Merge pull request #899 from mshabunin:pr718

modules/datasets/include/opencv2/datasets/track_alov.hpp

+/*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) 2014, Itseez 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 the copyright holders 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 Itseez Inc 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*/
+
+#ifndef OPENCV_DATASETS_TRACK_ALOV_HPP
+#define OPENCV_DATASETS_TRACK_ALOV_HPP
+
+#include <string>
+#include <vector>
+
+#include "opencv2/datasets/dataset.hpp"
+#include "opencv2/datasets/util.hpp"
+
+using namespace std;
+
+namespace cv
+{
+namespace datasets
+{
+
+//! @addtogroup datasets_track
+//! @{
+
+struct TRACK_alovObj : public Object
+{
+    int id;
+    std::string imagePath;
+    vector <Point2f> gtbb;
+};
+
+const string sectionNames[] = { "01-Light", "02-SurfaceCover", "03-Specularity", "04-Transparency", "05-Shape", "06-MotionSmoothness", "07-MotionCoherence",
+"08-Clutter", "09-Confusion", "10-LowContrast", "11-Occlusion", "12-MovingCamera", "13-ZoomingCamera", "14-LongDuration" };
+
+const int sectionSizes[] = { 33, 15, 18, 20, 24, 22, 12, 15, 37, 23, 34, 22, 29, 10 };
+
+class CV_EXPORTS TRACK_alov : public Dataset
+{
+public:
+    static Ptr<TRACK_alov> create();
+
+    virtual void load(const std::string &path) = 0;
+
+    //Load only frames with annotations (~every 5-th frame)
+    virtual void loadAnnotatedOnly(const std::string &path) = 0;
+
+    virtual int getDatasetsNum() = 0;
+
+    virtual int getDatasetLength(int id) = 0;
+
+    virtual bool initDataset(int id) = 0;
+
+    virtual bool getNextFrame(Mat &frame) = 0;
+    virtual vector <Point2f> getNextGT() = 0;
+
+    //Get frame/GT by datasetID (1..N) frameID (1..K)
+    virtual bool getFrame(Mat &frame, int datasetID, int frameID) = 0;
+    virtual vector <Point2f> getGT(int datasetID, int frameID) = 0;
+
+protected:
+    vector <vector <Ptr<TRACK_alovObj> > > data;
+    int activeDatasetID;
+    int frameCounter;
+};
+
+//! @}
+
+}
+}
+
+#endif

modules/datasets/src/track_alov.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) 2014, Itseez 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 the copyright holders 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 Itseez Inc 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 "opencv2/datasets/track_alov.hpp"
+
+#include <sys/stat.h>
+#include <opencv2/core.hpp>
+#include <opencv2/highgui.hpp>
+
+using namespace std;
+
+namespace cv
+{
+namespace datasets
+{
+
+class TRACK_alovImpl : public TRACK_alov
+{
+public:
+    //Constructor
+    TRACK_alovImpl()
+    {
+        activeDatasetID = 1;
+        frameCounter = 0;
+    }
+    //Destructor
+    virtual ~TRACK_alovImpl() {}
+
+    //Load Dataset
+    virtual void load(const string &path);
+    virtual void loadAnnotatedOnly(const std::string &path);
+
+protected:
+    virtual int getDatasetsNum();
+
+    virtual int getDatasetLength(int id);
+
+    virtual bool initDataset(int id);
+
+    virtual bool getNextFrame(Mat &frame);
+    virtual bool getFrame(Mat &frame, int datasetID, int frameID);
+
+    virtual vector <Point2f> getNextGT();
+    virtual vector <Point2f> getGT(int datasetID, int frameID);
+
+    void loadDataset(const string &path);
+    void loadDatasetAnnotatedOnly(const string &path);
+
+    string fullFramePath(string rootPath, int sectionID, int videoID, int frameID);
+    string fullAnnoPath(string rootPath, int sectionID, int videoID);
+};
+
+
+void TRACK_alovImpl::load(const string &path)
+{
+    loadDataset(path);
+}
+
+void TRACK_alovImpl::loadAnnotatedOnly(const string &path)
+{
+    loadDatasetAnnotatedOnly(path);
+}
+
+string TRACK_alovImpl::fullFramePath(string rootPath, int sectionID, int videoID, int frameID)
+{
+    string out;
+    char videoNum[9];
+    sprintf(videoNum, "%u", videoID+1);
+    char frameNum[9];
+    sprintf(frameNum, "%u", frameID);
+    out = rootPath + "/imagedata++/" + sectionNames[sectionID] + "/" + sectionNames[sectionID] + "_video";
+
+    for (unsigned int i = 0; i < 5 - strlen(videoNum); ++i)
+    {
+        out += "0";
+    }
+
+    out += videoNum;
+    out += "/";
+
+    for (unsigned int i = 0; i < 8 - strlen(frameNum); ++i)
+    {
+        out += "0";
+    }
+
+    out += frameNum;
+    out += ".jpg";
+    return out;
+}
+
+string TRACK_alovImpl::fullAnnoPath(string rootPath, int sectionID, int videoID)
+{
+    string out;
+    char videoNum[9];
+    sprintf(videoNum, "%u", videoID+1);
+    out = rootPath + "/alov300++_rectangleAnnotation_full/" + sectionNames[sectionID] + "/" + sectionNames[sectionID] + "_video";
+
+    for (unsigned int i = 0; i < 5 - strlen(videoNum); ++i)
+    {
+        out += "0";
+    }
+
+    out += videoNum;
+    out += ".ann";
+    return out;
+}
+
+inline bool fileExists(const std::string& name)
+{
+    struct stat buffer;
+    return (stat(name.c_str(), &buffer) == 0);
+}
+
+void TRACK_alovImpl::loadDataset(const string &rootPath)
+{
+    vector <int> datasetsLengths;
+
+    printf("ALOV300++ Dataset Initialization...\n");
+
+    //Load frames
+    //Loop for all sections of ALOV300++ (14 sections)
+    for (int i = 0; i < 14; i++)
+    {
+        //Loop for all videos in section
+        for (int k = 0; k < sectionSizes[i]; k++)
+        {
+            vector <Ptr<TRACK_alovObj> > objects;
+
+            //Make a list of datasets lengths
+            int currFrameID = 0;
+
+            for (;;)
+            {
+                currFrameID++;
+                string fullPath = fullFramePath(rootPath, i, k, currFrameID);
+                if (!fileExists(fullPath))
+                    break;
+
+                //Make ALOV300++ Object
+                Ptr<TRACK_alovObj> currObj(new TRACK_alovObj);
+                currObj->imagePath = fullPath;
+                currObj->id = currFrameID;
+
+                currObj->gtbb.push_back(Point2d(0, 0));
+                currObj->gtbb.push_back(Point2d(0, 0));
+                currObj->gtbb.push_back(Point2d(0, 0));
+                currObj->gtbb.push_back(Point2d(0, 0));
+
+                //Add object to storage
+                objects.push_back(currObj);
+
+            }
+
+            datasetsLengths.push_back(currFrameID - 1);
+            data.push_back(objects);
+        }
+    }
+
+    //Load annotations
+    //Loop for all sections of ALOV300++ (14 sections)
+    int currDatasetID = 0;
+    for (int i = 0; i < 14; i++)
+    {
+        //Loop for all videos in section
+        for (int k = 0; k < sectionSizes[i]; k++)
+        {
+            currDatasetID++;
+
+            //Open dataset's ground truth (annotation) file
+            string annoPath = fullAnnoPath(rootPath, i, k);
+            ifstream annoList(annoPath.c_str());
+            if (!annoList.is_open())
+            {
+                printf("Error: Can't open annotation file *.ANN!!!\n");
+                break;
+            }
+
+            //Ground Truth data
+            int n = 0;
+            double    x1 = 0, y1 = 0,
+                x2 = 0, y2 = 0,
+                x3 = 0, y3 = 0,
+                x4 = 0, y4 = 0;
+
+            do
+            {
+                //Make ALOV300++ Object
+                string tmp;
+
+                getline(annoList, tmp);
+                std::istringstream in(tmp);
+                in >> n >> x1 >> y1 >> x2 >> y2 >> x3 >> y3 >> x4 >> y4;
+
+                Ptr<TRACK_alovObj> currObj = data[currDatasetID-1][n-1];
+
+                currObj->gtbb.clear();
+                currObj->gtbb.push_back(Point2d(x1, y1));
+                currObj->gtbb.push_back(Point2d(x2, y2));
+                currObj->gtbb.push_back(Point2d(x3, y3));
+                currObj->gtbb.push_back(Point2d(x4, y4));
+
+            } while (annoList.good());
+        }
+    }
+
+    return;
+}
+
+void TRACK_alovImpl::loadDatasetAnnotatedOnly(const string &rootPath)
+{
+    vector <int> datasetsLengths;
+    int currDatasetID = 0;
+
+    printf("ALOV300++ Annotated Dataset Initialization...\n");
+
+    //Loop for all sections of ALOV300++ (14 sections)
+    for (int i = 0; i < 14; i++)
+    {
+        //Loop for all videos in section
+        for (int k = 0; k < sectionSizes[i]; k++)
+        {
+            vector <Ptr<TRACK_alovObj> > objects;
+            currDatasetID++;
+
+            //Open dataset's ground truth (annotation) file
+            string annoPath = fullAnnoPath(rootPath, i, k);
+            ifstream annoList(annoPath.c_str());
+            if (!annoList.is_open())
+            {
+                printf("Error: Can't open annotation file *.ANN!!!\n");
+                break;
+            }
+
+            int framesNum = 0;
+
+            do
+            {
+                //Make  ALOV300++ Object
+                Ptr<TRACK_alovObj> currObj(new TRACK_alovObj);
+                string tmp;
+                framesNum++;
+
+                //Ground Truth data
+                int    n = 0;
+                double x1 = 0, y1 = 0,
+                    x2 = 0, y2 = 0,
+                    x3 = 0, y3 = 0,
+                    x4 = 0, y4 = 0;
+
+                getline(annoList, tmp);
+                std::istringstream in(tmp);
+                in >> n >> x1 >> y1 >> x2 >> y2 >> x3 >> y3 >> x4 >> y4;
+
+                currObj->gtbb.push_back(Point2d(x1, y1));
+                currObj->gtbb.push_back(Point2d(x2, y2));
+                currObj->gtbb.push_back(Point2d(x3, y3));
+                currObj->gtbb.push_back(Point2d(x4, y4));
+
+                string fullPath = fullFramePath(rootPath, i, k, n);
+                if (!fileExists(fullPath))
+                    break;
+
+                currObj->imagePath = fullPath;
+                currObj->id = n;
+
+                //Add object to storage
+                objects.push_back(currObj);
+
+            } while (annoList.good());
+
+            datasetsLengths.push_back(framesNum-1);
+            data.push_back(objects);
+        }
+    }
+
+    return;
+}
+
+int TRACK_alovImpl::getDatasetsNum()
+{
+    return (int)(data.size());
+}
+
+int TRACK_alovImpl::getDatasetLength(int id)
+{
+    if (id > 0 && id <= (int)data.size())
+        return (int)(data[id - 1].size());
+    else
+    {
+        printf("Dataset ID is out of range...\nAllowed IDs are: 1~%d\n", (int)data.size());
+        return -1;
+    }
+}
+
+bool TRACK_alovImpl::initDataset(int id)
+{
+    if (id > 0 && id <= (int)data.size())
+    {
+        activeDatasetID = id;
+        return true;
+    }
+    else
+    {
+        printf("Dataset ID is out of range...\nAllowed IDs are: 1~%d\n", (int)data.size());
+        return false;
+    }
+}
+
+bool  TRACK_alovImpl::getNextFrame(Mat &frame)
+{
+    if (frameCounter >= (int)data[activeDatasetID - 1].size())
+        return false;
+    string imgPath = data[activeDatasetID - 1][frameCounter]->imagePath;
+    frame = imread(imgPath);
+    frameCounter++;
+    return !frame.empty();
+}
+
+bool  TRACK_alovImpl::getFrame(Mat &frame, int datasetID, int frameID)
+{
+    if (frameID > (int)data[datasetID-1].size())
+        return false;
+    string imgPath = data[datasetID-1][frameID-1]->imagePath;
+    frame = imread(imgPath);
+    return !frame.empty();
+}
+
+Ptr<TRACK_alov> TRACK_alov::create()
+{
+    return Ptr<TRACK_alovImpl>(new TRACK_alovImpl);
+}
+
+vector <Point2f> TRACK_alovImpl::getNextGT()
+{
+    Ptr <TRACK_alovObj> currObj = data[activeDatasetID - 1][frameCounter - 1];
+    return currObj->gtbb;
+}
+
+vector <Point2f> TRACK_alovImpl::getGT(int datasetID, int frameID)
+{
+    Ptr <TRACK_alovObj> currObj = data[datasetID - 1][frameID - 1];
+    return currObj->gtbb;
+}
+
+}
+}

modules/tracking/CMakeLists.txt

 set(the_description "Tracking API")
-ocv_define_module(tracking opencv_imgproc opencv_core opencv_video opencv_highgui opencv_plot OPTIONAL opencv_datasets WRAP python)
+ocv_define_module(tracking opencv_imgproc opencv_core opencv_video opencv_highgui opencv_dnn opencv_plot OPTIONAL opencv_datasets WRAP python)
\ No newline at end of file

modules/tracking/doc/tracking.bib

@@ -93,3 +93,10 @@
   keywords={computer vision;feature extraction;image colour analysis;image representation;image sequences;adaptive color attributes;benchmark color sequences;color features;color representations;computer vision;image description;real-time visual tracking;tracking-by-detection framework;Color;Computational modeling;Covariance matrices;Image color analysis;Kernel;Target tracking;Visualization;Adaptive Dimensionality Reduction;Appearance Model;Color Features;Visual Tracking},
   doi={10.1109/CVPR.2014.143},
 }
+
+@inproceedings{GOTURN,
+title={Learning to Track at 100 FPS with Deep Regression Networks},
+author={Held, David and Thrun, Sebastian and Savarese, Silvio},
+booktitle = {European Conference Computer Vision (ECCV)},
+year      = {2016}
+}

modules/tracking/include/opencv2/tracking/tracker.hpp

@@ -57,7 +57,7 @@
 /*
  * Partially based on:
  * ====================================================================================================================
- * 	- [AAM] S. Salti, A. Cavallaro, L. Di Stefano, Adaptive Appearance Modeling for Video Tracking: Survey and Evaluation
+ *   - [AAM] S. Salti, A. Cavallaro, L. Di Stefano, Adaptive Appearance Modeling for Video Tracking: Survey and Evaluation
  *  - [AMVOT] X. Li, W. Hu, C. Shen, Z. Zhang, A. Dick, A. van den Hengel, A Survey of Appearance Models in Visual Object Tracking
  *
  * This Tracking API has been designed with PlantUML. If you modify this API please change UML files under modules/tracking/doc/uml
@@ -200,7 +200,7 @@ class CV_EXPORTS TrackerFeatureSet
   bool blockAddTrackerFeature;
 
   std::vector<std::pair<String, Ptr<TrackerFeature> > > features;  //list of features
-  std::vector<Mat> responses;				//list of response after compute
+  std::vector<Mat> responses;        //list of response after compute
 
 };
 
@@ -567,7 +567,7 @@ class CV_EXPORTS_W Tracker : public virtual Algorithm
 
   Ptr<TrackerModel> getModel()
   {
-	  return model;
+    return model;
   }
 
  protected:
@@ -806,7 +806,7 @@ class CV_EXPORTS TrackerSamplerCSC : public TrackerSamplerAlgorithm
     Params();
     float initInRad;        //!< radius for gathering positive instances during init
     float trackInPosRad;    //!< radius for gathering positive instances during tracking
-    float searchWinSize;	//!< size of search window
+    float searchWinSize;  //!< size of search window
     int initMaxNegNum;      //!< # negative samples to use during init
     int trackMaxPosNum;     //!< # positive samples to use during training
     int trackMaxNegNum;     //!< # negative samples to use during training
@@ -1090,12 +1090,12 @@ class CV_EXPORTS TrackerMIL : public Tracker
   {
     Params();
     //parameters for sampler
-    float samplerInitInRadius;	//!< radius for gathering positive instances during init
+    float samplerInitInRadius;  //!< radius for gathering positive instances during init
     int samplerInitMaxNegNum;  //!< # negative samples to use during init
     float samplerSearchWinSize;  //!< size of search window
     float samplerTrackInRadius;  //!< radius for gathering positive instances during tracking
-    int samplerTrackMaxPosNum;	//!< # positive samples to use during tracking
-    int samplerTrackMaxNegNum;	//!< # negative samples to use during tracking
+    int samplerTrackMaxPosNum;  //!< # positive samples to use during tracking
+    int samplerTrackMaxNegNum;  //!< # negative samples to use during tracking
     int featureSetNumFeatures;  //!< # features
 
     void read( const FileNode& fn );
@@ -1205,56 +1205,85 @@ class CV_EXPORTS TrackerTLD : public Tracker
 class CV_EXPORTS TrackerKCF : public Tracker
 {
 public:
-	/**
-	* \brief Feature type to be used in the tracking grayscale, colornames, compressed color-names
-	* The modes available now:
-	-   "GRAY" -- Use grayscale values as the feature
-	-   "CN" -- Color-names feature
-	*/
-	enum MODE {
-		GRAY = (1u << 0),
-		CN = (1u << 1),
-		CUSTOM = (1u << 2)
-	};
-
-	struct CV_EXPORTS Params
-	{
-		/**
-		* \brief Constructor
-		*/
-		Params();
-
-		/**
-		* \brief Read parameters from file, currently unused
-		*/
-		void read(const FileNode& /*fn*/);
-
-		/**
-		* \brief Read parameters from file, currently unused
-		*/
-		void write(FileStorage& /*fs*/) const;
-
-		double sigma;                 //!<  gaussian kernel bandwidth
-		double lambda;                //!<  regularization
-		double interp_factor;         //!<  linear interpolation factor for adaptation
-		double output_sigma_factor;   //!<  spatial bandwidth (proportional to target)
-		double pca_learning_rate;     //!<  compression learning rate
-		bool resize;                  //!<  activate the resize feature to improve the processing speed
-		bool split_coeff;             //!<  split the training coefficients into two matrices
-		bool wrap_kernel;             //!<  wrap around the kernel values
-		bool compress_feature;        //!<  activate the pca method to compress the features
-		int max_patch_size;           //!<  threshold for the ROI size
-		int compressed_size;          //!<  feature size after compression
-		unsigned int desc_pca;        //!<  compressed descriptors of TrackerKCF::MODE
-		unsigned int desc_npca;       //!<  non-compressed descriptors of TrackerKCF::MODE
-	};
-
-	virtual void setFeatureExtractor(void(*)(const Mat, const Rect, Mat&), bool pca_func = false);
-
-	/** @brief Constructor
-	@param parameters KCF parameters TrackerKCF::Params
-	*/
-	BOILERPLATE_CODE("KCF", TrackerKCF);
+  /**
+  * \brief Feature type to be used in the tracking grayscale, colornames, compressed color-names
+  * The modes available now:
+  -   "GRAY" -- Use grayscale values as the feature
+  -   "CN" -- Color-names feature
+  */
+  enum MODE {
+    GRAY = (1u << 0),
+    CN = (1u << 1),
+    CUSTOM = (1u << 2)
+  };
+
+  struct CV_EXPORTS Params
+  {
+    /**
+    * \brief Constructor
+    */
+    Params();
+
+    /**
+    * \brief Read parameters from file, currently unused
+    */
+    void read(const FileNode& /*fn*/);
+
+    /**
+    * \brief Read parameters from file, currently unused
+    */
+    void write(FileStorage& /*fs*/) const;
+
+    double sigma;                 //!<  gaussian kernel bandwidth
+    double lambda;                //!<  regularization
+    double interp_factor;         //!<  linear interpolation factor for adaptation
+    double output_sigma_factor;   //!<  spatial bandwidth (proportional to target)
+    double pca_learning_rate;     //!<  compression learning rate
+    bool resize;                  //!<  activate the resize feature to improve the processing speed
+    bool split_coeff;             //!<  split the training coefficients into two matrices
+    bool wrap_kernel;             //!<  wrap around the kernel values
+    bool compress_feature;        //!<  activate the pca method to compress the features
+    int max_patch_size;           //!<  threshold for the ROI size
+    int compressed_size;          //!<  feature size after compression
+    unsigned int desc_pca;        //!<  compressed descriptors of TrackerKCF::MODE
+    unsigned int desc_npca;       //!<  non-compressed descriptors of TrackerKCF::MODE
+  };
+
+  virtual void setFeatureExtractor(void(*)(const Mat, const Rect, Mat&), bool pca_func = false);
+
+  /** @brief Constructor
+  @param parameters KCF parameters TrackerKCF::Params
+  */
+  BOILERPLATE_CODE("KCF", TrackerKCF);
+};
+
+/** @brief GOTURN (@cite GOTURN) is kind of trackers based on Convolutional Neural Networks (CNN). While taking all advantages of CNN trackers,
+ *  GOTURN is much faster due to offline training without online fine-tuning nature.
+ *  GOTURN tracker addresses the problem of single target tracking: given a bounding box label of an object in the first frame of the video,
+ *  we track that object through the rest of the video. NOTE: Current method of GOTURN does not handle occlusions; however, it is fairly
+ *  robust to viewpoint changes, lighting changes, and deformations.
+ *  Inputs of GOTURN are two RGB patches representing Target and Search patches resized to 227x227.
+ *  Outputs of GOTURN are predicted bounding box coordinates, relative to Search patch coordinate system, in format X1,Y1,X2,Y2.
+ *  Original paper is here: <http://davheld.github.io/GOTURN/GOTURN.pdf>
+ *  As long as original authors implementation: <https://github.com/davheld/GOTURN#train-the-tracker>
+ *  Implementation of training algorithm is placed in separately here due to 3d-party dependencies:
+ *  <https://github.com/Auron-X/GOTURN_Training_Toolkit>
+ *  GOTURN architecture goturn.prototxt and trained model goturn.caffemodel are accessible on opencv_extra GitHub repository.
+*/
+class CV_EXPORTS TrackerGOTURN : public Tracker
+{
+public:
+  struct CV_EXPORTS Params
+  {
+    Params();
+    void read(const FileNode& /*fn*/);
+    void write(FileStorage& /*fs*/) const;
+  };
+
+  /** @brief Constructor
+  @param parameters GOTURN parameters TrackerGOTURN::Params
+  */
+  BOILERPLATE_CODE("GOTURN", TrackerGOTURN);
 };
 
 /************************************ MultiTracker Class ---By Laksono Kurnianggoro---) ************************************/
@@ -1266,103 +1295,103 @@ class CV_EXPORTS_W MultiTracker
 {
 public:
 
-	/**
-	* \brief Constructor.
-	* In the case of trackerType is given, it will be set as the default algorithm for all trackers.
-	* @param trackerType the name of the tracker algorithm to be used
-	*/
-	CV_WRAP MultiTracker(const String& trackerType = "");
-
-	/**
-	* \brief Destructor
-	*/
-	~MultiTracker();
-
-	/**
-	* \brief Add a new object to be tracked.
-	* The defaultAlgorithm will be used the newly added tracker.
-	* @param image input image
-	* @param boundingBox a rectangle represents ROI of the tracked object
-	*/
-	CV_WRAP bool add(const Mat& image, const Rect2d& boundingBox);
-
-	/**
-	* \brief Add a new object to be tracked.
-	* @param trackerType the name of the tracker algorithm to be used
-	* @param image input image
-	* @param boundingBox a rectangle represents ROI of the tracked object
-	*/
-	CV_WRAP bool add(const String& trackerType, const Mat& image, const Rect2d& boundingBox);
-
-	/**
-	* \brief Add a set of objects to be tracked.
-	* @param trackerType the name of the tracker algorithm to be used
-	* @param image input image
-	* @param boundingBox list of the tracked objects
-	*/
-	CV_WRAP bool add(const String& trackerType, const Mat& image, std::vector<Rect2d> boundingBox);
-
-	/**
-	* \brief Add a set of objects to be tracked using the defaultAlgorithm tracker.
-	* @param image input image
-	* @param boundingBox list of the tracked objects
-	*/
-	CV_WRAP bool add(const Mat& image, std::vector<Rect2d> boundingBox);
-
-	/**
-	* \brief Update the current tracking status.
-	* The result will be saved in the internal storage.
-	* @param image input image
-	*/
-	bool update(const Mat& image);
-
-	//!<  storage for the tracked objects, each object corresponds to one tracker algorithm.
-	std::vector<Rect2d> objects;
-
-	/**
-	* \brief Update the current tracking status.
-	* @param image input image
-	* @param boundingBox the tracking result, represent a list of ROIs of the tracked objects.
-	*/
-	CV_WRAP bool update(const Mat& image, CV_OUT std::vector<Rect2d> & boundingBox);
+  /**
+  * \brief Constructor.
+  * In the case of trackerType is given, it will be set as the default algorithm for all trackers.
+  * @param trackerType the name of the tracker algorithm to be used
+  */
+  CV_WRAP MultiTracker(const String& trackerType = "");
+
+  /**
+  * \brief Destructor
+  */
+  ~MultiTracker();
+
+  /**
+  * \brief Add a new object to be tracked.
+  * The defaultAlgorithm will be used the newly added tracker.
+  * @param image input image
+  * @param boundingBox a rectangle represents ROI of the tracked object
+  */
+  CV_WRAP bool add(const Mat& image, const Rect2d& boundingBox);
+
+  /**
+  * \brief Add a new object to be tracked.
+  * @param trackerType the name of the tracker algorithm to be used
+  * @param image input image
+  * @param boundingBox a rectangle represents ROI of the tracked object
+  */
+  CV_WRAP bool add(const String& trackerType, const Mat& image, const Rect2d& boundingBox);
+
+  /**
+  * \brief Add a set of objects to be tracked.
+  * @param trackerType the name of the tracker algorithm to be used
+  * @param image input image
+  * @param boundingBox list of the tracked objects
+  */
+  CV_WRAP bool add(const String& trackerType, const Mat& image, std::vector<Rect2d> boundingBox);
+
+  /**
+  * \brief Add a set of objects to be tracked using the defaultAlgorithm tracker.
+  * @param image input image
+  * @param boundingBox list of the tracked objects
+  */
+  CV_WRAP bool add(const Mat& image, std::vector<Rect2d> boundingBox);
+
+  /**
+  * \brief Update the current tracking status.
+  * The result will be saved in the internal storage.
+  * @param image input image
+  */
+  bool update(const Mat& image);
+
+  //!<  storage for the tracked objects, each object corresponds to one tracker algorithm.
+  std::vector<Rect2d> objects;
+
+  /**
+  * \brief Update the current tracking status.
+  * @param image input image
+  * @param boundingBox the tracking result, represent a list of ROIs of the tracked objects.
+  */
+  CV_WRAP bool update(const Mat& image, CV_OUT std::vector<Rect2d> & boundingBox);
 
 protected:
-	//!<  storage for the tracker algorithms.
-	std::vector< Ptr<Tracker> > trackerList;
+  //!<  storage for the tracker algorithms.
+  std::vector< Ptr<Tracker> > trackerList;
 
-	//!<  default algorithm for the tracking method.
-	String defaultAlgorithm;
+  //!<  default algorithm for the tracking method.
+  String defaultAlgorithm;
 };
 
 class ROISelector {
 public:
-	Rect2d select(Mat img, bool fromCenter = true);
-	Rect2d select(const cv::String& windowName, Mat img, bool showCrossair = true, bool fromCenter = true);
-	void select(const cv::String& windowName, Mat img, std::vector<Rect2d> & boundingBox, bool fromCenter = true);
+  Rect2d select(Mat img, bool fromCenter = true);
+  Rect2d select(const cv::String& windowName, Mat img, bool showCrossair = true, bool fromCenter = true);
+  void select(const cv::String& windowName, Mat img, std::vector<Rect2d> & boundingBox, bool fromCenter = true);
 
-	struct handlerT{
-		// basic parameters
-		bool isDrawing;
-		Rect2d box;
-		Mat image;
+  struct handlerT{
+    // basic parameters
+    bool isDrawing;
+    Rect2d box;
+    Mat image;
 
-		// parameters for drawing from the center
-		bool drawFromCenter;
-		Point2f center;
+    // parameters for drawing from the center
+    bool drawFromCenter;
+    Point2f center;
 
-		// initializer list
-		handlerT() : isDrawing(false), drawFromCenter(true) {};
-	}selectorParams;
+    // initializer list
+    handlerT() : isDrawing(false), drawFromCenter(true) {};
+  }selectorParams;
 
-	// to store the tracked objects
-	std::vector<handlerT> objects;
+  // to store the tracked objects
+  std::vector<handlerT> objects;
 
 private:
-	static void mouseHandler(int event, int x, int y, int flags, void *param);
-	void opencv_mouse_callback(int event, int x, int y, int, void *param);
+  static void mouseHandler(int event, int x, int y, int flags, void *param);
+  void opencv_mouse_callback(int event, int x, int y, int, void *param);
 
-	// save the keypressed characted
-	int key;
+  // save the keypressed characted
+  int key;
 };
 
 Rect2d CV_EXPORTS_W selectROI(Mat img, bool fromCenter = true);
@@ -1379,45 +1408,45 @@ void CV_EXPORTS_W selectROI(const cv::String& windowName, Mat img, std::vector<R
 class CV_EXPORTS MultiTracker_Alt
 {
 public:
-	/** @brief Constructor for Multitracker
-	*/
-	MultiTracker_Alt()
-	{
-		targetNum = 0;
-	}
-
-	/** @brief Add a new target to a tracking-list and initialize the tracker with a know bounding box that surrounding the target
-	@param image The initial frame
-	@param boundingBox The initial boundig box of target
-	@param tracker_algorithm_name Multi-tracker algorithm name
-
-	@return True if new target initialization went succesfully, false otherwise
-	*/
-	bool addTarget(const Mat& image, const Rect2d& boundingBox, String tracker_algorithm_name);
-
-	/** @brief Update all trackers from the tracking-list, find a new most likely bounding boxes for the targets
-	@param image The current frame
-
-	@return True means that all targets were located and false means that tracker couldn't locate one of the targets in
-	current frame. Note, that latter *does not* imply that tracker has failed, maybe target is indeed
-	missing from the frame (say, out of sight)
-	*/
-	bool update(const Mat& image);
-
-	/** @brief Current number of targets in tracking-list
-	*/
-	int targetNum;
-
-	/** @brief Trackers list for Multi-Object-Tracker
-	*/
-	std::vector <Ptr<Tracker> > trackers;
-
-	/** @brief Bounding Boxes list for Multi-Object-Tracker
-	*/
-	std::vector <Rect2d> boundingBoxes;
-	/** @brief List of randomly generated colors for bounding boxes display
-	*/
-	std::vector<Scalar> colors;
+  /** @brief Constructor for Multitracker
+  */
+  MultiTracker_Alt()
+  {
+    targetNum = 0;
+  }
+
+  /** @brief Add a new target to a tracking-list and initialize the tracker with a know bounding box that surrounding the target
+  @param image The initial frame
+  @param boundingBox The initial boundig box of target
+  @param tracker_algorithm_name Multi-tracker algorithm name
+
+  @return True if new target initialization went succesfully, false otherwise
+  */
+  bool addTarget(const Mat& image, const Rect2d& boundingBox, String tracker_algorithm_name);
+
+  /** @brief Update all trackers from the tracking-list, find a new most likely bounding boxes for the targets
+  @param image The current frame
+
+  @return True means that all targets were located and false means that tracker couldn't locate one of the targets in
+  current frame. Note, that latter *does not* imply that tracker has failed, maybe target is indeed
+  missing from the frame (say, out of sight)
+  */
+  bool update(const Mat& image);
+
+  /** @brief Current number of targets in tracking-list
+  */
+  int targetNum;
+
+  /** @brief Trackers list for Multi-Object-Tracker
+  */
+  std::vector <Ptr<Tracker> > trackers;
+
+  /** @brief Bounding Boxes list for Multi-Object-Tracker
+  */
+  std::vector <Rect2d> boundingBoxes;
+  /** @brief List of randomly generated colors for bounding boxes display
+  */
+  std::vector<Scalar> colors;
 };
 
 /** @brief Multi Object Tracker for TLD. TLD is a novel tracking framework that explicitly decomposes
@@ -1436,17 +1465,17 @@ occlusions, object absence etc.
 class CV_EXPORTS MultiTrackerTLD : public MultiTracker_Alt
 {
 public:
-	/** @brief Update all trackers from the tracking-list, find a new most likely bounding boxes for the targets by
-	optimized update method using some techniques to speedup calculations specifically for MO TLD. The only limitation
-	is that	all target bounding boxes should have approximately same aspect ratios. Speed boost is around 20%
+  /** @brief Update all trackers from the tracking-list, find a new most likely bounding boxes for the targets by
+  optimized update method using some techniques to speedup calculations specifically for MO TLD. The only limitation
+  is that all target bounding boxes should have approximately same aspect ratios. Speed boost is around 20%
 
-	@param image The current frame.
+  @param image The current frame.
 
-	@return True means that all targets were located and false means that tracker couldn't locate one of the targets in
-	current frame. Note, that latter *does not* imply that tracker has failed, maybe target is indeed
-	missing from the frame (say, out of sight)
-	*/
-	bool update_opt(const Mat& image);
+  @return True means that all targets were located and false means that tracker couldn't locate one of the targets in
+  current frame. Note, that latter *does not* imply that tracker has failed, maybe target is indeed
+  missing from the frame (say, out of sight)
+  */
+  bool update_opt(const Mat& image);
 };
 
 //! @}

modules/tracking/perf/perf_Tracker.cpp

@@ -343,3 +343,73 @@ PERF_TEST_P(tracking, tld, testing::Combine(TESTSET_NAMES, SEGMENTS))
   SANITY_CHECK( bbs_mat, 15, ERROR_RELATIVE );
 
 }
+
+PERF_TEST_P(tracking, GOTURN, testing::Combine(TESTSET_NAMES, SEGMENTS))
+{
+  string video = get<0>(GetParam());
+  int segmentId = get<1>(GetParam());
+
+  int startFrame;
+  string prefix;
+  string suffix;
+  string datasetMeta = getDataPath(TRACKING_DIR + "/" + video + "/" + video + ".yml");
+  checkData(datasetMeta, startFrame, prefix, suffix);
+  int gtStartFrame = startFrame;
+
+  vector<Rect> gtBBs;
+  string gtFile = getDataPath(TRACKING_DIR + "/" + video + "/gt.txt");
+  if (!getGroundTruth(gtFile, gtBBs))
+    FAIL() << "Ground truth file " << gtFile << " can not be read" << endl;
+  int bbCounter = (int)gtBBs.size();
+
+  Mat frame;
+  bool initialized = false;
+  vector<Rect> bbs;
+
+  Ptr<Tracker> tracker = Tracker::create("GOTURN");
+  string folder = TRACKING_DIR + "/" + video + "/" + FOLDER_IMG;
+  int numSegments = (sizeof(SEGMENTS) / sizeof(int));
+  int endFrame = 0;
+  getSegment(segmentId, numSegments, bbCounter, startFrame, endFrame);
+
+  Rect currentBBi = gtBBs[startFrame - gtStartFrame];
+  Rect2d currentBB(currentBBi);
+
+  TEST_CYCLE_N(1)
+  {
+    VideoCapture c;
+    c.open(getDataPath(TRACKING_DIR + "/" + video + "/" + FOLDER_IMG + "/" + video + ".webm"));
+    c.set(CAP_PROP_POS_FRAMES, startFrame);
+    for (int frameCounter = startFrame; frameCounter < endFrame; frameCounter++)
+    {
+      c >> frame;
+
+      if (frame.empty())
+      {
+        break;
+      }
+
+      if (!initialized)
+      {
+        if (!tracker->init(frame, currentBB))
+        {
+          FAIL() << "Could not initialize tracker" << endl;
+          return;
+        }
+        initialized = true;
+      }
+      else if (initialized)
+      {
+        tracker->update(frame, currentBB);
+      }
+      bbs.push_back(currentBB);
+
+    }
+  }
+  //save the bounding boxes in a Mat
+  Mat bbs_mat((int)bbs.size(), 4, CV_32F);
+  getMatOfRects(bbs, bbs_mat);
+
+  SANITY_CHECK(bbs_mat, 15, ERROR_RELATIVE);
+
+}

modules/tracking/samples/goturnTracker.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) 2013, OpenCV Foundation, 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 the copyright holders 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*/
+
+//Demo of GOTURN tracker
+//In order to use GOTURN tracker, GOTURN architecture goturn.prototxt and goturn.caffemodel are required to exist in root folder.
+//There are 2 ways to get caffemodel:
+//1 - Train you own GOTURN model using <https://github.com/Auron-X/GOTURN_Training_Toolkit>
+//2 - Download pretrained caffemodel from <https://github.com/opencv/opencv_extra>
+
+#include "opencv2/datasets/track_alov.hpp"
+#include <opencv2/core/utility.hpp>
+#include <opencv2/tracking.hpp>
+#include <opencv2/videoio.hpp>
+#include <opencv2/highgui.hpp>
+#include <iostream>
+
+using namespace std;
+using namespace cv;
+using namespace cv::datasets;
+
+#define NUM_TEST_FRAMES 1000
+
+static Mat image;
+static bool paused;
+static bool selectObjects = false;
+static bool startSelection = false;
+Rect2d boundingBox;
+
+static const char* keys =
+{ "{@dataset_path     |true| Dataset path     }"
+"{@dataset_id     |1| Dataset ID     }"
+};
+
+static void onMouse(int event, int x, int y, int, void*)
+{
+    if (!selectObjects)
+    {
+        switch (event)
+        {
+        case EVENT_LBUTTONDOWN:
+            //set origin of the bounding box
+            startSelection = true;
+            boundingBox.x = x;
+            boundingBox.y = y;
+            boundingBox.width = boundingBox.height = 0;
+            break;
+        case EVENT_LBUTTONUP:
+            //sei with and height of the bounding box
+            boundingBox.width = std::abs(x - boundingBox.x);
+            boundingBox.height = std::abs(y - boundingBox.y);
+            paused = false;
+            selectObjects = true;
+            startSelection = false;
+            break;
+        case EVENT_MOUSEMOVE:
+
+            if (startSelection && !selectObjects)
+            {
+                //draw the bounding box
+                Mat currentFrame;
+                image.copyTo(currentFrame);
+                rectangle(currentFrame, Point((int)boundingBox.x, (int)boundingBox.y), Point(x, y), Scalar(255, 0, 0), 2, 1);
+                imshow("GOTURN Tracking", currentFrame);
+            }
+            break;
+        }
+    }
+}
+
+static void help()
+{
+    cout << "\nThis example is a simple demo of GOTURN tracking on ALOV300++ dataset"
+        "ALOV dataset contains videos with ID range: 1~314\n"
+        "-- pause video [p] and draw a bounding boxes around the targets to start the tracker\n"
+        "Example:\n"
+        "./goturnTracker <dataset_path> <dataset_id>\n"
+        << endl;
+
+    cout << "\n\nHot keys: \n"
+        "\tq - quit the program\n"
+        "\tp - pause video\n";
+}
+
+int main(int argc, char *argv[])
+{
+    CommandLineParser parser(argc, argv, keys);
+    string datasetRootPath = parser.get<string>(0);
+    int datasetID = parser.get<int>(1);
+
+    if (datasetRootPath.empty())
+    {
+        help();
+        return -1;
+    }
+
+    Mat frame;
+    paused = false;
+    namedWindow("GOTURN Tracking", 0);
+    setMouseCallback("GOTURN Tracking", onMouse, 0);
+
+    //Create GOTURN tracker
+    Ptr<Tracker> tracker = Tracker::create("GOTURN");
+
+    //Load and init full ALOV300++ dataset with a given datasetID, as alternative you can use loadAnnotatedOnly(..)
+    //to load only frames with labled ground truth ~ every 5-th frame
+    Ptr<cv::datasets::TRACK_alov> dataset = TRACK_alov::create();
+    dataset->load(datasetRootPath);
+    dataset->initDataset(datasetID);
+
+    //Read first frame
+    dataset->getNextFrame(frame);
+    frame.copyTo(image);
+    rectangle(image, boundingBox, Scalar(255, 0, 0), 2, 1);
+    imshow("GOTURN Tracking", image);
+
+    bool initialized = false;
+    paused = true;
+    int frameCounter = 0;
+
+    //Time measurment
+    int64 e3 = getTickCount();
+
+    for (;;)
+    {
+        if (!paused)
+        {
+            //Time measurment
+            int64 e1 = getTickCount();
+            if (initialized){
+                if (!dataset->getNextFrame(frame))
+                    break;
+                frame.copyTo(image);
+            }
+
+            if (!initialized && selectObjects)
+            {
+                //Initialize the tracker and add targets
+                if (!tracker->init(frame, boundingBox))
+                {
+                    cout << "Tracker Init Error!!!";
+                    return 0;
+                }
+                rectangle(frame, boundingBox, Scalar(0, 0, 255), 2, 1);
+                initialized = true;
+            }
+            else if (initialized)
+            {
+                //Update all targets
+                if (tracker->update(frame, boundingBox))
+                {
+                    rectangle(frame, boundingBox, Scalar(0, 0, 255), 2, 1);
+                }
+            }
+            imshow("GOTURN Tracking", frame);
+            frameCounter++;
+            //Time measurment
+            int64 e2 = getTickCount();
+            double t1 = (e2 - e1) / getTickFrequency();
+            cout << frameCounter << "\tframe :  " << t1 * 1000.0 << "ms" << endl;
+        }
+
+        char c = (char)waitKey(2);
+        if (c == 'q')
+            break;
+        if (c == 'p')
+            paused = !paused;
+    }
+
+    //Time measurment
+    int64 e4 = getTickCount();
+    double t2 = (e4 - e3) / getTickFrequency();
+    cout << "Average Time for Frame:  " << t2 * 1000.0 / frameCounter << "ms" << endl;
+    cout << "Average FPS:  " << 1.0 / t2*frameCounter << endl;
+
+
+    waitKey(0);
+
+    return 0;
+}

modules/tracking/src/gtrTracker.cpp

+/*///////////////////////////////////////////////////////////////////////////////////////
+//
+//  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) 2013, OpenCV Foundation, 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 the copyright holders 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 "gtrTracker.hpp"
+
+
+namespace cv
+{
+
+TrackerGOTURN::Params::Params(){}
+
+void TrackerGOTURN::Params::read(const cv::FileNode& /*fn*/){}
+
+void TrackerGOTURN::Params::write(cv::FileStorage& /*fs*/) const {}
+
+
+Ptr<TrackerGOTURN> TrackerGOTURN::createTracker(const TrackerGOTURN::Params &parameters)
+{
+    return Ptr<gtr::TrackerGOTURNImpl>(new gtr::TrackerGOTURNImpl(parameters));
+}
+
+namespace gtr
+{
+
+class TrackerGOTURNModel : public TrackerModel{
+public:
+    TrackerGOTURNModel(TrackerGOTURN::Params){}
+    Rect2d getBoundingBox(){ return boundingBox_; }
+    void setBoudingBox(Rect2d boundingBox){ boundingBox_ = boundingBox; }
+    Mat getImage(){ return image_; }
+    void setImage(const Mat& image){ image.copyTo(image_); }
+protected:
+    Rect2d boundingBox_;
+    Mat image_;
+    void modelEstimationImpl(const std::vector<Mat>&){}
+    void modelUpdateImpl(){}
+};
+
+TrackerGOTURNImpl::TrackerGOTURNImpl(const TrackerGOTURN::Params &parameters) :
+    params(parameters){
+    isInit = false;
+};
+
+void TrackerGOTURNImpl::read(const cv::FileNode& fn)
+{
+    params.read(fn);
+}
+
+void TrackerGOTURNImpl::write(cv::FileStorage& fs) const
+{
+    params.write(fs);
+}
+
+bool TrackerGOTURNImpl::initImpl(const Mat& image, const Rect2d& boundingBox)
+{
+    //Make a simple model from frame and bounding box
+    model = Ptr<TrackerGOTURNModel>(new TrackerGOTURNModel(params));
+    ((TrackerGOTURNModel*)static_cast<TrackerModel*>(model))->setImage(image);
+    ((TrackerGOTURNModel*)static_cast<TrackerModel*>(model))->setBoudingBox(boundingBox);
+
+    //Load GOTURN architecture from *.prototxt and pretrained weights from *.caffemodel
+    String modelTxt = "goturn.prototxt";
+    String modelBin = "goturn.caffemodel";
+    Ptr<dnn::Importer> importer;
+    try                                     //Import GOTURN model
+    {
+        importer = dnn::createCaffeImporter(modelTxt, modelBin);
+    }
+    catch (const cv::Exception &err)        //Importer can throw errors, we will catch them
+    {
+        std::cerr << err.msg << std::endl;
+    }
+    if (!importer)
+    {
+        cvError(CV_StsError, "cv::gtr::InitImpl", "GOTURN network loading error...", "gtrTracker.cpp", 117);
+    }
+
+    importer->populateNet(net);
+    importer.release();                     //We don't need importer anymore
+
+    return true;
+}
+
+bool TrackerGOTURNImpl::updateImpl(const Mat& image, Rect2d& boundingBox)
+{
+    int INPUT_SIZE = 227;
+    //Using prevFrame & prevBB from model and curFrame GOTURN calculating curBB
+    Mat curFrame = image.clone();
+    Mat prevFrame = ((TrackerGOTURNModel*)static_cast<TrackerModel*>(model))->getImage();
+    Rect2d prevBB = ((TrackerGOTURNModel*)static_cast<TrackerModel*>(model))->getBoundingBox();
+    Rect2d curBB;
+
+    float padTargetPatch = 2.0;
+    Rect2f searchPatchRect, targetPatchRect;
+    Point2f currCenter, prevCenter;
+    Mat prevFramePadded, curFramePadded;
+    Mat searchPatch, targetPatch;
+
+    prevCenter.x = (float)(prevBB.x + prevBB.width / 2);
+    prevCenter.y = (float)(prevBB.y + prevBB.height / 2);
+
+    targetPatchRect.width = (float)(prevBB.width*padTargetPatch);
+    targetPatchRect.height = (float)(prevBB.height*padTargetPatch);
+    targetPatchRect.x = (float)(prevCenter.x - prevBB.width*padTargetPatch / 2.0 + targetPatchRect.width);
+    targetPatchRect.y = (float)(prevCenter.y - prevBB.height*padTargetPatch / 2.0 + targetPatchRect.height);
+
+    copyMakeBorder(prevFrame, prevFramePadded, (int)targetPatchRect.height, (int)targetPatchRect.height, (int)targetPatchRect.width, (int)targetPatchRect.width, BORDER_REPLICATE);
+    targetPatch = prevFramePadded(targetPatchRect).clone();
+
+    copyMakeBorder(curFrame, curFramePadded, (int)targetPatchRect.height, (int)targetPatchRect.height, (int)targetPatchRect.width, (int)targetPatchRect.width, BORDER_REPLICATE);
+    searchPatch = curFramePadded(targetPatchRect).clone();
+
+    //Preprocess
+    //Resize
+    resize(targetPatch, targetPatch, Size(INPUT_SIZE, INPUT_SIZE));
+    resize(searchPatch, searchPatch, Size(INPUT_SIZE, INPUT_SIZE));
+
+    //Mean Subtract
+    targetPatch = targetPatch - 128;
+    searchPatch = searchPatch - 128;
+
+    //Convert to Float type
+    targetPatch.convertTo(targetPatch, CV_32F);
+    searchPatch.convertTo(searchPatch, CV_32F);
+
+    dnn::Blob targetBlob = dnn::Blob(targetPatch);
+    dnn::Blob searchBlob = dnn::Blob(searchPatch);
+
+    net.setBlob(".data1", targetBlob);
+    net.setBlob(".data2", searchBlob);
+
+    net.forward();
+    dnn::Blob res = net.getBlob("scale");
+
+    Mat resMat = res.matRefConst().reshape(1, 1);
+
+    curBB.x = targetPatchRect.x + (resMat.at<float>(0) * targetPatchRect.width / INPUT_SIZE) - targetPatchRect.width;
+    curBB.y = targetPatchRect.y + (resMat.at<float>(1) * targetPatchRect.height / INPUT_SIZE) - targetPatchRect.height;
+    curBB.width = (resMat.at<float>(2) - resMat.at<float>(0)) * targetPatchRect.width / INPUT_SIZE;
+    curBB.height = (resMat.at<float>(3) - resMat.at<float>(1)) * targetPatchRect.height / INPUT_SIZE;
+
+    //Predicted BB
+    boundingBox = curBB;
+
+    //Set new model image and BB from current frame
+    ((TrackerGOTURNModel*)static_cast<TrackerModel*>(model))->setImage(curFrame);
+    ((TrackerGOTURNModel*)static_cast<TrackerModel*>(model))->setBoudingBox(curBB);
+    return true;
+}
+
+}
+
+}

modules/tracking/src/gtrTracker.hpp

+
+/*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) 2013, OpenCV Foundation, 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 the copyright holders 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*/
+
+#ifndef OPENCV_GOTURN_TRACKER
+#define OPENCV_GOTURN_TRACKER
+
+#include "precomp.hpp"
+#include "opencv2/video/tracking.hpp"
+#include "opencv2/dnn.hpp"
+#include "gtrUtils.hpp"
+#include "opencv2/imgproc.hpp"
+#include<algorithm>
+#include<limits.h>
+
+namespace cv
+{
+namespace gtr
+{
+
+class TrackerGOTURNImpl : public TrackerGOTURN
+{
+public:
+    TrackerGOTURNImpl(const TrackerGOTURN::Params &parameters = TrackerGOTURN::Params());
+    void read(const FileNode& fn);
+    void write(FileStorage& fs) const;
+    bool initImpl(const Mat& image, const Rect2d& boundingBox);
+    bool updateImpl(const Mat& image, Rect2d& boundingBox);
+
+    TrackerGOTURN::Params params;
+
+    dnn::Net net;
+};
+
+}
+}
+
+#endif

modules/tracking/src/gtrUtils.cpp

+/*///////////////////////////////////////////////////////////////////////////////////////
+ //
+ //  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) 2013, OpenCV Foundation, 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 the copyright holders 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 "gtrUtils.hpp"
+
+
+namespace cv
+{
+namespace gtr
+{
+
+double generateRandomLaplacian(double b, double m)
+{
+    double t = (double)rand() / (RAND_MAX);
+    double n = (double)rand() / (RAND_MAX);
+
+    if (t > 0.5)
+        return m + b*log(n);
+    else
+        return m - b*log(n);
+}
+
+Rect2f anno2rect(vector<Point2f> annoBB)
+{
+    Rect2f rectBB;
+    rectBB.x = min(annoBB[0].x, annoBB[1].x);
+    rectBB.y = min(annoBB[0].y, annoBB[2].y);
+    rectBB.width = fabs(annoBB[0].x - annoBB[1].x);
+    rectBB.height = fabs(annoBB[0].y - annoBB[2].y);
+
+    return rectBB;
+}
+
+vector <TrainingSample> gatherFrameSamples(Mat prevFrame, Mat currFrame, Rect2f prevBB, Rect2f currBB)
+{
+    vector <TrainingSample> trainingSamples;
+    Point2f currCenter, prevCenter;
+    Rect2f targetPatchRect, searchPatchRect;
+    Mat targetPatch, searchPatch;
+    Mat prevFramePadded, currFramePadded;
+
+    //Crop Target Patch
+
+    //Padding
+
+    //Previous frame GTBBs center
+    prevCenter.x = prevBB.x + prevBB.width / 2;
+    prevCenter.y = prevBB.y + prevBB.height / 2;
+
+    targetPatchRect.width = (float)(prevBB.width*padTarget);
+    targetPatchRect.height = (float)(prevBB.height*padTarget);
+    targetPatchRect.x = (float)(prevCenter.x - prevBB.width*padTarget / 2.0 + targetPatchRect.width);
+    targetPatchRect.y = (float)(prevCenter.y - prevBB.height*padTarget / 2.0 + targetPatchRect.height);
+
+    copyMakeBorder(prevFrame, prevFramePadded, (int)targetPatchRect.height, (int)targetPatchRect.height, (int)targetPatchRect.width, (int)targetPatchRect.width, BORDER_REPLICATE);
+
+    targetPatch = prevFramePadded(targetPatchRect);
+
+
+    for (int i = 0; i < samplesInFrame; i++)
+    {
+        TrainingSample sample;
+
+        //Current frame GTBBs center
+        currCenter.x = (float)(currBB.x + currBB.width / 2.0);
+        currCenter.y = (float)(currBB.y + currBB.height / 2.0);
+
+        //Generate and add random Laplacian distribution (Scaling from target size)
+        double dx, dy, ds;
+        dx = generateRandomLaplacian(bX, 0)*prevBB.width;
+        dy = generateRandomLaplacian(bY, 0)*prevBB.height;
+        ds = generateRandomLaplacian(bS, 1);
+
+        //Limit coefficients
+        dx = min(dx, (double)prevBB.width);
+        dx = max(dx, (double)-prevBB.width);
+        dy = min(dy, (double)prevBB.height);
+        dy = max(dy, (double)-prevBB.height);
+        ds = min(ds, Ymax);
+        ds = max(ds, Ymin);
+
+        searchPatchRect.width = (float)(prevBB.width*padSearch*ds);
+        searchPatchRect.height =(float)(prevBB.height*padSearch*ds);
+        searchPatchRect.x = (float)(currCenter.x + dx - searchPatchRect.width / 2.0 + searchPatchRect.width);
+        searchPatchRect.y = (float)(currCenter.y + dy - searchPatchRect.height / 2.0 + searchPatchRect.height);
+        copyMakeBorder(currFrame, currFramePadded, (int)searchPatchRect.height, (int)searchPatchRect.height, (int)searchPatchRect.width, (int)searchPatchRect.width, BORDER_REPLICATE);
+        searchPatch = currFramePadded(searchPatchRect);
+
+        //Calculate Relative GTBB in search patch
+        Rect2f relGTBB;
+        relGTBB.width = currBB.width;
+        relGTBB.height = currBB.height;
+        relGTBB.x = currBB.x - searchPatchRect.x + searchPatchRect.width;
+        relGTBB.y = currBB.y - searchPatchRect.y + searchPatchRect.height;
+
+        //Link to the sample struct
+        sample.targetPatch = targetPatch.clone();
+        sample.searchPatch = searchPatch.clone();
+        sample.targetBB = relGTBB;
+
+        trainingSamples.push_back(sample);
+    }
+
+    return trainingSamples;
+}
+
+}
+}

modules/tracking/src/gtrUtils.hpp

+#ifndef OPENCV_GTR_UTILS
+#define OPENCV_GTR_UTILS
+
+#include "precomp.hpp"
+#include <vector>
+#include "opencv2/highgui.hpp"
+#include <opencv2/datasets/track_alov.hpp>
+
+namespace cv
+{
+namespace gtr
+{
+
+//Number of samples in batch
+const int samplesInBatch = 50;
+
+//Number of samples to mine from video frame
+const int samplesInFrame = 10;
+
+//Number of samples to mine from still image
+const int samplesInImage = 10;
+
+//Padding coefficients for Target/Search Region
+const double padTarget = 2.0;
+const double padSearch = 2.0;
+
+//Scale parameters for Laplace distribution for Translation/Scale
+const double bX = 1.0/10;
+const double bY = 1.0/10;
+const double bS = 1.0/15;
+
+//Limits of scale changes
+const double Ymax = 1.4;
+const double Ymin = 0.6;
+
+//Lower boundary constraints for random samples (sample should include X% of target BB)
+const double minX = 0.5;
+const double minY = 0.5;
+
+//Structure of sample for training
+struct TrainingSample
+{
+    Mat targetPatch;
+    Mat searchPatch;
+    //Output bounding box on search patch
+    Rect2f targetBB;
+};
+
+//Laplacian distribution
+double generateRandomLaplacian(double b, double m);
+
+//Convert ALOV300++ anno coordinates to Rectangle BB
+Rect2f anno2rect(vector<Point2f> annoBB);
+
+//Gather samples from random video frame
+vector <TrainingSample> gatherFrameSamples(Mat prevFrame, Mat currFrame, Rect2f prevBB, Rect2f currBB);
+
+}
+}
+
+#endif

modules/tracking/src/tracker.cpp

@@ -111,6 +111,7 @@ Ptr<Tracker> Tracker::create( const String& trackerType )
   BOILERPLATE_CODE("MEDIANFLOW",TrackerMedianFlow);
   BOILERPLATE_CODE("TLD",TrackerTLD);
   BOILERPLATE_CODE("KCF",TrackerKCF);
+  BOILERPLATE_CODE("GOTURN", TrackerGOTURN);
   return Ptr<Tracker>();
 }
 

modules/tracking/test/test_trackerOPE.cpp

@@ -422,6 +422,20 @@ TEST_P(OPE_Overlap, TLD)
   RecordProperty( "ratioSuccess", test.getRatioSucc() );
 }
 
+TEST_P(OPE_Distance, GOTURN)
+{
+  TrackerOPETest test(Tracker::create("GOTURN"), TrackerOPETest::DISTANCE, dataset, threshold);
+  test.run();
+  RecordProperty("ratioSuccess", test.getRatioSucc());
+}
+
+TEST_P(OPE_Overlap, GOTURN)
+{
+  TrackerOPETest test(Tracker::create("GOTURN"), TrackerOPETest::OVERLAP, dataset, threshold);
+  test.run();
+  RecordProperty("ratioSuccess", test.getRatioSucc());
+}
+
 INSTANTIATE_TEST_CASE_P( Tracking, OPE_Distance, testing::Combine( TESTSET_NAMES, LOCATION_ERROR_THRESHOLD ) );
 
 INSTANTIATE_TEST_CASE_P( Tracking, OPE_Overlap, testing::Combine( TESTSET_NAMES, OVERLAP_THRESHOLD ) );

modules/tracking/test/test_trackerSRE.cpp

@@ -529,6 +529,20 @@ TEST_P(SRE_Overlap, TLD)
   RecordProperty( "ratioSuccess", test.getRatioSucc() );
 }
 
+TEST_P(SRE_Distance, GOTURN)
+{
+  TrackerSRETest test(Tracker::create("GOTURN"), TrackerSRETest::DISTANCE, dataset, shift, threshold);
+  test.run();
+  RecordProperty("ratioSuccess", test.getRatioSucc());
+}
+
+TEST_P(SRE_Overlap, GOTURN)
+{
+  TrackerSRETest test(Tracker::create("GOTURN"), TrackerSRETest::OVERLAP, dataset, shift, threshold);
+  test.run();
+  RecordProperty("ratioSuccess", test.getRatioSucc());
+}
+
 INSTANTIATE_TEST_CASE_P( Tracking, SRE_Distance, testing::Combine( TESTSET_NAMES, SPATIAL_SHIFTS, LOCATION_ERROR_THRESHOLD ) );
 
 INSTANTIATE_TEST_CASE_P( Tracking, SRE_Overlap, testing::Combine( TESTSET_NAMES, SPATIAL_SHIFTS, OVERLAP_THRESHOLD ) );

modules/tracking/test/test_trackerTRE.cpp

@@ -389,7 +389,7 @@ void TrackerTRETest::checkDataTest()
   gt2.close();
 
   if( segmentIdx == ( sizeof ( SEGMENTS)/sizeof(int) ) )
-	endFrame = (int)bbs.size();
+  endFrame = (int)bbs.size();
 }
 
 void TrackerTRETest::run()
@@ -499,6 +499,20 @@ TEST_P(TRE_Overlap, TLD)
   RecordProperty( "ratioSuccess", test.getRatioSucc() );
 }
 
+TEST_P(TRE_Distance, GOTURN)
+{
+  TrackerTRETest test(Tracker::create("GOTURN"), TrackerTRETest::DISTANCE, dataset, threshold, segment);
+  test.run();
+  RecordProperty("ratioSuccess", test.getRatioSucc());
+}
+
+TEST_P(TRE_Overlap, GOTURN)
+{
+  TrackerTRETest test(Tracker::create("GOTURN"), TrackerTRETest::OVERLAP, dataset, threshold, segment);
+  test.run();
+  RecordProperty("ratioSuccess", test.getRatioSucc());
+}
+
 INSTANTIATE_TEST_CASE_P( Tracking, TRE_Distance, testing::Combine( TESTSET_NAMES, SEGMENTS, LOCATION_ERROR_THRESHOLD ) );
 
 INSTANTIATE_TEST_CASE_P( Tracking, TRE_Overlap, testing::Combine( TESTSET_NAMES, SEGMENTS, OVERLAP_THRESHOLD ) );