hybridtracker.cpp

//*M///////////////////////////////////////////////////////////////////////////////////////
//
//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
//  By downloading, copying, installing or using the software you agree to this license.
//  If you do not agree to this license, do not download, install,
//  copy or use the software.
//
//
//                                License Agreement
//                       For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2008-2011, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
//   * Redistribution's of source code must retain the above copyright notice,
//     this list of conditions and the following disclaimer.
//
//   * Redistribution's in binary form must reproduce the above copyright notice,
//     this list of conditions and the following disclaimer in the documentation
//     and/or other materials provided with the distribution.
//
//   * The name of Intel Corporation may not be used to endorse or promote products
//     derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
#include "opencv2/contrib/hybridtracker.hpp"

using namespace cv;
using namespace std;

CvHybridTrackerParams::CvHybridTrackerParams(float _ft_tracker_weight, float _ms_tracker_weight,
            CvFeatureTrackerParams _ft_params,
            CvMeanShiftTrackerParams _ms_params,
            CvMotionModel)
{
    ft_tracker_weight = _ft_tracker_weight;
    ms_tracker_weight = _ms_tracker_weight;
    ft_params = _ft_params;
    ms_params = _ms_params;
}

CvMeanShiftTrackerParams::CvMeanShiftTrackerParams(int _tracking_type, CvTermCriteria _term_crit)
{
    tracking_type = _tracking_type;
    term_crit = _term_crit;
}

CvHybridTracker::CvHybridTracker() {

}

CvHybridTracker::CvHybridTracker(HybridTrackerParams _params) :
    params(_params) {
    params.ft_params.feature_type = CvFeatureTrackerParams::SIFT;
    mstracker = new CvMeanShiftTracker(params.ms_params);
    fttracker = new CvFeatureTracker(params.ft_params);
}

CvHybridTracker::~CvHybridTracker() {
    if (mstracker != NULL)
        delete mstracker;
    if (fttracker != NULL)
        delete fttracker;
}

inline float CvHybridTracker::getL2Norm(Point2f p1, Point2f p2) {
    float distance = (p1.x - p2.x) * (p1.x - p2.x) + (p1.y - p2.y) * (p1.y
            - p2.y);
    return sqrt(distance);
}

Mat CvHybridTracker::getDistanceProjection(Mat image, Point2f center) {
    Mat hist(image.size(), CV_64F);

    double lu = getL2Norm(Point(0, 0), center);
    double ru = getL2Norm(Point(0, image.size().width), center);
    double rd = getL2Norm(Point(image.size().height, image.size().width),
            center);
    double ld = getL2Norm(Point(image.size().height, 0), center);

    double max = (lu < ru) ? lu : ru;
    max = (max < rd) ? max : rd;
    max = (max < ld) ? max : ld;

    for (int i = 0; i < hist.rows; i++)
        for (int j = 0; j < hist.cols; j++)
            hist.at<double> (i, j) = 1.0 - (getL2Norm(Point(i, j), center)
                    / max);

    return hist;
}

Mat CvHybridTracker::getGaussianProjection(Mat image, int ksize, double sigma,
        Point2f center) {
    Mat kernel = getGaussianKernel(ksize, sigma, CV_64F);
    double max = kernel.at<double> (ksize / 2);

    Mat hist(image.size(), CV_64F);
    for (int i = 0; i < hist.rows; i++)
        for (int j = 0; j < hist.cols; j++) {
            int pos = cvRound(getL2Norm(Point(i, j), center));
            if (pos < ksize / 2.0)
                hist.at<double> (i, j) = 1.0 - (kernel.at<double> (pos) / max);
        }

    return hist;
}

void CvHybridTracker::newTracker(Mat image, Rect selection) {
    prev_proj = Mat::zeros(image.size(), CV_64FC1);
    prev_center = Point2f(selection.x + selection.width / 2.0f, selection.y
            + selection.height / 2.0f);
    prev_window = selection;

    mstracker->newTrackingWindow(image, selection);
    fttracker->newTrackingWindow(image, selection);

    samples = cvCreateMat(2, 1, CV_32FC1);
    labels = cvCreateMat(2, 1, CV_32SC1);

    ittr = 0;
}

void CvHybridTracker::updateTracker(Mat image) {
    ittr++;

    //copy over clean images: TODO
    mstracker->updateTrackingWindow(image);
    fttracker->updateTrackingWindowWithFlow(image);

    if (params.motion_model == CvMotionModel::EM)
        updateTrackerWithEM(image);
    else
        updateTrackerWithLowPassFilter(image);

    // Regression to find new weights
    Point2f ms_center = mstracker->getTrackingEllipse().center;
    Point2f ft_center = fttracker->getTrackingCenter();

#ifdef DEBUG_HYTRACKER
    circle(image, ms_center, 3, Scalar(0, 0, 255), -1, 8);
    circle(image, ft_center, 3, Scalar(255, 0, 0), -1, 8);
    putText(image, "ms", Point(ms_center.x+2, ms_center.y), FONT_HERSHEY_PLAIN, 0.75, Scalar(255, 255, 255));
    putText(image, "ft", Point(ft_center.x+2, ft_center.y), FONT_HERSHEY_PLAIN, 0.75, Scalar(255, 255, 255));
#endif

    double ms_len = getL2Norm(ms_center, curr_center);
    double ft_len = getL2Norm(ft_center, curr_center);
    double total_len = ms_len + ft_len;

    params.ms_tracker_weight *= (ittr - 1);
    params.ms_tracker_weight += (float)((ms_len / total_len));
    params.ms_tracker_weight /= ittr;
    params.ft_tracker_weight *= (ittr - 1);
    params.ft_tracker_weight += (float)((ft_len / total_len));
    params.ft_tracker_weight /= ittr;

    circle(image, prev_center, 3, Scalar(0, 0, 0), -1, 8);
    circle(image, curr_center, 3, Scalar(255, 255, 255), -1, 8);

    prev_center = curr_center;
    prev_window.x = (int)(curr_center.x-prev_window.width/2.0);
    prev_window.y = (int)(curr_center.y-prev_window.height/2.0);

    mstracker->setTrackingWindow(prev_window);
    fttracker->setTrackingWindow(prev_window);
}

void CvHybridTracker::updateTrackerWithEM(Mat image) {
    Mat ms_backproj = mstracker->getHistogramProjection(CV_64F);
    Mat ms_distproj = getDistanceProjection(image, mstracker->getTrackingCenter());
    Mat ms_proj = ms_backproj.mul(ms_distproj);

    float dist_err = getL2Norm(mstracker->getTrackingCenter(), fttracker->getTrackingCenter());
    Mat ft_gaussproj = getGaussianProjection(image, cvRound(dist_err), -1, fttracker->getTrackingCenter());
    Mat ft_distproj = getDistanceProjection(image, fttracker->getTrackingCenter());
    Mat ft_proj = ft_gaussproj.mul(ft_distproj);

    Mat proj = params.ms_tracker_weight * ms_proj + params.ft_tracker_weight * ft_proj + prev_proj;

    int sample_count = countNonZero(proj);
    cvReleaseMat(&samples);
    cvReleaseMat(&labels);
    samples = cvCreateMat(sample_count, 2, CV_32FC1);
    labels = cvCreateMat(sample_count, 1, CV_32SC1);

    int count = 0;
    for (int i = 0; i < proj.rows; i++)
        for (int j = 0; j < proj.cols; j++)
            if (proj.at<double> (i, j) > 0) {
                samples->data.fl[count * 2] = (float)i;
                samples->data.fl[count * 2 + 1] = (float)j;
                count++;
            }

    cv::Mat lbls;

    EM em_model(1, EM::COV_MAT_SPHERICAL, TermCriteria(TermCriteria::COUNT + TermCriteria::EPS, 10000, 0.001));
    em_model.train(cvarrToMat(samples), noArray(), lbls);
    if(labels)
        lbls.copyTo(cvarrToMat(labels));

    Mat em_means = em_model.get<Mat>("means");
    curr_center.x = (float)em_means.at<float>(0, 0);
    curr_center.y = (float)em_means.at<float>(0, 1);
}

void CvHybridTracker::updateTrackerWithLowPassFilter(Mat) {
    RotatedRect ms_track = mstracker->getTrackingEllipse();
    Point2f ft_center = fttracker->getTrackingCenter();

    float a = params.low_pass_gain;
    curr_center.x = (1 - a) * prev_center.x + a * (params.ms_tracker_weight * ms_track.center.x + params.ft_tracker_weight * ft_center.x);
    curr_center.y = (1 - a) * prev_center.y + a * (params.ms_tracker_weight * ms_track.center.y + params.ft_tracker_weight * ft_center.y);
}

Rect CvHybridTracker::getTrackingWindow() {
    return prev_window;
}