main.cpp

/**
 *  MPAI CAE-ARP Video Analyser.
 *
 *	Implements MPAI CAE-ARP Video Analyser Technical Specification.
 *	It identifies Irregularities on the Preservation Audio-Visual File, providing:
 *	- Irregularity Files;
 *	- Irregularity Images.
 *
 *	WARNING:
 *	Currently, this program is only compatible with the Studer A810 and videos recorded in PAL standard.
 *
 *  @author Nadir Dalla Pozza
 *	@copyright 2022, Audio Innova S.r.l.
 *	@credits Niccolò Pretto, Nadir Dalla Pozza, Sergio Canazza
 *	@license GPL v3.0
 *	@version 1.0.1
 *	@maintainer Nadir Dalla Pozza
 *	@email nadir.dallapozza@unipd.it
 *	@status Production
 */
#include <filesystem>
#include <fstream>
#include <iostream>
#include <stdlib.h>
#include <sys/timeb.h>

#include <boost/program_options.hpp>
#include <boost/uuid/uuid.hpp>            // uuid class
#include <boost/uuid/uuid_generators.hpp> // generators
#include <boost/uuid/uuid_io.hpp>         // streaming operators etc.
#include <boost/lexical_cast.hpp>

#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/imgcodecs.hpp>
#include <opencv2/imgproc.hpp>

#include <nlohmann/json.hpp>

#include "opencv2/core.hpp"
#include "opencv2/calib3d.hpp"
#include "opencv2/highgui.hpp"
#include "opencv2/imgproc.hpp"
#include "opencv2/features2d.hpp"
#include "opencv2/xfeatures2d.hpp"

#include "utility.h"
#include "forAudioAnalyser.h"

#include "lib/colors.h"
#include "lib/time.h"
#include "lib/parser.h"

using namespace cv;
using namespace std;
using json = nlohmann::json;
namespace fs = std::filesystem;
namespace po = boost::program_options;


// For capstan detection, there are two alternative approaches:
// Generalized Hough Transform and SURF.
bool useSURF = true;

bool savingPinchRoller = false;
bool pinchRollerRect = false;
bool savingBrand = false;
bool endTapeSaved = false;
cv::Mat myFrame;
float mediaPrevFrame = 0;
bool firstBrand = true;	// The first frame containing brands on tape must be saved
float firstInstant = 0;
string fileName, extension;

// Path variables
fs::path outputPath;
fs::path irregularityImagesPath;
fs::path videoPath;
// JSON files
json configurationFile;
json irregularityFileOutput1;
json irregularityFileOutput2;
// RotatedRect identifying the processing area
RotatedRect rect, rectTape, rectCapstan;

// Structs

// config.json parameters
struct Config {
	fs::path workingPath;
	string filesName;
	bool brands;
	float speed;
};

struct Threshold {
	float percentual;
	int angle;
	int scale;
	int pos;
};

struct SceneObject {
	int minDist;
	Threshold threshold;
};

Config config;
SceneObject tape;
SceneObject capstan;

void saveFile(fs::path fileName, auto content, bool append) {
	ofstream outputFile;
	if (append) {
		outputFile.open(fileName, ios::app);
	} else {
		outputFile.open(fileName);
	}
	outputFile << content << endl;
	outputFile.close();
}

/**
 * @brief Get operation arguments from command line or config.json file.
 *
 * @param argc Command line arguments count;
 * @param argv Command line arguments.
 * @return true if input configuration is valid;
 * @return false otherwise.
 */
bool getArguments(int argc, char** argv) {
	// Read configuration file
	ifstream iConfig("config/config.json");
	iConfig >> configurationFile;

	if (argc == 1) {
		// Read from JSON file
		string wp = configurationFile["WorkingPath"];

		config = {
			fs::path(wp),
			configurationFile["FilesName"],
			configurationFile["Brands"],
			configurationFile["Speed"]
		};

	} else {
		// Get from command line
		try {
			po::options_description desc(
				"A tool that implements MPAI CAE-ARP Video Analyser Technical Specification.\n"
				"By default, the configuartion parameters are loaded from config/config.json file,\n"
				"but, alternately, you can pass command line arguments to replace them"
			);
			desc.add_options()
				("help,h", "Display this help message")
				("working-path,w", po::value<string>()->required(), "Specify the Working Path, where all input files are stored")
				("files-name,f", po::value<string>()->required(), "Specify the name of the Preservation files (without extension)")
				("brands,b", po::value<bool>()->required(), "Specify if the tape presents brands on its surface")
				("speed,s", po::value<float>()->required(), "Specify the speed at which the tape was read");
			po::variables_map vm;
			po::store(po::command_line_parser(argc, argv).options(desc).run(), vm);
			if (vm.count("help")) {
				cout << desc << "\n";
				return false;
			}
			po::notify(vm);

			// Access the stored options
			config.workingPath = fs::path(vm["working-path"].as<string>());
			config.filesName = vm["files-name"].as<string>();
			config.brands = vm["brands"].as<bool>();
			config.speed = vm["speed"].as<float>();

		} catch (po::invalid_command_line_syntax& e) {
			cerr << RED << BOLD << "The command line syntax is invalid: " << END << RED << e.what() << END << endl;
			return false;
		} catch (po::required_option& e) {
			cerr << "Error: " << e.what() << endl;
			return false;
		}
	}

	capstan = {
		configurationFile["MinDistCapstan"],
		{
			configurationFile["CapstanThresholdPercentual"],
			configurationFile["AngleThreshCapstan"],
			configurationFile["ScaleThreshCapstan"],
			configurationFile["PosThreshCapstan"]
		}
	};

	tape = {
		configurationFile["MinDist"],
		{
			configurationFile["TapeThresholdPercentual"],
			configurationFile["AngleThresh"],
			configurationFile["ScaleThresh"],
			configurationFile["PosThresh"]
		}
	};

	return true;
}


/**
 * @brief Identifies the Regions Of Interest (ROIs) on the video,
 * which are:
 * - The reading head;
 * - The tape area under the tape head (computed on the basis of the detected reading head);
 * - The capstan.
 *
 * @return true if some areas have been detected;
 * @return false otherwise.
 */
bool findProcessingAreas() {

	/*********************************************************************************************/
	/*********************************** READING HEAD DETECTION **********************************/
	/*********************************************************************************************/

	// Obtain grayscale version of myFrame
	Mat myFrameGrayscale;
	cvtColor(myFrame, myFrameGrayscale, COLOR_BGR2GRAY);
	// Get input shape in grayscale
	Mat templateImage = imread("input/readingHead.png", IMREAD_GRAYSCALE);
	// Downsample myFrameGrayscale in half pixels for performance reasons
	Mat myFrameGrayscaleHalf;
	pyrDown(myFrameGrayscale, myFrameGrayscaleHalf, Size(myFrame.cols/2, myFrame.rows/2));
	// Downsample tapeShape in half pixels
	Mat templateImageHalf;
	pyrDown(templateImage, templateImageHalf, Size(templateImage.cols/2, templateImage.rows/2));

	// Process only the bottom-central portion of the input video -> best results with our videos
	Rect readingHeadProcessingAreaRect(myFrameGrayscaleHalf.cols/4, myFrameGrayscaleHalf.rows/2, myFrameGrayscaleHalf.cols/2, myFrameGrayscaleHalf.rows/2);
	Mat processingImage = myFrameGrayscaleHalf(readingHeadProcessingAreaRect);
	// Select the template to be detected
	Mat templateShape = templateImageHalf;

	// Algorithm and parameters
	Ptr<GeneralizedHoughGuil> alg = createGeneralizedHoughGuil();

	vector<Vec4f> positionsPos, positionsNeg;
	Mat votesPos, votesNeg;
	TickMeter tm;
	int oldPosThresh = tape.threshold.pos;
	RotatedRect rectPos, rectNeg;
	ofstream myFile;
	Point2f pts[4];

	// Find the best matches for positive and negative angles
	// If there are more than one shapes, then choose the one with the highest score
	// If there are more than one with the same highest score, then arbitrarily choose the latest
	double maxValPos = 0, maxValNeg = 0;
	int indexPos = 0, indexNeg = 0;

	alg -> setMinDist(tape.minDist);
	alg -> setLevels(360);
	alg -> setDp(2);
	alg -> setMaxBufferSize(1000);

	alg -> setAngleStep(1);
	alg -> setAngleThresh(tape.threshold.angle);

	alg -> setMinScale(0.9);
	alg -> setMaxScale(1.1);
	alg -> setScaleStep(0.01);
	alg -> setScaleThresh(tape.threshold.scale);

	alg -> setPosThresh(tape.threshold.pos);

	alg -> setCannyLowThresh(150); // Old: 100
	alg -> setCannyHighThresh(240); // Old: 300

	alg -> setTemplate(templateShape);

	cout << DARK_CYAN << "Reading head" << END << endl;
	tm.start();
	// Invoke utility.h function
	
	detectShape(alg, templateShape, tape.threshold.pos, positionsPos, votesPos, positionsNeg, votesNeg, processingImage);
	tm.stop();
	cout << "Reading head detection time: " << tm.getTimeMilli() << " ms" << endl;

	for (int i = 0; i < votesPos.size().width; i++) {
		if (votesPos.at<int>(i) >= maxValPos) {
			maxValPos = votesPos.at<int>(i);
			indexPos = i;
		}
	}

	for (int i = 0; i < votesNeg.size().width; i++) {
		if (votesNeg.at<int>(i) >= maxValNeg) {
			maxValNeg = votesNeg.at<int>(i);
			indexNeg = i;
		}
	}

	// The color is progressively darkened to emphasize that the algorithm found more than one shape
	if (positionsPos.size() > 0)
		rectPos = drawShapes(myFrame, positionsPos[indexPos], Scalar(0, 0, 255-indexPos*64), templateImageHalf.cols, templateImageHalf.rows, myFrameGrayscaleHalf.cols/4, myFrameGrayscaleHalf.rows/2, 2);
	if (positionsNeg.size() > 0)
		rectNeg = drawShapes(myFrame, positionsNeg[indexNeg], Scalar(128, 128, 255-indexNeg*64), templateImageHalf.cols, templateImageHalf.rows, myFrameGrayscaleHalf.cols/4, myFrameGrayscaleHalf.rows/2, 2);

	myFile.open("log.txt", ios::app);

	if (maxValPos > 0)
		if (maxValNeg > 0)
			if (maxValPos > maxValNeg) {
				myFile << "READING HEAD: Positive angle is best, match number: " << indexPos << endl;
				rect = rectPos;
			} else {
				myFile << "READING HEAD: Negative angle is best, match number: " << indexNeg << endl;
				rect = rectNeg;
			}
		else {
			myFile << "READING HEAD: Positive angle is the only choice, match number: " << indexPos << endl;
			rect = rectPos;
		}
	else if (maxValNeg > 0) {
		myFile << "READING HEAD: Negative angle is the only choice, match number: " << indexNeg << endl;
		rect = rectNeg;
	} else {
		myFile.close();
		return false;
	}
	cout << endl;

	rect.points(pts);

	/*********************************************************************************************/
	/************************************ TAPE AREA DETECTION ************************************/
	/*********************************************************************************************/

	// Compute area basing on reading head detection
	Vec4f positionTape( rect.center.x, rect.center.y + rect.size.height / 2 + 20 * (rect.size.width / 200), 1, rect.angle );
	rectTape = drawShapes(myFrame, positionTape, Scalar(0, 255-indexPos*64, 0), rect.size.width, 50 * (rect.size.width / 200), 0, 0, 1);

	myFile << "Tape area:" << endl;
	myFile << "  Center (x, y): (" << rectTape.center.x << ", " << rectTape.center.y << ")" << endl;
	myFile << "  Size (w, h): (" << rectTape.size.width << ", " << rectTape.size.height << ")" << endl;
	myFile << "  Angle (deg): (" << rectTape.angle << ")" << endl;

	json autoJSON;
	autoJSON["PreservationAudioVisualFile"] = fileName;
	autoJSON["RotatedRect"] = {
		{
			"CenterX", rectTape.center.x
		}, {
			"CenterY", rectTape.center.y
		}, {
			"Width", rectTape.size.width
		}, {
			"Height", rectTape.size.height
		}, {
			"Angle", rectTape.angle
		}
	};

	saveFile(fs::path("./" + fileName + ".json"), autoJSON.dump(4), false);

	/*********************************************************************************************/
	/************************************* CAPSTAN DETECTION *************************************/
	/*********************************************************************************************/

	// Read template image - it is smaller than before, therefore there is no need to downsample
	templateShape = imread("input/capstanBERIO058prova.png", IMREAD_GRAYSCALE); // WORKING
	// templateShape = imread("../input/capstanBERIO058.png", IMREAD_GRAYSCALE);

	cout << DARK_CYAN << "Capstan" << END << endl;

	if (useSURF) {

		// Step 1: Detect the keypoints using SURF Detector, compute the descriptors
		int minHessian = 100;
		Ptr<xfeatures2d::SURF> detector = xfeatures2d::SURF::create(minHessian);
		vector<KeyPoint> keypoints_object, keypoints_scene;
		Mat descriptors_object, descriptors_scene;

		tm.reset();
		tm.start();
		detector->detectAndCompute(templateShape, noArray(), keypoints_object, descriptors_object);
		detector->detectAndCompute(myFrameGrayscale, noArray(), keypoints_scene, descriptors_scene);
		tm.stop();
		cout << "Capstan detection time: " << tm.getTimeMilli() << " ms" << endl;

		// Step 2: Matching descriptor vectors with a FLANN based matcher
		// Since SURF is a floating-point descriptor NORM_L2 is used
		Ptr<DescriptorMatcher> matcher = DescriptorMatcher::create(DescriptorMatcher::FLANNBASED);
		vector<vector<DMatch>> knn_matches;
		matcher->knnMatch(descriptors_object, descriptors_scene, knn_matches, 2);
		//-- Filter matches using the Lowe's ratio test
		const float ratio_thresh = 0.75f;
		vector<DMatch> good_matches;
		for (size_t i = 0; i < knn_matches.size(); i++) {
			if (knn_matches[i][0].distance < ratio_thresh * knn_matches[i][1].distance) {
				good_matches.push_back(knn_matches[i][0]);
			}
		}
		// Draw matches
		Mat img_matches;
		drawMatches(templateShape, keypoints_object, myFrameGrayscale, keypoints_scene, good_matches, img_matches, Scalar::all(-1), Scalar::all(-1), vector<char>(), DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS);
		// Localize the object
		vector<Point2f> obj;
		vector<Point2f> scene;
		for (size_t i = 0; i < good_matches.size(); i++) {
			// Get the keypoints from the good matches
			obj.push_back(keypoints_object[good_matches[i].queryIdx].pt);
			scene.push_back(keypoints_scene[good_matches[i].trainIdx].pt);
		}
		Mat H = findHomography(obj, scene, RANSAC);
		// Get the corners from the image_1 ( the object to be "detected" )
		vector<Point2f> obj_corners(4);
		obj_corners[0] = Point2f(0, 0);
		obj_corners[1] = Point2f((float)templateShape.cols, 0);
		obj_corners[2] = Point2f((float)templateShape.cols, (float)templateShape.rows);
		obj_corners[3] = Point2f(0, (float)templateShape.rows);
		vector<Point2f> scene_corners(4);
		perspectiveTransform( obj_corners, scene_corners, H);

		// Find average
		float capstanX = (scene_corners[0].x + scene_corners[1].x + scene_corners[2].x + scene_corners[3].x) / 4;
		float capstanY = (scene_corners[0].y + scene_corners[1].y + scene_corners[2].y + scene_corners[3].y) / 4;

		// In the following there are two alterations to cut the first 20 horizontal pixels and the first 90 vertical pixels from the found rectangle:
		// +10 in X for centering and -20 in width
		// +45 in Y for centering and -90 in height
		Vec4f positionCapstan(capstanX + 10, capstanY + 45, 1, 0);
		rectCapstan = drawShapes(myFrame, positionCapstan, Scalar(255-indexPos*64, 0, 0), templateShape.cols - 20, templateShape.rows - 90, 0, 0, 1);

	} else {

		// Process only right portion of the image, wherw the capstain always appears
		int capstanProcessingAreaRectX = myFrame.cols*3/4;
		int capstanProcessingAreaRectY = myFrame.rows/2;
		int capstanProcessingAreaRectWidth = myFrame.cols/4;
		int capstanProcessingAreaRectHeight = myFrame.rows/2;
		Rect capstanProcessingAreaRect(capstanProcessingAreaRectX, capstanProcessingAreaRectY, capstanProcessingAreaRectWidth, capstanProcessingAreaRectHeight);
		Mat capstanProcessingAreaGrayscale = myFrameGrayscale(capstanProcessingAreaRect);
		// Reset algorithm and set parameters
		alg = createGeneralizedHoughGuil();

		alg -> setMinDist(capstan.minDist);
		alg -> setLevels(360);
		alg -> setDp(2);
		alg -> setMaxBufferSize(1000);

		alg -> setAngleStep(1);
		alg -> setAngleThresh(capstan.threshold.angle);

		alg -> setMinScale(0.9);
		alg -> setMaxScale(1.1);
		alg -> setScaleStep(0.01);
		alg -> setScaleThresh(capstan.threshold.scale);

		alg -> setPosThresh(capstan.threshold.pos);

		alg -> setCannyLowThresh(150);
		alg -> setCannyHighThresh(240);

		alg -> setTemplate(templateShape);

		oldPosThresh = capstan.threshold.pos;

		vector<Vec4f> positionsC1Pos, positionsC1Neg;
		Mat votesC1Pos, votesC1Neg;

		tm.reset();
		tm.start();
		detectShape(alg, templateShape, capstan.threshold.pos, positionsC1Pos, votesC1Pos, positionsC1Neg, votesC1Neg, capstanProcessingAreaGrayscale);
		tm.stop();
		cout << "Capstan detection time: " << tm.getTimeMilli() << " ms" << endl;

		// Find the best matches for positive and negative angles
		// If there are more than one shapes, then choose the one with the highest score
		// If there are more than one with the same highest score, then choose the latest
		maxValPos = 0, maxValNeg = 0, indexPos = 0, indexNeg = 0;

		for (int i = 0; i < votesC1Pos.size().width; i++) {
			if (votesC1Pos.at<int>(i) >= maxValPos) {
				maxValPos = votesC1Pos.at<int>(i);
				indexPos = i;
			}
		}

		for (int i = 0; i < votesC1Neg.size().width; i++) {
			if (votesC1Neg.at<int>(i) >= maxValNeg) {
				maxValNeg = votesC1Neg.at<int>(i);
				indexNeg = i;
			}
		}

		RotatedRect rectCapstanPos, rectCapstanNeg;
		if (positionsC1Pos.size() > 0)
			rectCapstanPos = drawShapes(myFrame, positionsC1Pos[indexPos], Scalar(255-indexPos*64, 0, 0), templateShape.cols-22, templateShape.rows-92, capstanProcessingAreaRectX+11, capstanProcessingAreaRectY+46, 1);
		if (positionsC1Neg.size() > 0)
			rectCapstanNeg = drawShapes(myFrame, positionsC1Neg[indexNeg], Scalar(255-indexNeg*64, 128, 0), templateShape.cols-22, templateShape.rows-92, capstanProcessingAreaRectX+11, capstanProcessingAreaRectY+46, 1);

		if (maxValPos > 0)
			if (maxValNeg > 0)
				if (maxValPos > maxValNeg) {
					myFile << "CAPSTAN: Positive is best, match number: " << indexPos << endl;
					rectCapstan = rectCapstanPos;
				} else {
					myFile << "CAPSTAN: Negative is best, match number: " << indexNeg << endl;
					rectCapstan = rectCapstanNeg;
				}
			else {
				myFile << "CAPSTAN: Positive is the only choice, match number: " << indexPos << endl;
				rectCapstan = rectCapstanPos;
			}
		else if (maxValNeg > 0) {
			myFile << "CAPSTAN: Negative is the only choice, match number: " << indexNeg << endl;
			rectCapstan = rectCapstanNeg;
		} else {
			myFile.close();
			return false;
		}

	}

	myFile << "Capstan ROI:" << endl;
	myFile << "  Center (x, y): (" << rectCapstan.center.x << ", " << rectCapstan.center.y << ")" << endl;
	myFile << "  Size (w, h): (" << rectCapstan.size.width << ", " << rectCapstan.size.height << ")" << endl;
	myFile << "  Angle (deg): (" << rectCapstan.angle << ")" << endl;
	myFile.close();
	cout << endl;

	// Save the image containing the detected areas
	cv::imwrite(outputPath.string() + "/tapeAreas.jpg", myFrame);

	return true;
}


/**
 * @brief Compares two consecutive video frames and establish if there potentially is an Irregularity.
 * The comparison is pixel-wise and based on threshold values set on config.json file.
 *
 * @param prevFrame the frame before the current one;
 * @param currentFrame the current frame;
 * @param msToEnd the number of milliseconds left before the end of the video. Useful for capstan analysis.
 * @return true if a potential Irregularity has been found;
 * @return false otherwise.
 */
bool frameDifference(cv::Mat prevFrame, cv::Mat currentFrame, int msToEnd) {

	/*********************************************************************************************/
	/********************************** Capstan analysis *****************************************/
	/*********************************************************************************************/

	// In the last minute of the video, check for pinchRoller position for endTape event
	if (!endTapeSaved && msToEnd < 60000) {

		// Capstan area
		int capstanAreaPixels = rectCapstan.size.width * rectCapstan.size.height;
		float capstanDifferentPixelsThreshold = capstanAreaPixels * capstan.threshold.percentual / 100;

		// Extract matrices corresponding to the processing area
		// CODE FROM https://answers.opencv.org/question/497/extract-a-rotatedrect-area/

		// matrices we'll use
		Mat M, rotatedPrevFrame, croppedPrevFrame, rotatedCurrentFrame, croppedCurrentFrame;
		// get angle and size from the bounding box
		float angle = rectCapstan.angle;
		Size rect_size = rectCapstan.size;
		// thanks to http://felix.abecassis.me/2011/10/opencv-rotation-deskewing/
		if (rectCapstan.angle < -45.) {
			angle += 90.0;
			swap(rect_size.width, rect_size.height);
		}
		// get the rotation matrix
		M = getRotationMatrix2D(rectCapstan.center, angle, 1.0);
		// perform the affine transformation
		cv::warpAffine(prevFrame, rotatedPrevFrame, M, prevFrame.size(), INTER_CUBIC);
		cv::warpAffine(currentFrame, rotatedCurrentFrame, M, currentFrame.size(), INTER_CUBIC);
		// crop the resulting image
		cv::getRectSubPix(rotatedPrevFrame, rect_size, rectCapstan.center, croppedPrevFrame);
		cv::getRectSubPix(rotatedCurrentFrame, rect_size, rectCapstan.center, croppedCurrentFrame);

		// END CODE FROM https://answers.opencv.org/question/497/extract-a-rotatedrect-area/

		cv::Mat differenceFrame = difference(croppedPrevFrame, croppedCurrentFrame);

		int blackPixelsCapstan = 0;

		for (int i = 0; i < croppedCurrentFrame.rows; i++) {
			for (int j = 0; j < croppedCurrentFrame.cols; j++) {
				if (differenceFrame.at<cv::Vec3b>(i, j)[0] == 0) {
					// There is a black pixel, then there is a difference between previous and current frames
					blackPixelsCapstan++;
				}
			}
		}

		if (blackPixelsCapstan > capstanDifferentPixelsThreshold) {
			savingPinchRoller = true;
			endTapeSaved = true; // Never check again for end tape instant
			return true;
		} else {
			savingPinchRoller = false;
		}
	} else {
		savingPinchRoller = false; // It will already be false before the last minute of the video. After having saved the capstan, the next time reset the variable to not save again
	}

	/*********************************************************************************************/
	/************************************ Tape analysis ******************************************/
	/*********************************************************************************************/

	// Tape area
    int tapeAreaPixels = rectTape.size.width * rectTape.size.height;
	float tapeDifferentPixelsThreshold = tapeAreaPixels * tape.threshold.percentual / 100;

	// Extract matrices corresponding to the processing area
	// CODE FROM https://answers.opencv.org/question/497/extract-a-rotatedrect-area/

	// matrices we'll use
	Mat M, rotatedPrevFrame, croppedPrevFrame, rotatedCurrentFrame, croppedCurrentFrame;
	// get angle and size from the bounding box
	float angle = rectTape.angle;
	Size rect_size = rectTape.size;
	// thanks to http://felix.abecassis.me/2011/10/opencv-rotation-deskewing/
	if (rectTape.angle < -45.) {
		angle += 90.0;
		swap(rect_size.width, rect_size.height);
	}
	// get the rotation matrix
	M = getRotationMatrix2D(rectTape.center, angle, 1.0);
	// perform the affine transformation
	cv::warpAffine(prevFrame, rotatedPrevFrame, M, prevFrame.size(), INTER_CUBIC);
	cv::warpAffine(currentFrame, rotatedCurrentFrame, M, currentFrame.size(), INTER_CUBIC);
	// crop the resulting image
	cv::getRectSubPix(rotatedPrevFrame, rect_size, rectTape.center, croppedPrevFrame);
	cv::getRectSubPix(rotatedCurrentFrame, rect_size, rectTape.center, croppedCurrentFrame);

	// END CODE FROM https://answers.opencv.org/question/497/extract-a-rotatedrect-area/

	cv::Mat differenceFrame = difference(croppedPrevFrame, croppedCurrentFrame);

	int decEnd = (msToEnd % 1000) / 100;
	int secEnd = (msToEnd - (msToEnd % 1000)) / 1000;
	int minEnd = secEnd / 60;
	secEnd = secEnd % 60;


	/************************************* Segment analysis **************************************/

  	int blackPixels = 0;
	float mediaCurrFrame;
	int totColoreCF = 0;

	for (int i = 0; i < croppedCurrentFrame.rows; i++) {
		for (int j = 0; j < croppedCurrentFrame.cols; j++) {
			totColoreCF += croppedCurrentFrame.at<cv::Vec3b>(i, j)[0] + croppedCurrentFrame.at<cv::Vec3b>(i, j)[1] + croppedCurrentFrame.at<cv::Vec3b>(i, j)[2];
			if (differenceFrame.at<cv::Vec3b>(i, j)[0] == 0) {
				blackPixels++;
			}
		}
	}
	mediaCurrFrame = totColoreCF/tapeAreaPixels;

	/************************************* Decision stage ****************************************/

	bool isIrregularity = false;

	if (blackPixels > tapeDifferentPixelsThreshold) { // The threshold must be passed

		/***** AVERAGE_COLOR-BASED DECISION *****/
		if (mediaPrevFrame > (mediaCurrFrame + 7) || mediaPrevFrame < (mediaCurrFrame - 7)) { // They are not similar for color average
			isIrregularity = true;
		}

		/***** BRANDS MANAGEMENT *****/
		if (config.brands) {
			// At the beginning of the video, wait at least 5 seconds before the next Irregularity to consider it as a brand.
			// It is not guaranteed that it will be the first brand, but it is generally a safe approach to have a correct image
			if (firstBrand) {
				if (firstInstant - msToEnd > 5000) {
					firstBrand = false;
					savingBrand = true;
					isIrregularity = true;
				}
			// In the following iterations reset savingBrand, since we are no longer interested in brands.
			} else
				savingBrand = false;
		}

	}

	// Update mediaPrevFrame
	mediaPrevFrame = mediaCurrFrame;

	return isIrregularity;
}


/**
 * @brief video processing phase, where each frame is analysed.
 * It saves the IrregularityImages and updates the IrregularityFiles if an Irregularity is found
 *
 * @param videoCapture the input Preservation Audio-Visual File;
 */
void processing(cv::VideoCapture videoCapture) {

	// Video duration
	int frameNumbers_v = videoCapture.get(CAP_PROP_FRAME_COUNT);
	float fps_v = videoCapture.get(CAP_PROP_FPS); // FPS can be non-integers!!!
	float videoLength = (float) frameNumbers_v / fps_v; // [s]
	int videoLength_ms = videoLength * 1000;

    int savedFrames = 0, unsavedFrames = 0;
	float lastSaved = -160;
	// Whenever we find an Irregularity, we want to skip a lenght equal to the Studer reading head (3 cm = 1.18 inches).
	int savingRate = 79; // [ms]. Time taken to cross 3 cm at 15 ips, or 1.5 cm at 7.5 ips. The considered lengths are the widths of the tape areas.
	// The following condition constitutes a valid approach if the tape areas have widths always equal to the reading head
	if (config.speed == 7.5)
		savingRate = 157; // Time taken to cross 3 cm at 7.5 ips

	// The first frame of the video won't be processed
    cv::Mat prevFrame;
	videoCapture >> prevFrame;
	firstInstant = videoLength_ms - videoCapture.get(CAP_PROP_POS_MSEC);

    while (videoCapture.isOpened()) {

		cv::Mat frame;
        videoCapture >> frame;

        if (!frame.empty()) {

			int ms = videoCapture.get(CAP_PROP_POS_MSEC);
			int msToEnd = videoLength_ms - ms;
			if (ms == 0) // With OpenCV library, this happens at the last few frames of the video before realising that "frame" is empty.
				break;

			// Variables to display program status
			int secToEnd = msToEnd / 1000;
			int minToEnd = (secToEnd / 60) % 60;
			secToEnd = secToEnd % 60;

			string secStrToEnd = to_string(secToEnd), minStrToEnd = to_string(minToEnd);
			if (minToEnd < 10)
				minStrToEnd = "0" + minStrToEnd;
			if (secToEnd < 10)
				secStrToEnd = "0" + secStrToEnd;

			// Display program status
			cout << "\rIrregularities: " << savedFrames << ".   ";
			cout << "Remaining video time [mm:ss]: " << minStrToEnd << ":" << secStrToEnd << flush;

			if ((ms - lastSaved > savingRate) && frameDifference(prevFrame, frame, msToEnd)) {

				// An Irregularity has been found!

				// De-interlacing frame
				cv::Mat oddFrame(frame.rows/2, frame.cols, CV_8UC3);
				cv::Mat evenFrame(frame.rows/2, frame.cols, CV_8UC3);
				separateFrame(frame, oddFrame, evenFrame);

				// Extract the image corresponding to the ROIs
				Point2f pts[4];
				if (savingPinchRoller)
					rectCapstan.points(pts);
				else
					rectTape.points(pts);
				cv::Mat subImage(frame, cv::Rect(100, min(pts[1].y, pts[2].y), frame.cols - 100, static_cast<int>(rectTape.size.height)));

				// De-interlacing
				cv::Mat oddSubImage(subImage.rows/2, subImage.cols, CV_8UC3);
				int evenSubImageRows = subImage.rows/2;
				if (subImage.rows % 2 != 0) // If the found rectangle is of odd height, we must increase evenSubImage height by 1, otherwise we have segmentation_fault!!!
					evenSubImageRows += 1;
				cv::Mat evenSubImage(evenSubImageRows, subImage.cols, CV_8UC3);
				separateFrame(subImage, oddSubImage, evenSubImage);

				string timeLabel = getTimeLabel(ms, ":");
				string safeTimeLabel = getTimeLabel(ms, "-");

				string irregularityImageFilename = to_string(savedFrames) + "_" + safeTimeLabel + ".jpg";
				cv::imwrite(irregularityImagesPath / irregularityImageFilename, oddFrame);

				// Append Irregularity information to JSON
				boost::uuids::uuid uuid = boost::uuids::random_generator()();
				irregularityFileOutput1["Irregularities"] += {
					{
						"IrregularityID", boost::lexical_cast<string>(uuid)
					}, {
						"Source", "v"
					}, {
						"TimeLabel", timeLabel
					}
				};
				irregularityFileOutput2["Irregularities"] += {
					{
						"IrregularityID", boost::lexical_cast<string>(uuid)
					}, {
						"Source", "v"
					}, {
						"TimeLabel", timeLabel
					}, {
						"ImageURI", irregularityImagesPath.string() + "/" + irregularityImageFilename
					}
				};

				lastSaved = ms;
				savedFrames++;

			} else {
				unsavedFrames++;
			}

			prevFrame = frame;

	    } else {
			cout << endl << "Empty frame!" << endl;
	    	videoCapture.release();
	    	break;
	    }
	}

	ofstream myFile;
	myFile.open("log.txt", ios::app);
	myFile << "Saved frames are: " << savedFrames << endl;
	myFile.close();

}


/*************************************************************************************************/
/********************************************* MAIN **********************************************/
/*************************************************************************************************/

/**
 * @brief main program, organised as:
 * - Get input from command line or config.json file;
 * - Check input parameters;
 * - Creation of output directories;
 * - Regions Of Interest (ROIs) detection;
 * - Irregularities detection;
 * - Saving of output IrregularityFiles.
 *
 * @param argc Command line arguments count;
 * @param argv Command line arguments.
 * @return int program status.
 */
int main(int argc, char** argv) {

	json irregularityFileInput;
	fs::path irregularityFileInputPath;
	/*********************************************************************************************/
	/*************************************** CONFIGURATION ***************************************/
	/*********************************************************************************************/

	// Get the input from config.json or command line
	try {
		bool continueExecution = getArguments(argc, argv);
		if (!continueExecution) {
			return 0;
		}
	} catch (nlohmann::detail::type_error e) {
		cerr << RED << "config.json error!" << endl << e.what() << END << endl;
		return -1;
	}

	videoPath = config.workingPath / "PreservationAudioVisualFile" / config.filesName;
    if (findFileName(videoPath, fileName, extension) == -1) {
        cerr << RED << BOLD << "config.json error!" << END << endl << RED << videoPath.string() << " cannot be found or opened." << END << endl;
        return -1;
    }

	irregularityFileInputPath = config.workingPath / "temp" / fileName / "AudioAnalyser_IrregularityFileOutput1.json";


	// Input JSON check
	ifstream iJSON(irregularityFileInputPath);
	if (iJSON.fail()) {
		cerr << RED << BOLD << "config.json error!" << END << endl << RED << irregularityFileInputPath.string() << " cannot be found or opened." << END << endl;
		return -1;
	}
	if (config.speed != 7.5 && config.speed != 15) {
		cerr << RED << BOLD << "config.json error!" << END << endl << RED << "Speed parameter must be 7.5 or 15 ips." << END << endl;
		return -1;
	}
	if (tape.threshold.percentual < 0 || tape.threshold.percentual > 100) {
		cerr << RED << BOLD << "config.json error!" << END << endl << RED << "TapeThresholdPercentual parameter must be a percentage value." << END << endl;
		return -1;
	}
	if (capstan.threshold.percentual < 0 || capstan.threshold.percentual > 100) {
		cerr << RED << BOLD << "config.json error!" << END << endl << RED << "CapstanThresholdPercentual parameter must be a percentage value." << END << endl;
		return -1;
	}

	// Adjust input paramenters (considering given ones as pertinent to a speed reference = 7.5)
	if (config.brands) {
		if (config.speed == 15)
			tape.threshold.percentual += 6;
	} else
		if (config.speed == 15)
			tape.threshold.percentual += 20;
		else
			tape.threshold.percentual += 21;

    cout << endl;
	cout << "Parameters:" << endl;
	cout << "    Brands: " << config.brands << endl;
	cout << "    Speed: " << config.speed << endl;
    cout << "    ThresholdPercentual: " << tape.threshold.percentual << endl;
	cout << "    ThresholdPercentualCapstan: " << capstan.threshold.percentual << endl;
	cout << endl;

	// Read input JSON
	iJSON >> irregularityFileInput;

	/*********************************************************************************************/
	/*********************************** MAKE OUTPUT DIRECTORY ***********************************/
	/*********************************************************************************************/

	// Make directory with fileName name
	outputPath = config.workingPath / "temp" / fileName;
	int outputFileNameDirectory = create_directory(outputPath);
	// Get now time
	time_t t = chrono::system_clock::to_time_t(chrono::system_clock::now());
    string ts = ctime(&t);
	// Write useful info to log file
	fs::path logPath = outputPath / "log.txt";

	string logInfo = fileName + '\n' + "tsh: " + to_string(tape.threshold.percentual) + "   tshp: " + to_string(capstan.threshold.percentual) + '\n' + ts;
	saveFile(logPath, logInfo, true);


	/*********************************************************************************************/
	/************************************** AREAS DETECTION **************************************/
	/*********************************************************************************************/

	cv::VideoCapture videoCapture(videoPath);
    if (!videoCapture.isOpened()) {
        cerr << RED << BOLD << "Video unreadable." << END << endl;
        return -1;
    }

	// Get total number of frames
	int totalFrames = videoCapture.get(CAP_PROP_FRAME_COUNT);
	// Set frame position to half video length
	videoCapture.set(CAP_PROP_POS_FRAMES, totalFrames/2);
	// Get frame
	videoCapture >> myFrame;

	cout << "Video resolution: " << myFrame.cols << "x" << myFrame.rows << endl << endl;

	// Find the processing area corresponding to the tape area over the reading head
	bool found = findProcessingAreas();

	// Reset frame position
	videoCapture.set(CAP_PROP_POS_FRAMES, 0);

	// Write useful information to log file
	string message;
	if (found) {
		message = "Processing areas found!\n";
		cout << message;
		saveFile(logPath, message, true);
	} else {
		message = "Processing area not found. Try changing JSON parameters.\n";
		cout << message;
		saveFile(logPath, message, true);
		return -1; // Program terminated early
	}

	/*********************************************************************************************/
	/***************************** MAKE ADDITIONAL OUTPUT DIRECTORIES ****************************/
	/*********************************************************************************************/

	irregularityImagesPath = outputPath / "IrregularityImages";
	int fullFrameDirectory = fs::create_directory(irregularityImagesPath);

	/*********************************************************************************************/
	/**************************************** PROCESSING *****************************************/
	/*********************************************************************************************/

	cout << '\n' << CYAN << "Starting processing..." << END << '\n';

	// Processing timer
	time_t startTimer, endTimer;
	startTimer = time(NULL);

	processing(videoCapture);

	endTimer = time(NULL);
	float min = (endTimer - startTimer) / 60;
	float sec = (endTimer - startTimer) % 60;

	string result("Processing elapsed time: " + to_string((int)min) + ":" + to_string((int)sec));
	cout << endl << result << endl;

	saveFile("log.txt", result + '\n', true);

	/*********************************************************************************************/