Commands cheatsheet

CMakeLists.txt

@@ -6,40 +6,24 @@ find_package(catkin REQUIRED COMPONENTS
   std_msgs
   image_transport
   cv_bridge
-  message_generation
 )
 find_package(OpenCV)
-find_package( PkgConfig REQUIRED )
-pkg_check_modules ( ncurses++ REQUIRED ncurses++ )
 
-add_message_files(
-  FILES
-  BoundingBox.msg
-)
-
-generate_messages(
-    DEPENDENCIES
-    std_msgs
-)
 catkin_package(CATKIN_DEPENDS
   roscpp
   std_msgs
   image_transport
   cv_bridge
-  message_runtime
 )
 
-
 include_directories  (${catkin_INCLUDE_DIRS})
 
 add_executable       (papillon src/papillon.cpp)
-add_executable       (control src/control.cpp)
-add_dependencies( papillon ${PROJECT_NAME}_generate_messages_cpp )
 target_link_libraries(papillon ${catkin_LIBRARIES})
-target_link_libraries(control ${catkin_LIBRARIES} ncurses)
 set_property         (TARGET papillon APPEND PROPERTY INCLUDE_DIRECTORIES ${OpenCV_INCLUDE_DIRS})
 set_property         (TARGET papillon APPEND PROPERTY INCLUDE_DIRECTORIES ${catkin_INCLUDE_DIRS})
 set_property         (TARGET papillon APPEND PROPERTY LINK_LIBRARIES      ${OpenCV_LIBRARIES})
+set( CMAKE_EXPORT_COMPILE_COMMANDS 1 )
 
 
 install(TARGETS papillon DESTINATION ${CATKIN_PACKAGE_BIN_DESTINATION})

README.md

@@ -1,22 +1,3 @@
 # Papillon
 
-Comportements émergents avec les drônes bebop
-
-## Utilisation
-
-Il faut en premier lieu se connecter en WiFi au drone.
-
-Il faut ensuite bien entendu lancer `roscore`.
-
-Une fois ce dernier lancé, on peut lancer le driver du drone afin qu'il soit connecté à ROS à l'aide de la commande :
-```bash
-  $ roslaunch bebop_driver bebop_node.launch
-```
-Il faut ensuite lancer les deux noeuds de `papillon` à l'aide des commandes :
-```bash
-  $ rosrun papillon papillon
-  $ rosrun papillon control
-```
-Enfin, sur le noeud `control`, lors de l'appui sur la touche **F** le drone décolle et commence à suivre les mouvements qu'il détecte.
-
-Lors d'un appui sur la touche **Espace**, le drone arrête de suivre et se pose.
+## Comportements émergents avec les drônes bebop

commandes

@@ -1,6 +0,0 @@
-rosrun demo_teleop key_teleop.py --persist /cmd_vel:=/bebop/cmd_vel
-rostopic pub --once /bebop/land std_msgs/Empty
-rostopic pub --once /bebop/takeoff std_msgs/Empty
-catkin_make --pkg=papillon
-catkin_make -DCMAKE_BUILD_TYPE=Debug
-roslaunch bebop_driver bebop_node.launch

commands

@@ -0,0 +1 @@
+catkin_make -DCMAKE_BUILD_TYPE=Debug

msg/BoundingBox.msg

@@ -1,4 +0,0 @@
-float64 x
-float64 y
-float64 width
-float64 height

package.xml

@@ -45,13 +45,10 @@
   <build_depend>std_msgs</build_depend>
   <build_depend>image_transport</build_depend>
   <build_depend>cv_bridge</build_depend>
-  <build_depend>message_generation</build_depend>
-  <build_depend>ncurses++</build_depend>
   <run_depend>roscpp</run_depend>
   <run_depend>std_msgs</run_depend>
   <run_depend>image_transport</run_depend>
   <run_depend>cv_bridge</run_depend>
-  <run_depend>message_runtime</run_depend>
 
 
 

src/control.cpp

@@ -1,137 +0,0 @@
-#include "ros/ros.h"
-#include <papillon/BoundingBox.h>
-#include <geometry_msgs/Twist.h>
-#include <std_msgs/Empty.h>
-#include <ncurses.h>
-
-#include <opencv/cv.h>
-
-#include <sstream>
-
-using namespace std;
-
-class Drone_control {
-	public:
-		constexpr static float MAX_ANG_VEL = 0.3;
-
-		ros::NodeHandle n;
-		ros::Publisher pub_cmd;
-		ros::Publisher pub_takeoff;
-		ros::Publisher pub_land;
-		ros::Subscriber sub_box;
-		std_msgs::Empty nope;
-
-		float THRES_TURN = 0.1;
-		bool emergency = false;
-
-		// 1 left | 0 center | -1 right
-		vector<int> decision_acc;
-		const static int NB_ACC = 5;
-
-
-		Drone_control() : n("~") {
-			pub_cmd = n.advertise<geometry_msgs::Twist>("/bebop/cmd_vel", 1);
-			pub_takeoff = n.advertise<std_msgs::Empty>("/bebop/takeoff", 1);
-			pub_land = n.advertise<std_msgs::Empty>("/bebop/land", 1);
-			sub_box = n.subscribe<papillon::BoundingBox>("/papillon/bbox", 1, &Drone_control::on_msg, this);
-
-			// Initialise decision accumulator
-			for (size_t i = 0; i < NB_ACC; ++i)
-				decision_acc.push_back(0);
-
-			ROS_INFO("Built !");
-		}
-
-
-		// This processes an image and publishes the result.
-		void on_msg(const papillon::BoundingBox::ConstPtr& bbox) {
-			geometry_msgs::Twist twist = geometry_msgs::Twist();
-			cv::Point2f r_center = cv::Point2f(bbox->x + bbox->width/2, bbox->y + bbox->height/2);
-			if (!emergency) {
-				decision_acc.pop_back();
-				// Image is 1.0 in width and height
-				if (r_center.x < 0.5 - THRES_TURN)
-					decision_acc.insert(decision_acc.begin(), 1);
-				else if (r_center.x > 0.5 + THRES_TURN)
-					decision_acc.insert(decision_acc.begin(), -1);
-				else
-					decision_acc.insert(decision_acc.begin(), 0);
-
-				int moy = 0;
-				for (const auto& d : decision_acc)
-					moy += d;
-				twist.angular.z = moy / (float)NB_ACC * MAX_ANG_VEL;
-			}
-
-			pub_cmd.publish(twist);
-		}
-
-		void key_spin() {
-			// >>>>> nCurses initization
-			initscr();
-			keypad(stdscr, TRUE);
-			cbreak();
-			noecho();
-			timeout(0); // Non blocking getch()
-			// <<<<< nCurses initization
-
-			int c;
-
-			printw(" ------------------------\n");
-			printw("|      Papillon HQ       |\n");
-			printw(" ------------------------ \n");
-			printw("| Press F to take off.   |\n");
-			printw("| Press SPACEBAR to land.|\n");
-			printw("| Press Q to exit.       |\n");
-			printw(" ------------------------\n");
-
-			bool stop = false;
-
-			ros::Rate rate(1.0/30.0);  // 30 Hz
-			while(!stop)
-			{
-				c = getch();
-
-				switch(c)
-				{
-				case 'f':
-				{
-					// Takeoff
-					ROS_DEBUG_STREAM("Taking off...\r");
-					pub_takeoff.publish(nope);
-					emergency = false;
-					printw("Taking off...\r");
-					break;
-				}
-				case ' ':
-				{
-					// Takeoff
-					ROS_DEBUG_STREAM("Landing...\r");
-					pub_land.publish(nope);
-					printw("Landing...\r");
-					emergency = true;
-					break;
-				}
-				case 'q':
-				case 'Q':
-				{
-					ROS_DEBUG_STREAM("EXIT\r");
-					stop = true;
-				}
-				}
-				ros::spinOnce();
-			}
-			printw("\nExiting\n");
-		}
-};
-
-
-int main(int argc, char **argv)
-{
-	ros::init(argc, argv, "control");
-	Drone_control con=Drone_control();
-	con.key_spin();
-
-	return 0;
-}
-

src/papillon.cpp

@@ -2,207 +2,233 @@
 #include <image_transport/image_transport.h>
 #include <cv_bridge/cv_bridge.h>
 #include <sensor_msgs/image_encodings.h>
-#include <papillon/BoundingBox.h>
+#include <geometry_msgs/Twist.h>
+#include <typeinfo>
 
 #include <opencv/cv.h>
 
 #include <sstream>
 
+using namespace cv;
 using namespace std;
 
 class Traite_image {
 	public:
-		const static int SENSITIVITY_VALUE = 20;
-		const static int BLUR_SIZE = 9;
-		const static int CLOSE_SIZE = 30;
-		const static int ERODE_SIZE = 2;
-		const static int NB_FRAME_DROP = 1;
+		const static int THRESHOLD_DETECT_SENSITIVITY = 10;
+		const static int BLUR_SIZE = 5;
+		const static int THRESHOLD_MOV = 5;
+		constexpr static float MOVEMENT_THRES = 0.05;
+
+		constexpr static float FLOW_MIN_QUAL = 0.01;
+		const static int FLOW_MIN_DIST = 20;
 
 
-		vector<cv::Mat> prevs;
-		cv::Mat last_T;
+		Mat prev;
+
+		// Stabilisation transformation matrices
+		Mat T, last_T;
+
 		bool first = true;
+
+		// Features vectors
+		vector <Point2f> prev_ftr, cur_ftr;
+
+		// Downsize factor
 		int resize_f = 1;
 
+		int theObject[2] = {0,0};
+		Rect objectBoundingRectangle = Rect(0,0,0,0);
+
 		ros::NodeHandle n;
 
 		image_transport::ImageTransport it;
-		image_transport::Subscriber     sub;
 		image_transport::Publisher      pub_img;
-		image_transport::Publisher      pub_thres;
         ros::Publisher                pub_cmd;
+		image_transport::Subscriber     sub;
 
 
 		Traite_image() : n("~"),it(n) {
-			pub_img = it.advertise("/papillon/image_out", 1);
-			pub_thres = it.advertise("/papillon/thres_out", 1);
-			pub_cmd = n.advertise<papillon::BoundingBox>("/papillon/bbox", 1);
-			sub = it.subscribe("/bebop/image_raw", 1, [this](const sensor_msgs::ImageConstPtr& img) -> void { this->on_image(img);},ros::VoidPtr(),image_transport::TransportHints("compressed"));
+			pub_img = it.advertise("/image_out", 1);
+			pub_cmd = n.advertise<geometry_msgs::Twist>("/vrep/drone/cmd_vel", 1);
+			sub = it.subscribe("/usb_cam/image_raw", 1, [this](const sensor_msgs::ImageConstPtr& img) -> void { this->on_image(img);},ros::VoidPtr(),image_transport::TransportHints("compressed"));
 		}
 
 
-		// Callback function executedwhen recieving an image from the camera
-		// Publishes to /papillon/image_out
+		// This processes an image and publishes the result.
 		void on_image(const sensor_msgs::ImageConstPtr& msg) {
 
 			cv_bridge::CvImageConstPtr bridge_input;
 			try {
 				bridge_input = cv_bridge::toCvShare(msg,sensor_msgs::image_encodings::RGB8);
 			}
-			catch (cv::Exception& e) {
+			catch (Exception& e) {
 				std::ostringstream errstr;
 				errstr << "cv_bridge exception caught: " << e.what();
 				return;
 			}
 
-			const cv::Mat& input = bridge_input->image;
-			cv::Mat next;
-			resize(input, next, cv::Size(input.size().width/resize_f, input.size().height/resize_f));
-			cv::Mat output;
-			next.copyTo(output);
-			
-			cv::cvtColor(next, next, cv::COLOR_BGR2GRAY);
-			
+			//Mat& input  = const_cast<Mat&>(bridge_input->image);
+			const Mat& input = bridge_input->image;
+			Mat next;
+			resize(input, next, Size(input.size().width/resize_f, input.size().height/resize_f));
+			Mat output;// = input.clone(); //   (input.rows, input.cols, CV_32FC2);
+			//ROS_INFO("got input");
 			if (first) {
-				for (int i = 0; i < NB_FRAME_DROP; ++i) {
-					prevs.push_back(next.clone());
-				}
+				prev = next.clone();
 				first = false;
-				ROS_INFO("Ready");
+				ROS_INFO("first done");
 			}
 
-			cv::Mat next_stab;
-			stabiliseImg(prevs.back(), next, next_stab);
-			int crop_ratio = 6;
-			float crop_x = next_stab.size().width/crop_ratio;
-			float crop_y = next_stab.size().height/crop_ratio;
-			float crop_w = next_stab.size().width*(1-2.0/crop_ratio);
-			float crop_h = next_stab.size().height*(1-2.0/crop_ratio);
-			cv::Rect myROI(crop_x, crop_y, crop_w, crop_h);
-			cv::Mat next_stab_cropped = next_stab(myROI);
-			cv::Mat prev_cropped = prevs.back()(myROI);
-			cv::Mat closed_thres;
-			searchForMovement(prev_cropped, next_stab_cropped, output, closed_thres);
+			Mat next_stab;
+			trackFeatures(prev, next);
+			stabiliseImg(next, next_stab);
+			trackingOptFlow(prev, next_stab, next_stab);
+			Mat next_stab2;
+			trackFeatures(prev, next);
+			stabiliseImg(next, next_stab2);
+			trackingOptFlow(prev, next_stab2, output);
+			//searchForMovementOptFlow(prev_cropped, next_stab_cropped, output);
 
 
 			pub_img.publish(cv_bridge::CvImage(msg->header, "rgb8", output).toImageMsg());
-			pub_thres.publish(cv_bridge::CvImage(msg->header, "mono8", closed_thres).toImageMsg());
+			// bridge_input is handled by a smart-pointer. No explicit delete needed.
 
-			prevs.pop_back();
-			prevs.insert(prevs.begin(), next.clone());
+            droneTracking(Rect(Point(0,0), output.size()));
+
+			//ROS_INFO("pub");
+
+			prev = next.clone();
 		}
 
-		void stabiliseImg(cv::Mat prev_grey, cv::Mat cur_grey, cv::Mat &output){
+
+		void trackFeatures(Mat prev, Mat cur) {
+			Mat cur_grey, prev_grey;
+			cvtColor(cur, cur_grey, COLOR_BGR2GRAY);
+			cvtColor(prev, prev_grey, COLOR_BGR2GRAY);
+
 			// vector from prev to cur
-			vector <cv::Point2f> prev_corner, cur_corner;
-			vector <cv::Point2f> prev_corner2, cur_corner2;
+			vector <Point2f> prev_corner, cur_corner;
 			vector <uchar> status;
 			vector <float> err;
 
-			cv::goodFeaturesToTrack(prev_grey, prev_corner, 200, 0.01, 30);
-			cv::calcOpticalFlowPyrLK(prev_grey, cur_grey, prev_corner, cur_corner, status, err);
+			goodFeaturesToTrack(prev_grey, prev_corner, 200, FLOW_MIN_QUAL, FLOW_MIN_DIST);
+			calcOpticalFlowPyrLK(prev_grey, cur_grey, prev_corner, cur_corner, status, err);
 
 			// weed out bad matches
+			prev_ftr.resize(0);
+			cur_ftr.resize(0);
 			for(size_t i=0; i < status.size(); i++) {
 				if(status[i]) {
-					prev_corner2.push_back(prev_corner[i]);
-					cur_corner2.push_back(cur_corner[i]);
+					prev_ftr.push_back(prev_corner[i]);
+					cur_ftr.push_back(cur_corner[i]);
+				}
 			}
 		}
 
-			cv::Mat T = estimateRigidTransform(prev_corner2, cur_corner2, true); // false = rigid transform, no scaling/shearing
 
-			if(T.data == NULL) {
+		void stabiliseImg(Mat cur, Mat &output){
+			T = estimateRigidTransform(prev_ftr, cur_ftr, true); // false = rigid transform, no scaling/shearing
+
+			if(T.data == NULL)
 				last_T.copyTo(T);
-			}
+			else
 				T.copyTo(last_T);
 
-			cv::warpAffine(cur_grey, output, T, cur_grey.size(),cv::INTER_CUBIC|cv::WARP_INVERSE_MAP);
+			Mat cur2;
+
+			warpAffine(cur, cur2, T, cur.size(),INTER_NEAREST|WARP_INVERSE_MAP);
+
+			cur2.copyTo(output);
 		}
 
-		void searchForMovement(cv::Mat prev_grey, cv::Mat cur_grey, cv::Mat &output, cv::Mat &out2){
-			cv::GaussianBlur(prev_grey, prev_grey, cv::Size(BLUR_SIZE,BLUR_SIZE), 3.0);
-			cv::GaussianBlur(cur_grey, cur_grey, cv::Size(BLUR_SIZE,BLUR_SIZE), 3.0);
 
-			// Subtract the 2 last frames and threshold them
-			cv::Mat thres;
-			cv::absdiff(prev_grey,cur_grey,thres);
+		void warpPoints(vector<Point2f> p, vector<Point2f> &p_warp, Mat T, bool invert=false) {
+			Mat H;
+			if(invert)
+				invertAffineTransform(T, H);
 
-			cv::Mat element = getStructuringElement( cv::MORPH_ELLIPSE,
-                                   cv::Size( 2*ERODE_SIZE + 1, 2*ERODE_SIZE+1 ),
-                                   cv::Point( ERODE_SIZE, ERODE_SIZE ) );
-			// Apply the erode operation
-			cv::erode(thres, thres, element );
+			p_warp.clear();
+			for(size_t i=0; i < p.size(); ++i) {
+				Mat src(3/*rows*/,1 /* cols */,CV_64F);
 
-			cv::threshold(thres, thres, SENSITIVITY_VALUE, 255, cv::THRESH_BINARY);
+				src.at<double>(0,0)=p[i].x;
+				src.at<double>(1,0)=p[i].y;
+				src.at<double>(2,0)=1.0;
 
-			// Intermediate output
-			thres.copyTo(out2);
-
-			cv::Mat closed_thres;
-			cv::Mat structuringElement = getStructuringElement(cv::MORPH_ELLIPSE, cv::Size(CLOSE_SIZE, CLOSE_SIZE));
-			cv::morphologyEx( thres, closed_thres, cv::MORPH_CLOSE, structuringElement );
-
-			bool objectDetected = false;
-			cv::Mat temp;
-			closed_thres.copyTo(temp);
-			vector< vector<cv::Point> > contours;
-			//find contours of filtered image using openCV findContours function
-			cv::findContours(temp,contours,CV_RETR_EXTERNAL,CV_CHAIN_APPROX_SIMPLE );// retrieves external contours
-
-			//if contours vector is not empty, we have found some objects
-			if(contours.size()>0){
-				vector<cv::Rect> nc_rects; // Non connected rectangles
-				for(size_t i=0; i<contours.size();i++)
-				{
-					nc_rects.push_back(cv::boundingRect(contours[i]));
-				}
-
-				vector<cv::Rect> c_rects; // Connected rectangles
-				cleanBBoxes(nc_rects, cur_grey.size(), c_rects);
-				if (c_rects.size() > 0) {
-					for (const auto& rect : c_rects)
-						cv::rectangle(output, rect, cv::Scalar(0, 255, 0), 2);
-
-					cv::Rect objBRect = c_rects.front();
-					papillon::BoundingBox bbox = papillon::BoundingBox();
-					bbox.x = objBRect.x / (float)cur_grey.size().width;
-					bbox.y = objBRect.y / (float)cur_grey.size().height;
-					bbox.width = objBRect.width / (float)cur_grey.size().width;
-					bbox.height = objBRect.height / (float)cur_grey.size().height;
-					pub_cmd.publish(bbox);
+				Mat dst = H*src; //USE MATRIX ALGEBRA
+				p_warp.push_back(Point2f(dst.at<double>(0,0),dst.at<double>(1,0)));
 			}
 		}
+
+
+		void trackingOptFlow(Mat prev, Mat cur, Mat &output) {
+			cur.copyTo(output);
+			vector <Point2f> cur_ftr_stab;
+
+			//T = estimateRigidTransform(prev_ftr, cur_ftr, true); // false = rigid transform, no scaling/shearing
+			//if(T.data == NULL)
+			//	last_T.copyTo(T);
+			//else
+			//	T.copyTo(last_T);
+
+			warpPoints(cur_ftr, cur_ftr_stab, T, true);
+
+			vector <Point2f> objects;
+			vector <float> flow_norm;
+			for(size_t i=0; i < prev_ftr.size(); ++i) {
+				flow_norm.push_back(norm(prev_ftr[i] - cur_ftr_stab[i]) / prev.size().height);
+				line(output, prev_ftr[i], cur_ftr[i], Scalar(200,0,0),1);
+				line(output, prev_ftr[i], cur_ftr_stab[i], Scalar(0,200,0),1);
 			}
 
-		void cleanBBoxes(vector<cv::Rect> nc_rects, cv::Size img, vector<cv::Rect> &c_rects) {
-			int max = 0;
-			for (auto r : nc_rects) {
-				cv::Point center = rectCenter(r);
-				if (r.width > img.width / 2 || r.height > img.height / 8)
-					continue;
-
-				if (center.y > img.height * 2 / 3 || center.y < img.height / 3)
-					continue;
-
-				int r_area = r.area();
-				if (max < r_area) {
-					max = r_area;
-					c_rects.insert(c_rects.begin(), r);
-				} else {
-					c_rects.push_back(r);
-				}
+			for(size_t i=0; i < flow_norm.size(); ++i) {
+				if(flow_norm[i] > MOVEMENT_THRES) {
+					objects.push_back(cur_ftr_stab[i]);
+					prev_ftr.erase(prev_ftr.begin() + i);
+					cur_ftr.erase(cur_ftr.begin() + i);
 				}
 			}
 
-		cv::Point rectCenter(cv::Rect r) {
-			return cv::Point(r.x + r.width/2, r.y + r.height / 2);
+			for(size_t i=0; i < objects.size(); ++i) {
+				circle(output, objects[i], 5, Scalar(0, 200, 0), 1);
+			}
 		}
 
-		inline bool isFlowCorrect(cv::Point2f u)
+
+		inline bool isFlowCorrect(Point2f u)
 		{
 			return !cvIsNaN(u.x) && !cvIsNaN(u.y) && fabs(u.x) < 1e9 && fabs(u.y) < 1e9;
 		}
+
+
+        void droneTracking(Rect img_size)
+        {
+            Point2f centre_image = Point2f(img_size.width/2, img_size.height/2);
+            Point2f centre_rect = Point2f(objectBoundingRectangle.x + objectBoundingRectangle.width/2, objectBoundingRectangle.y + objectBoundingRectangle.height/2);
+
+            geometry_msgs::Twist twist = geometry_msgs::Twist();
+
+            if(centre_rect.x < centre_image.x-THRESHOLD_MOV)
+            {
+                twist.angular.z = 0.2;
+            }
+            else if(centre_rect.x > centre_image.x+THRESHOLD_MOV)
+            {
+                twist.angular.z = -0.2;
+            }
+
+            pub_cmd.publish(twist);
+        }
+
+
+		//int to string helper function
+		string intToString(int number){
+
+			//this function has a number input and string output
+			std::stringstream ss;
+			ss << number;
+			return ss.str();
+		}
 };
 
 

src/videostab.cpp

@@ -0,0 +1,258 @@
+#include <opencv2/opencv.hpp>
+#include <iostream>
+#include <cassert>
+#include <cmath>
+#include <fstream>
+
+using namespace std;
+using namespace cv;
+
+// This video stablisation smooths the global trajectory using a sliding average window
+
+const int SMOOTHING_RADIUS = 30; // In frames. The larger the more stable the video, but less reactive to sudden panning
+const int HORIZONTAL_BORDER_CROP = 20; // In pixels. Crops the border to reduce the black borders from stabilisation being too noticeable.
+
+// 1. Get previous to current frame transformation (dx, dy, da) for all frames
+// 2. Accumulate the transformations to get the image trajectory
+// 3. Smooth out the trajectory using an averaging window
+// 4. Generate new set of previous to current transform, such that the trajectory ends up being the same as the smoothed trajectory
+// 5. Apply the new transformation to the video
+
+struct TransformParam
+{
+    TransformParam() {}
+    TransformParam(double _dx, double _dy, double _da) {
+        dx = _dx;
+        dy = _dy;
+        da = _da;
+    }
+
+    double dx;
+    double dy;
+    double da; // angle
+};
+
+struct Trajectory
+{
+    Trajectory() {}
+    Trajectory(double _x, double _y, double _a) {
+        x = _x;
+        y = _y;
+        a = _a;
+    }
+
+    double x;
+    double y;
+    double a; // angle
+};
+
+int main(int argc, char **argv)
+{
+    if(argc < 2) {
+        cout << "./VideoStab [video.avi]" << endl;
+        return 0;
+    }
+
+    // For further analysis
+    ofstream out_transform("prev_to_cur_transformation.txt");
+    ofstream out_trajectory("trajectory.txt");
+    ofstream out_smoothed_trajectory("smoothed_trajectory.txt");
+    ofstream out_new_transform("new_prev_to_cur_transformation.txt");
+
+    VideoCapture cap(argv[1]);
+    assert(cap.isOpened());
+
+    Mat cur, cur_grey;
+    Mat prev, prev_grey;
+
+    cap >> prev;
+    cvtColor(prev, prev_grey, COLOR_BGR2GRAY);
+
+    // Step 1 - Get previous to current frame transformation (dx, dy, da) for all frames
+    vector <TransformParam> prev_to_cur_transform; // previous to current
+
+    int k=1;
+    int max_frames = cap.get(CV_CAP_PROP_FRAME_COUNT);
+    Mat last_T;
+
+    while(true) {
+        cap >> cur;
+
+        if(cur.data == NULL) {
+            break;
+        }
+
+        cvtColor(cur, cur_grey, COLOR_BGR2GRAY);
+
+        // vector from prev to cur
+        vector <Point2f> prev_corner, cur_corner;
+        vector <Point2f> prev_corner2, cur_corner2;
+        vector <uchar> status;
+        vector <float> err;
+
+        goodFeaturesToTrack(prev_grey, prev_corner, 200, 0.01, 30);
+        calcOpticalFlowPyrLK(prev_grey, cur_grey, prev_corner, cur_corner, status, err);
+
+        // weed out bad matches
+        for(size_t i=0; i < status.size(); i++) {
+            if(status[i]) {
+                prev_corner2.push_back(prev_corner[i]);
+                cur_corner2.push_back(cur_corner[i]);
+            }
+        }
+
+        // translation + rotation only
+        Mat T = estimateRigidTransform(prev_corner2, cur_corner2, false); // false = rigid transform, no scaling/shearing
+
+        // in rare cases no transform is found. We'll just use the last known good transform.
+        if(T.data == NULL) {
+            last_T.copyTo(T);
+        }
+
+        T.copyTo(last_T);
+
+        // decompose T
+        double dx = T.at<double>(0,2);
+        double dy = T.at<double>(1,2);
+        double da = atan2(T.at<double>(1,0), T.at<double>(0,0));
+
+        prev_to_cur_transform.push_back(TransformParam(dx, dy, da));
+
+        out_transform << k << " " << dx << " " << dy << " " << da << endl;
+
+        cur.copyTo(prev);
+        cur_grey.copyTo(prev_grey);
+
+        cout << "Frame: " << k << "/" << max_frames << " - good optical flow: " << prev_corner2.size() << endl;
+        k++;
+    }
+
+    // Step 2 - Accumulate the transformations to get the image trajectory
+
+    // Accumulated frame to frame transform
+    double a = 0;
+    double x = 0;
+    double y = 0;
+
+    vector <Trajectory> trajectory; // trajectory at all frames
+
+    for(size_t i=0; i < prev_to_cur_transform.size(); i++) {
+        x += prev_to_cur_transform[i].dx;
+        y += prev_to_cur_transform[i].dy;
+        a += prev_to_cur_transform[i].da;
+
+        trajectory.push_back(Trajectory(x,y,a));
+
+        out_trajectory << (i+1) << " " << x << " " << y << " " << a << endl;
+    }
+
+    // Step 3 - Smooth out the trajectory using an averaging window
+    vector <Trajectory> smoothed_trajectory; // trajectory at all frames
+
+    for(size_t i=0; i < trajectory.size(); i++) {
+        double sum_x = 0;
+        double sum_y = 0;
+        double sum_a = 0;
+        int count = 0;
+
+        for(int j=-SMOOTHING_RADIUS; j <= SMOOTHING_RADIUS; j++) {
+            if(i+j >= 0 && i+j < trajectory.size()) {
+                sum_x += trajectory[i+j].x;
+                sum_y += trajectory[i+j].y;
+                sum_a += trajectory[i+j].a;
+
+                count++;
+            }
+        }
+
+        double avg_a = sum_a / count;
+        double avg_x = sum_x / count;
+        double avg_y = sum_y / count;
+
+        smoothed_trajectory.push_back(Trajectory(avg_x, avg_y, avg_a));
+
+        out_smoothed_trajectory << (i+1) << " " << avg_x << " " << avg_y << " " << avg_a << endl;
+    }
+
+    // Step 4 - Generate new set of previous to current transform, such that the trajectory ends up being the same as the smoothed trajectory
+    vector <TransformParam> new_prev_to_cur_transform;
+
+    // Accumulated frame to frame transform
+    a = 0;
+    x = 0;
+    y = 0;
+
+    for(size_t i=0; i < prev_to_cur_transform.size(); i++) {
+        x += prev_to_cur_transform[i].dx;
+        y += prev_to_cur_transform[i].dy;
+        a += prev_to_cur_transform[i].da;
+
+        // target - current
+        double diff_x = smoothed_trajectory[i].x - x;
+        double diff_y = smoothed_trajectory[i].y - y;
+        double diff_a = smoothed_trajectory[i].a - a;
+
+        double dx = prev_to_cur_transform[i].dx + diff_x;
+        double dy = prev_to_cur_transform[i].dy + diff_y;
+        double da = prev_to_cur_transform[i].da + diff_a;
+
+        new_prev_to_cur_transform.push_back(TransformParam(dx, dy, da));
+
+        out_new_transform << (i+1) << " " << dx << " " << dy << " " << da << endl;
+    }
+
+    // Step 5 - Apply the new transformation to the video
+    cap.set(CV_CAP_PROP_POS_FRAMES, 0);
+    Mat T(2,3,CV_64F);
+
+    int vert_border = HORIZONTAL_BORDER_CROP * prev.rows / prev.cols; // get the aspect ratio correct
+
+    k=0;
+    while(k < max_frames-1) { // don't process the very last frame, no valid transform
+        cap >> cur;
+
+        if(cur.data == NULL) {
+            break;
+        }
+
+        T.at<double>(0,0) = cos(new_prev_to_cur_transform[k].da);
+        T.at<double>(0,1) = -sin(new_prev_to_cur_transform[k].da);
+        T.at<double>(1,0) = sin(new_prev_to_cur_transform[k].da);
+        T.at<double>(1,1) = cos(new_prev_to_cur_transform[k].da);
+
+        T.at<double>(0,2) = new_prev_to_cur_transform[k].dx;
+        T.at<double>(1,2) = new_prev_to_cur_transform[k].dy;
+
+        Mat cur2;
+
+        warpAffine(cur, cur2, T, cur.size());
+
+        cur2 = cur2(Range(vert_border, cur2.rows-vert_border), Range(HORIZONTAL_BORDER_CROP, cur2.cols-HORIZONTAL_BORDER_CROP));
+
+        // Resize cur2 back to cur size, for better side by side comparison
+        resize(cur2, cur2, cur.size());
+
+        // Now draw the original and stablised side by side for coolness
+        Mat canvas = Mat::zeros(cur.rows, cur.cols*2+10, cur.type());
+
+        cur.copyTo(canvas(Range::all(), Range(0, cur2.cols)));
+        cur2.copyTo(canvas(Range::all(), Range(cur2.cols+10, cur2.cols*2+10)));
+
+        // If too big to fit on the screen, then scale it down by 2, hopefully it'll fit :)
+        if(canvas.cols > 1920) {
+            resize(canvas, canvas, Size(canvas.cols/2, canvas.rows/2));
+        }
+
+        imshow("before and after", canvas);
+
+        //char str[256];
+        //sprintf(str, "images/%08d.jpg", k);
+        //imwrite(str, canvas);
+
+        waitKey(20);
+
+        k++;
+    }
+
+    return 0;
+}