How to use QAbstractVideoFilter to process frames from camera

DennisZhang

@Leonardo Thanks for your applying!

I want to implement ViBe Algorithm and Human Skeletonization by OpenGL ES 2.0(Shader Language), and because of heavy computation needed by these Algorithms, I must implement them using OpenGL ES 2.0.

I know there is a convenient function in QtQuick - that is ShaderEffect , but I can't use it because it limits too many functions such as creating external texture buffer for further computing, so I need to focus on the original way to use OpenGL ES (using C++) in cooperation with QAbstractVideoFilter.

SGaist

Hi and welcome to devnet,

You can find some more information in this blog post

Note that the finished signal is just for the example here to communicate a result to the outside. If you only modify the image then there's no need for it.

theshadowx

Hi @DennisZhang ,

I have the same problem as you. First when I emit the signal from QVideoFilterRunnable , I got :

Undefined reference to subFilterClass::finished(...)

So then I comment it out.

and in the QML : if I call onFinished, I got this error while executing the app:

Cannot assign to non-existent property "onFinished"

have you been able to solve this problem ?

agocs

There is no built-in finished() signal. You have to declare it yourself and can be called anything.

So in the QAbstractVideoFilter subclass, have

signals:
void finished();

then in the QVideoFilterRunnable, do

emit m_filter->finished();

when the computation is done.

belab

Hi DennisZhang,

Is your file processed by moc? I guess you forget the QOBJECT macro.

theshadowx

@agocs of course I declared it in signals

kolegs

I had the same problem, tried diffrent ways and it worked when I declared both childs of QVideoFilterRunnable and QAbstractVideoFilter in the same header and source file(so I have 2 files, one .h and one .cpp)

My header file looks like this:

class QRCode_filter : public QAbstractVideoFilter
{
    Q_OBJECT
public:
    QRCode_filter();
    QVideoFilterRunnable *createFilterRunnable();
private:
signals:
    void finished(QObject *result);
public slots:

};

class MyQRCodeFilterRunnable : public QVideoFilterRunnable
{
public:
    QVideoFrame run(QVideoFrame *input, const QVideoSurfaceFormat &surfaceFormat, RunFlags flags);
    MyQRCodeFilterRunnable(QRCode_filter *filter = NULL);
private:
    QRCode_filter *m_filter;
};

theshadowx

@kolegs Thanks for sharing this solution.

theshadowx

@kolegs Have you tried using OpenCV in

QVideoFrame run(QVideoFrame *input, const QVideoSurfaceFormat &surfaceFormat, RunFlags flags);

I have a problem about that :

QVideoFrame FilterRunnable::run(QVideoFrame *input, const QVideoSurfaceFormat &surfaceFormat, QVideoFilterRunnable::RunFlags flags)
{
    if(input->isValid()){
        input->map(QAbstractVideoBuffer::ReadWrite);
        /// QVideoFrame to cv::Mat
        Mat cvImg = Mat(input->height(),input->width(), CV_8UC3,input->bits(),input->bytesPerLine());

        /// Apply Filter
        Mat edges;

        /// to grayscale
        cvtColor(cvImg, edges, COLOR_YUV2GRAY_420);

        /// apply filters
        GaussianBlur(edges, edges, Size(7,7), 1.5, 1.5);
        Canny(edges, edges, 0, 30, 3);

        /// convert to YUV420
        cvtColor(edges,edges, COLOR_GRAY2RGB);
        cvtColor(edges,edges, COLOR_RGB2YUV_I420);

        ///  what to do here to send back the modified frame .....   ???
    }

     return *input;
}

kolegs

@theshadowx I've never used openCV so I'm not sure how to use this libraries, but about run() function as far as I know you need to modify the frame that you get(QVideoFrame *input) and simply return it in the same format.

medyakovvit

@theshadowx, i don't know how to work with YUV format, but with QVideoFrame::Format_RGB32 you can do something like that:

QVideoFrame ThresholdFilterRunnable::run(QVideoFrame *input, const QVideoSurfaceFormat &surfaceFormat, QVideoFilterRunnable::RunFlags flags)
{
    if (!input->isValid())
        return *input;

    input->map(QAbstractVideoBuffer::ReadWrite);

    cv::Mat mat = cv::Mat(input->height(),input->width(), CV_8UC4, input->bits());

    cv::Mat gray, threshold, color;
    cv::cvtColor(mat, gray, COLOR_RGB2GRAY);
    cv::threshold(gray, threshold, 150.0, 255.0, THRESH_BINARY);
    cv::cvtColor(threshold, color, COLOR_GRAY2RGB);
    this->toVideoFrame(input, color);

    input->unmap();
    return *input;
}

void ThresholdFilterRunnable::toVideoFrame(QVideoFrame *input, cv::Mat &mat)
{
    int aCols = 4*mat.cols;
    int aRows = mat.rows;
    uchar* inputBits = input->bits();

    for (int i=0; i<aRows; i++)
    {
        uchar* p = mat.ptr<uchar>(i);
        for (int j=0; j<aCols; j++)
        {
            int index = i*aCols + j;
            inputBits[index] = p[j];
        }
    }
}

I can't test it right now, because i'm in hospital.

theshadowx

@medyakovvit Hi, thanks for responding, I hope everything is fine for you.

if my camera takes video in RGB32 it would be easier as I can create QImage from the edge matrix and then create a QVideoFrame from QImage.:

 ///cv::Mat to QImage
 QImage ImgDest(edges.data, edges.cols, edges.rows, edges.step, QImage::Format_RGB888);
 QVideoFrame *output = new QVideoFrame(ImgDest);

And then I would return the output.

The problem (From what I understood) is that my camera generates images with YUV color scheme. which means that QVideoSurfaceFormat is in YUV, so if I use RGB32 format it won't work.

Unfortunately QImage doesn't support YUV format. :(

medyakovvit

@theshadowx So, if correctly understood this YUV420 and YUV format, Y - luma component, U and V - color component. And in single frame Ys go first, then Us and Vs. So if you need grayscale image, you can work with Ys and ignore Us and Vs. Maybe you can try:

QVideoFrame YUVFilterRunnable::run(QVideoFrame *input, const QVideoSurfaceFormat &surfaceFormat, QVideoFilterRunnable::RunFlags flags)
{
    if (!input->isValid())
        return *input;

    input->map(QAbstractVideoBuffer::ReadWrite);

    this->deleteColorComponentFromYUV(input);
    cv::Mat mat(input->height(),input->width(), CV_8UC1, input->bits()); // create grayscale mat with input's Ys only and avoid additional color conversion and copying

    cv::GaussianBlur(mat, mat, Size(7,7), 1.5, 1.5);
    cv::Canny(mat, mat, 0, 30, 3);

    input->unmap();
    return *input;
}

void YUVFilterRunnable::deleteColorComponentFromYUV(QVideoFrame *input)
{
    // Assign 0 to Us and Vs
    int firstU = input->width()*input->height(); // if i correctly understand YUV420
    int lastV = input->width()*input->height() + input->width()*input->height()/4*2;
    uchar* inputBits = input->bits();

    for (int i=firstU; i<lastV; i++)
        inputBits[i] = 0;
}

theshadowx

@medyakovvit

Yes that's right, there is one porblem in the code is that with you code I get a greenscale. So to convert to grayscale the solution is that instead of 0 for chroma, it should be 127 (http://stackoverflow.com/a/20609599/2775917) :

void YUVFilterRunnable::deleteColorComponentFromYUV(QVideoFrame *input)
{
    // Assign 0 to Us and Vs
    int firstU = input->width()*input->height(); // if i correctly understand YUV420
    int lastV = input->width()*input->height() + input->width()*input->height()/4*2;
    uchar* inputBits = input->bits();

    for (int i=firstU; i<lastV; i++)
        inputBits[i] = 127;
}

Thanks a lot for your help

medyakovvit

@theshadowx I'm glad to help

cs_a994

Hello!

Is it possible to use the QVideoFilterRunnuble without QML?

As an example I would like to embed it into Camera example ( http://doc.qt.io/qt-5/qtmultimediawidgets-camera-example.html ) to be able to change data format.

Thank you.

medyakovvit

@cs_a994 Don't think so. But you can try to subclass QAbstractVideoSurface.

cs_a994

@medyakovvit

Thank you. I know about this way. But use of filters seams to me more neat.

I tryed to use filters with QML ( in the Camera example ) and did not yet get a result.

medyakovvit

@cs_a994
OK. I f you have a questions about qml camera and filters, I will try to help you.

theshadowx

@cs_a994

I have put a QtQuick app in github using QAbstractVideoFilter with OpenCV.

I hope it helps you.

Cheers