isarandi's profile - activity

https://en.wikipedia.org/wiki/Cascadi...

@Junglee What is your application about, what is your goal? How long did you wait for it to disappear?

You probably need to wait until the background model adapts to the new appearance there. Or you may want to use a different background modeling approach. Perhaps a simple thresholded absolute difference with an empty scene is enough.

imshow expects BGR images. If the result is messed up, I think the problem is with the VideoCapture. Maybe VideoCapture gives RGB? Anyway you can use cv::cvtColor to convert between BGR and RGB.

cv::cvtColor(rgbMat, bgrMat, cv::COLOR_RGB2BGR);
cv::cvtColor(bgrMat, rgbMat, cv::COLOR_BGR2RGB);

Disclaimer: I only have experience with OpenCV on desktop.

Connected Component Labeling may be a good solution here. Take a look at this implementation.

Maybe the video is compressed?

I don't know the solution but why don't you write a function instead of copying code and suffixing variable names? Also in the line h1, w1 = img.shape[:2] you forgot to change img to img1.

You can extract rectangular regions from a Mat.

cv::Mat bigImage = cv::Mat::zeros(cv::Size(660,350));
cv::Mat smallImage = cv::Mat(bigImage, cv::Rect(0,0,110,70));

cv::Rect takes the x, y, width, height as constructor parameters.

Note that this will not copy the image data, it just creates another wrapper to the same image data. If you need data copy, use

cv::Mat smallImage = cv::Mat(bigImage, cv::Rect(0,0,110,70)).clone();

Now if you need to extract multiple such image parts, essentially slicing your image with a grid, you can create a for loop as follows.

cv::Size smallSize(110,70);
std::vector<Mat> smallImages;

for (int y = 0; y < bigImage.rows; y += smallSize.height)
{
    for (int x = 0; x < bigImage.cols; x += smallSize.width)
    {
        cv::Rect rect =  cv::Rect(x,y, smallSize.width, smallSize.height);
        smallImages.push_back(cv::Mat(bigImage, rect));
    }
}

Here I assumed that the size of the bigImage is an integer multiple of the size of the desired small images.

This is a texture problem. There are many texture descriptors that could be used here. It really depends on what the percentage should represent and what you want to do with it later. Maybe the simple approach of calculating the contrast of the image or the variance of the pixel values or the gradient strengths could work.

Mat can contain doubles as matrix elements. Google for OpenCV Basic Structures and you will find a page about Mat. Do you already have another double matrix class that you are trying to convert to Mat? Please be more specific. You are more likely to find help if you ask in a clear way.

Difference between what? How do you define the 'grain' of an image? Percentage of what? Are you using some kind of texture descriptors? Or is your question about choosing such texture descriptors?

Please spend more time to formulate your questions more clearly. Otherwise you will have a bad experience here and not know why. People will be helpful but first you need to put some effort into asking your question. What is your goal? You can even upload example images.

I'm trying to convert a column of a cv::Mat to an std::vector. Mat has such a type conversion operator but it doesn't do what I would expect. Let's create a 10 by 2 matrix and add ascending elements in row major order starting with 1.

cv::Mat m(10,2, CV_64F);

int num = 1;
for (int row = 0; row < m.rows; ++row)
{
    for (int col = 0; col < m.cols; ++col)
    {
        m.at<double>(row,col) = (double)(num++);
    }
}

std::vector<double> v = m.col(0);
cout << v[0] << endl; //outputs 1, OK
cout << v[1] << endl; //outputs 0 instead of 3
cout << v[2] << endl; //outputs 0 instead of 5, etc.

However, if I replace the line

std::vector<double> v = m.col(0);

with

std::vector<double> v = m.col(0).clone();

then it works as expected and outputs lines 1, 3, 5. With some other matrix sizes (e.g. 120x2) it even crashes at memcpy (inside cv::Mat::copyTo(cv::OutputArray)) with segmentation fault or at vector destruction.

I'm using OpenCV 2.4.8 Debug build, GCC 4.8.2, Ubuntu 14.04 LTS 64-bit.

I'm trying to step into OpenCV source files while debugging. I'm using Qt Creator on Ubuntu 14.04 64-bit with GDB and GCC. OpenCV version 2.4.8. I built OpenCV from source using CMAKE_BUILD_TYPE=Debug and later also with ENABLE_PROFILING just to make sure. Still, I cannot jump into the OpenCV source when I'm debugging. Do I need to set something in the .pro project file? Or maybe set up some Source Mapping in Qt Creator settings? Thanks!