Revision history [back]

I have a simple program that finds convex hulls and draws them onto my frame.

For my application, I am interested in one convex hull in particular shown below inside the labels S and V.

One way I thought of narrowing it down is looking for convex hulls which consist of parallel lines by performing a theta test for parallelism.

The structure returned by convexHull is of vector<vector<Point>>. Below is a sample image showing the points of one convex inside the container

My questions are:

1. Are the points returned in the order of (634, 599)->(387, 599)->....->(634, 597)->(634, 599)? If no, what is the order then?
2. Along with parallelism, length is another factor I had in mind to filter out useless convexes. Any better suggestions to detect this structure?

Any recommendations/responses are highly appreciated it!

I have a simple program that finds convex hulls and draws them onto my frame.

For my application, I am interested in one convex hull in particular shown below inside the labels S and V.

One way I thought of narrowing it down is looking for convex hulls which consist of parallel lines by performing a theta test for parallelism.

The structure returned by convexHull is of vector<vector<Point>>. Below is a sample image showing the points of one convex inside the container

My questions are:

1. Are the points returned in the order of Is (634, 599)->(387, 599)->....->(634, 597)->(634, 599)? the relation between the points? If no, what is the order then?
2. Along with parallelism, length is another factor I had in mind to filter out useless convexes. Any better suggestions to detect this structure?

Any recommendations/responses are highly appreciated it!

I have a simple program that finds convex hulls and draws them onto my frame.

For my application, I am interested in one convex hull in particular shown below inside the labels S and V.

One way I thought of narrowing it down is looking for convex hulls which consist of parallel lines by performing a theta test for parallelism.

The structure returned by convexHull is of vector<vector<Point>>. Below is a sample image showing the points of one convex inside the container

My questions are:

1. Is (634, 599)->(387, 599)->....->(634, 597)->(634, 599) the relation between the points? If no, what is then?
2. Along with parallelism, length is another factor I had in mind to filter out useless convexes. Any better suggestions to detect this structure?

Any recommendations/responses are highly appreciated it!

int main(int argc, const char * argv[])
{
Mat frame, grey, threshold, cannyEdge, gaussianBlur;
vector<vector<Point>> contours;

VideoCapture camera(0);

if(!camera.isOpened())
{
    cerr << "Error grabbing the camera\n";
    return -1;
}

while(true)
{
    camera.read(frame);

    cvtColor(frame, grey, COLOR_BGR2GRAY);

    for(int mainIndex = 0; mainIndex <= 12; ++mainIndex)
    {
        int autoThreshold = 200 - 15 * mainIndex;

        // TODO: tweak these values around
        Canny(grey, cannyEdge, 25, 2*autoThreshold, 3, true);
        GaussianBlur(cannyEdge, gaussianBlur, Size(3,3), 0);
        findContours(gaussianBlur, contours, hierarchy, CV_RETR_LIST, CHAIN_APPROX_SIMPLE);

        Mat contoursFrame = Mat::zeros(frame.size(), CV_8UC3);
        Mat convexHullFrame = Mat::zeros(frame.size(), CV_8UC3);

        convexes.resize(contours.size());

        for(size_t index = 0; index < contours.size(); ++index)
        {
            convexHull(contours[index], convexes[index]);
        }

        cout << "Found " << thresholdCentres.size() << " circles at threshold " << autoThreshold << endl;

        for(size_t index = 0; index < contours.size(); ++index)
        {
            Scalar colour = Scalar(rng.uniform(0, 255), rng.uniform(0, 255), rng.uniform(0, 255));
            drawContours(contoursFrame, contours, int(index), colour, 2, LINE_8);
            drawContours(convexHullFrame, convexes, int(index), colour, 2, LINE_8);
        }

        imshow("Convex", convexHullFrame);
        imshow("Contours", contoursFrame);
    }

    char tappedKey = waitKey(1);

    if(tappedKey == 27)
    {
        break;
    }
}

destroyAllWindows();

camera.release();

return 0;
}

I have a simple program that finds convex hulls and draws them onto my frame.

For my application, I am interested in one convex hull in particular shown below inside the labels S and V.

One way I thought of narrowing it down is looking for convex hulls which consist of parallel lines by performing a theta test for parallelism.

The structure returned by convexHull is of vector<vector<Point>>. Below is a sample image showing the points of one convex inside the container

My questions are:

1. Is (634, 599)->(387, 599)->....->(634, 597)->(634, 599) the relation between the points? If no, what is then?
2. Along with parallelism, length is another factor I had in mind to filter out useless convexes. Any better suggestions to detect this structure?

Any recommendations/responses are highly appreciated it!

Edit 1

Sample Image

Minimal Code

int main(int argc, const char * argv[])
{
 Mat frame, grey, threshold, cannyEdge, gaussianBlur;
 vector<vector<Point>> contours;

 VideoCapture camera(0);

 if(!camera.isOpened())
 {
     cerr << "Error grabbing the camera\n";
     return -1;
 }

 while(true)
 {
     camera.read(frame);

     cvtColor(frame, grey, COLOR_BGR2GRAY);

     for(int mainIndex = 0; mainIndex <= 12; ++mainIndex)
     {
         int autoThreshold = 200 - 15 * mainIndex;

         // TODO: tweak these values around
         Canny(grey, cannyEdge, 25, 2*autoThreshold, 3, true);
         GaussianBlur(cannyEdge, gaussianBlur, Size(3,3), 0);
         findContours(gaussianBlur, contours, hierarchy, CV_RETR_LIST, CHAIN_APPROX_SIMPLE);

         Mat contoursFrame = Mat::zeros(frame.size(), CV_8UC3);
         Mat convexHullFrame = Mat::zeros(frame.size(), CV_8UC3);

         convexes.resize(contours.size());

         for(size_t index = 0; index < contours.size(); ++index)
         {
             convexHull(contours[index], convexes[index]);
         }

         cout << "Found " << thresholdCentres.size() << " circles at threshold " << autoThreshold << endl;

         for(size_t index = 0; index < contours.size(); ++index)
         {
             Scalar colour = Scalar(rng.uniform(0, 255), rng.uniform(0, 255), rng.uniform(0, 255));
             drawContours(contoursFrame, contours, int(index), colour, 2, LINE_8);
             drawContours(convexHullFrame, convexes, int(index), colour, 2, LINE_8);
         }

         imshow("Convex", convexHullFrame);
         imshow("Contours", contoursFrame);
     }

     char tappedKey = waitKey(1);

     if(tappedKey == 27)
     {
         break;
     }
 }

 destroyAllWindows();

 camera.release();

 return 0;
}