Revision history [back]

I did, some time ago, a face tracking algorithm using template matching and kalman filter in order to reduce the window to perform template matching. I implemented a simple class that i called "FaceKalman" based on this mouse kalman. The code (cpp) is something like this:

FaceKalman::FaceKalman(){
    // TODO Auto-generated constructor stub

}

KalmanFilter FaceKalman::initKalman(Point coord)
{
     /* 
       (x, y, Vx, Vy)
        position of the object (x,y)
        velocity (Vx,Vy)
      */            


    this->measurement =  Mat(2,1,DataType<float>::type);

    KalmanFilter KF(4, 2, 0);

    KF.statePre.at<float>(0) = coord.x;
    KF.statePre.at<float>(1) = coord.y;
    KF.statePre.at<float>(2) = 0;
    KF.statePre.at<float>(3) = 0;

    //VELOCITY NOT TAKEN INTO ACCOUNT
    //KF.transitionMatrix = *(Mat_<float>(4, 4) << 1,0,0,0,   0,1,0,0,  0,0,1,0,  0,0,0,1);

    //VELOCITY TAKEN INTO ACCOUNT
    KF.transitionMatrix = *(Mat_<float>(4, 4) << 1,0,1,0,   0,1,0,1,  0,0,1,0,  0,0,0,1);


    setIdentity(KF.measurementMatrix);
    setIdentity(KF.processNoiseCov, Scalar::all(1e-4));
    setIdentity(KF.measurementNoiseCov, Scalar::all(1e-1));
    setIdentity(KF.errorCovPost, Scalar::all(.1));      

    return KF;
}

Point FaceKalman::updateKalman(KalmanFilter KF, Point coord)
{
    //predict
    Mat prediction = KF.predict();
    Point predictPt(prediction.at<float>(0),prediction.at<float>(1));

    this->measurement(0) = coord.x;
    this->measurement(1) = coord.y;

    //correct
    Mat estimated = KF.correct(this->measurement);
    Point estimatedPoint(estimated.at<float>(0),estimated.at<float>(1));        
    return estimatedPoint;
}

Then, you can use your FaceKalman class for face detection in order to reduce the window to look for the face.