.
2. Binarize, get blobs and sort the coins based on shape or area.

That's what I would try first, but maybe you have some other constraints?

This could be an easy way of doing background substraction. If you have ideal and stable conditions, once this is done, everything that will appear in the image will be detected.