Dual TV-L1 optical flow implementation

modules/video/doc/motion_analysis_and_object_tracking.rst

@@ -641,6 +641,72 @@ Calculate an optical flow using "SimpleFlow" algorithm.
 
 See [Tao2012]_. And site of project - http://graphics.berkeley.edu/papers/Tao-SAN-2012-05/.
 
+
+
+OpticalFlowDual_TVL1
+--------------------
+"Dual TV L1" Optical Flow Algorithm.
+
+.. ocv:class:: OpticalFlowDual_TVL12
+
+
+The class implements the "Dual TV L1" optical flow algorithm described in [Zach2007]_ and [Javier2012]_ .
+
+Here are important members of the class that control the algorithm, which you can set after constructing the class instance:
+
+    .. ocv:member:: double tau
+
+        Time step of the numerical scheme.
+
+    .. ocv:member:: double lambda
+
+        Weight parameter for the data term, attachment parameter. This is the most relevant parameter, which determines the smoothness of the output. The smaller this parameter is, the smoother the solutions we obtain. It depends on the range of motions of the images, so its value should be adapted to each image sequence.
+
+    .. ocv:member:: double theta
+
+        Weight parameter for (u - v)^2, tightness parameter. It serves as a link between the attachment and the regularization terms. In theory, it should have a small value in order to maintain both parts in correspondence. The method is stable for a large range of values of this parameter.
+
+    .. ocv:member:: int nscales
+
+        Number of scales used to create the pyramid of images.
+
+    .. ocv:member:: int warps
+
+        Number of warpings per scale. Represents the number of times that I1(x+u0) and grad( I1(x+u0) ) are computed per scale. This is a parameter that assures the stability of the method. It also affects the running time, so it is a compromise between speed and accuracy.
+
+    .. ocv:member:: double epsilon
+
+        Stopping criterion threshold used in the numerical scheme, which is a trade-off between precision and running time. A small value will yield more accurate solutions at the expense of a slower convergence.
+
+    .. ocv:member:: int iterations
+
+        Stopping criterion iterations number used in the numerical scheme.
+
+
+
+
+OpticalFlowDual_TVL1::operator()
+--------------------------------
+Calculates an optical flow.
+
+.. ocv:function:: void OpticalFlowDual_TVL1::operator ()(InputArray I0, InputArray I1, InputOutputArray flow)
+
+    :param prev: first 8-bit single-channel input image.
+
+    :param next: second input image of the same size and the same type as ``prev`` .
+
+    :param flow: computed flow image that has the same size as ``prev`` and type ``CV_32FC2`` .
+
+
+
+OpticalFlowDual_TVL1::collectGarbage
+------------------------------------
+Releases all inner buffers.
+
+.. ocv:function:: void OpticalFlowDual_TVL1::collectGarbage()
+
+
+
 .. [Bouguet00] Jean-Yves Bouguet. Pyramidal Implementation of the Lucas Kanade Feature Tracker.
 
 .. [Bradski98] Bradski, G.R. "Computer Vision Face Tracking for Use in a Perceptual User Interface", Intel, 1998
@@ -658,3 +724,7 @@ See [Tao2012]_. And site of project - http://graphics.berkeley.edu/papers/Tao-SA
 .. [Welch95] Greg Welch and Gary Bishop “An Introduction to the Kalman Filter”, 1995
 
 .. [Tao2012] Michael Tao, Jiamin Bai, Pushmeet Kohli and Sylvain Paris. SimpleFlow: A Non-iterative, Sublinear Optical Flow Algorithm. Computer Graphics Forum (Eurographics 2012)
+
+.. [Zach2007] C. Zach, T. Pock and H. Bischof. "A Duality Based Approach for Realtime TV-L1 Optical Flow", In Proceedings of Pattern Recognition (DAGM), Heidelberg, Germany, pp. 214-223, 2007
+
+.. [Javier2012] Javier Sanchez, Enric Meinhardt-Llopis and Gabriele Facciolo. "TV-L1 Optical Flow Estimation".

modules/video/include/opencv2/video/tracking.hpp

@@ -352,6 +352,105 @@ CV_EXPORTS_W void calcOpticalFlowSF(Mat& from,
                                     double upscale_sigma_color,
                                     double speed_up_thr);
 
+// Implementation of the Zach, Pock and Bischof Dual TV-L1 Optical Flow method
+//
+// see reference:
+//   [1] C. Zach, T. Pock and H. Bischof, "A Duality Based Approach for Realtime TV-L1 Optical Flow".
+//   [2] Javier Sanchez, Enric Meinhardt-Llopis and Gabriele Facciolo. "TV-L1 Optical Flow Estimation".
+class CV_EXPORTS OpticalFlowDual_TVL1
+{
+public:
+    OpticalFlowDual_TVL1();
+
+    void operator ()(InputArray I0, InputArray I1, InputOutputArray flow);
+
+    void collectGarbage();
+
+    /**
+     * Time step of the numerical scheme.
+     */
+    double tau;
+
+    /**
+     * Weight parameter for the data term, attachment parameter.
+     * This is the most relevant parameter, which determines the smoothness of the output.
+     * The smaller this parameter is, the smoother the solutions we obtain.
+     * It depends on the range of motions of the images, so its value should be adapted to each image sequence.
+     */
+    double lambda;
+
+    /**
+     * Weight parameter for (u - v)^2, tightness parameter.
+     * It serves as a link between the attachment and the regularization terms.
+     * In theory, it should have a small value in order to maintain both parts in correspondence.
+     * The method is stable for a large range of values of this parameter.
+     */
+    double theta;
+
+    /**
+     * Number of scales used to create the pyramid of images.
+     */
+    int nscales;
+
+    /**
+     * Number of warpings per scale.
+     * Represents the number of times that I1(x+u0) and grad( I1(x+u0) ) are computed per scale.
+     * This is a parameter that assures the stability of the method.
+     * It also affects the running time, so it is a compromise between speed and accuracy.
+     */
+    int warps;
+
+    /**
+     * Stopping criterion threshold used in the numerical scheme, which is a trade-off between precision and running time.
+     * A small value will yield more accurate solutions at the expense of a slower convergence.
+     */
+    double epsilon;
+
+    /**
+     * Stopping criterion iterations number used in the numerical scheme.
+     */
+    int iterations;
+
+    bool useInitialFlow;
+
+private:
+    void procOneScale(const Mat_<float>& I0, const Mat_<float>& I1, Mat_<float>& u1, Mat_<float>& u2);
+
+    std::vector<Mat_<float> > I0s;
+    std::vector<Mat_<float> > I1s;
+    std::vector<Mat_<float> > u1s;
+    std::vector<Mat_<float> > u2s;
+
+    Mat_<float> I1x_buf;
+    Mat_<float> I1y_buf;
+
+    Mat_<float> flowMap1_buf;
+    Mat_<float> flowMap2_buf;
+
+    Mat_<float> I1w_buf;
+    Mat_<float> I1wx_buf;
+    Mat_<float> I1wy_buf;
+
+    Mat_<float> grad_buf;
+    Mat_<float> rho_c_buf;
+
+    Mat_<float> v1_buf;
+    Mat_<float> v2_buf;
+
+    Mat_<float> p11_buf;
+    Mat_<float> p12_buf;
+    Mat_<float> p21_buf;
+    Mat_<float> p22_buf;
+
+    Mat_<float> div_p1_buf;
+    Mat_<float> div_p2_buf;
+
+    Mat_<float> u1x_buf;
+    Mat_<float> u1y_buf;
+    Mat_<float> u2x_buf;
+    Mat_<float> u2y_buf;
+};
+
 }
 
 #endif

modules/video/perf/perf_tvl1optflow.cpp

+#include "perf_precomp.hpp"
+
+using namespace std;
+using namespace cv;
+using namespace perf;
+
+typedef TestBaseWithParam< pair<string, string> > ImagePair;
+
+pair<string, string> impair(const char* im1, const char* im2)
+{
+    return make_pair(string(im1), string(im2));
+}
+
+PERF_TEST_P(ImagePair, OpticalFlowDual_TVL1, testing::Values(impair("cv/optflow/RubberWhale1.png", "cv/optflow/RubberWhale2.png")))
+{
+    declare.time(40);
+
+    Mat frame1 = imread(getDataPath(GetParam().first), IMREAD_GRAYSCALE);
+    Mat frame2 = imread(getDataPath(GetParam().second), IMREAD_GRAYSCALE);
+    ASSERT_FALSE(frame1.empty());
+    ASSERT_FALSE(frame2.empty());
+
+    Mat flow;
+
+    OpticalFlowDual_TVL1 tvl1;
+
+    TEST_CYCLE()
+    {
+        tvl1(frame1, frame2, flow);
+    }
+
+    SANITY_CHECK(flow);
+}

modules/video/test/test_tvl1optflow.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.
+//
+//
+//                        Intel License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000, Intel Corporation, 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 "test_precomp.hpp"
+#include <fstream>
+
+using namespace std;
+using namespace cv;
+using namespace cvtest;
+
+//#define DUMP
+
+namespace
+{
+    // first four bytes, should be the same in little endian
+    const float FLO_TAG_FLOAT = 202021.25f;  // check for this when READING the file
+    const char FLO_TAG_STRING[] = "PIEH";    // use this when WRITING the file
+
+    // binary file format for flow data specified here:
+    // http://vision.middlebury.edu/flow/data/
+    void writeOpticalFlowToFile(const Mat_<Point2f>& flow, const string& fileName)
+    {
+        ofstream file(fileName.c_str(), ios_base::binary);
+
+        file << FLO_TAG_STRING;
+
+        file.write((const char*) &flow.cols, sizeof(int));
+        file.write((const char*) &flow.rows, sizeof(int));
+
+        for (int i = 0; i < flow.rows; ++i)
+        {
+            for (int j = 0; j < flow.cols; ++j)
+            {
+                const Point2f u = flow(i, j);
+
+                file.write((const char*) &u.x, sizeof(float));
+                file.write((const char*) &u.y, sizeof(float));
+            }
+        }
+    }
+
+    // binary file format for flow data specified here:
+    // http://vision.middlebury.edu/flow/data/
+    void readOpticalFlowFromFile(Mat_<Point2f>& flow, const string& fileName)
+    {
+        ifstream file(fileName.c_str(), ios_base::binary);
+
+        float tag;
+        file.read((char*) &tag, sizeof(float));
+        CV_Assert( tag == FLO_TAG_FLOAT );
+
+        Size size;
+
+        file.read((char*) &size.width, sizeof(int));
+        file.read((char*) &size.height, sizeof(int));
+
+        flow.create(size);
+
+        for (int i = 0; i < flow.rows; ++i)
+        {
+            for (int j = 0; j < flow.cols; ++j)
+            {
+                Point2f u;
+
+                file.read((char*) &u.x, sizeof(float));
+                file.read((char*) &u.y, sizeof(float));
+
+                flow(i, j) = u;
+            }
+        }
+    }
+}
+
+TEST(Video_calcOpticalFlowDual_TVL1, Regression)
+{
+    const string frame1_path = TS::ptr()->get_data_path() + "optflow/RubberWhale1.png";
+    const string frame2_path = TS::ptr()->get_data_path() + "optflow/RubberWhale2.png";
+    const string gold_flow_path = TS::ptr()->get_data_path() + "optflow/tvl1_flow.flo";
+
+    Mat frame1 = imread(frame1_path, IMREAD_GRAYSCALE);
+    Mat frame2 = imread(frame2_path, IMREAD_GRAYSCALE);
+    ASSERT_FALSE(frame1.empty());
+    ASSERT_FALSE(frame2.empty());
+
+    Mat_<Point2f> flow;
+    OpticalFlowDual_TVL1 tvl1;
+
+    tvl1(frame1, frame2, flow);
+
+#ifdef DUMP
+    writeOpticalFlowToFile(flow, gold_flow_path);
+#else
+    Mat_<Point2f> gold;
+    readOpticalFlowFromFile(gold, gold_flow_path);
+    double err = norm(gold, flow, NORM_INF);
+    EXPECT_EQ(0.0f, err);
+#endif
+}

samples/cpp/tvl1_optical_flow.cpp

+#include <iostream>
+#include <fstream>
+
+#include "opencv2/video/tracking.hpp"
+#include "opencv2/highgui/highgui.hpp"
+
+using namespace cv;
+using namespace std;
+
+inline bool isFlowCorrect(Point2f u)
+{
+    return !cvIsNaN(u.x) && !cvIsNaN(u.y) && fabs(u.x) < 1e9 && fabs(u.y) < 1e9;
+}
+
+static Vec3b computeColor(float fx, float fy)
+{
+    static bool first = true;
+
+    // relative lengths of color transitions:
+    // these are chosen based on perceptual similarity
+    // (e.g. one can distinguish more shades between red and yellow
+    //  than between yellow and green)
+    const int RY = 15;
+    const int YG = 6;
+    const int GC = 4;
+    const int CB = 11;
+    const int BM = 13;
+    const int MR = 6;
+    const int NCOLS = RY + YG + GC + CB + BM + MR;
+    static Vec3i colorWheel[NCOLS];
+
+    if (first)
+    {
+        int k = 0;
+
+        for (int i = 0; i < RY; ++i, ++k)
+            colorWheel[k] = Vec3i(255, 255 * i / RY, 0);
+
+        for (int i = 0; i < YG; ++i, ++k)
+            colorWheel[k] = Vec3i(255 - 255 * i / YG, 255, 0);
+
+        for (int i = 0; i < GC; ++i, ++k)
+            colorWheel[k] = Vec3i(0, 255, 255 * i / GC);
+
+        for (int i = 0; i < CB; ++i, ++k)
+            colorWheel[k] = Vec3i(0, 255 - 255 * i / CB, 255);
+
+        for (int i = 0; i < BM; ++i, ++k)
+            colorWheel[k] = Vec3i(255 * i / BM, 0, 255);
+
+        for (int i = 0; i < MR; ++i, ++k)
+            colorWheel[k] = Vec3i(255, 0, 255 - 255 * i / MR);
+
+        first = false;
+    }
+
+    const float rad = sqrt(fx * fx + fy * fy);
+    const float a = atan2(-fy, -fx) / CV_PI;
+
+    const float fk = (a + 1.0f) / 2.0f * (NCOLS - 1);
+    const int k0 = static_cast<int>(fk);
+    const int k1 = (k0 + 1) % NCOLS;
+    const float f = fk - k0;
+
+    Vec3b pix;
+
+    for (int b = 0; b < 3; b++)
+    {
+        const float col0 = colorWheel[k0][b] / 255.0;
+        const float col1 = colorWheel[k1][b] / 255.0;
+
+        float col = (1 - f) * col0 + f * col1;
+
+        if (rad <= 1)
+            col = 1 - rad * (1 - col); // increase saturation with radius
+        else
+            col *= .75; // out of range
+
+        pix[2 - b] = static_cast<int>(255.0 * col);
+    }
+
+    return pix;
+}
+
+static void drawOpticalFlow(const Mat_<Point2f>& flow, Mat& dst, float maxmotion = -1)
+{
+    dst.create(flow.size(), CV_8UC3);
+    dst.setTo(Scalar::all(0));
+
+    // determine motion range:
+    float maxrad = maxmotion;
+
+    if (maxmotion <= 0)
+    {
+        maxrad = 1;
+        for (int y = 0; y < flow.rows; ++y)
+        {
+            for (int x = 0; x < flow.cols; ++x)
+            {
+                Point2f u = flow(y, x);
+
+                if (!isFlowCorrect(u))
+                    continue;
+
+                maxrad = max(maxrad, sqrt(u.x * u.x + u.y * u.y));
+            }
+        }
+    }
+
+    for (int y = 0; y < flow.rows; ++y)
+    {
+        for (int x = 0; x < flow.cols; ++x)
+        {
+            Point2f u = flow(y, x);
+
+            if (isFlowCorrect(u))
+                dst.at<Vec3b>(y, x) = computeColor(u.x / maxrad, u.y / maxrad);
+        }
+    }
+}
+
+// binary file format for flow data specified here:
+// http://vision.middlebury.edu/flow/data/
+static void writeOpticalFlowToFile(const Mat_<Point2f>& flow, const string& fileName)
+{
+    static const char FLO_TAG_STRING[] = "PIEH";
+
+    ofstream file(fileName.c_str(), ios_base::binary);
+
+    file << FLO_TAG_STRING;
+
+    file.write((const char*) &flow.cols, sizeof(int));
+    file.write((const char*) &flow.rows, sizeof(int));
+
+    for (int i = 0; i < flow.rows; ++i)
+    {
+        for (int j = 0; j < flow.cols; ++j)
+        {
+            const Point2f u = flow(i, j);
+
+            file.write((const char*) &u.x, sizeof(float));
+            file.write((const char*) &u.y, sizeof(float));
+        }
+    }
+}
+
+int main(int argc, const char* argv[])
+{
+    if (argc < 3)
+    {
+        cerr << "Usage : " << argv[0] << "<frame0> <frame1> [<output_flow>]" << endl;
+        return -1;
+    }
+
+    Mat frame0 = imread(argv[1], IMREAD_GRAYSCALE);
+    Mat frame1 = imread(argv[2], IMREAD_GRAYSCALE);
+
+    if (frame0.empty())
+    {
+        cerr << "Can't open image ["  << argv[1] << "]" << endl;
+        return -1;
+    }
+    if (frame1.empty())
+    {
+        cerr << "Can't open image ["  << argv[2] << "]" << endl;
+        return -1;
+    }
+
+    if (frame1.size() != frame0.size())
+    {
+        cerr << "Images should be of equal sizes" << endl;
+        return -1;
+    }
+
+    Mat_<Point2f> flow;
+    OpticalFlowDual_TVL1 tvl1;
+
+    const double start = getTickCount();
+    tvl1(frame0, frame1, flow);
+    const double timeSec = (getTickCount() - start) / getTickFrequency();
+    cout << "calcOpticalFlowDual_TVL1 : " << timeSec << " sec" << endl;
+
+    Mat out;
+    drawOpticalFlow(flow, out);
+
+    if (argc == 4)
+        writeOpticalFlowToFile(flow, argv[3]);
+
+    imshow("Flow", out);
+    waitKey();
+
+    return 0;
+}