int main(int argc, char ** argv) {
//The angle used for matching validation
double angle = 10;
//The distance threshold
double distance = 200;
//Determine if the KNN validation criterion is necessary
bool usingKnnCriterion = true;
//The horizon line
int horizonLine = 300;
//Using the terminal
bool terminal = false;
//For BRISK SURF Using radius = 0.20, threshold = 70
bool hamming=false;
std::string feat_detector = "BRISK";
std::string feat_descriptor = "SURF";
double hammingDistance = 0.28;
//double threshold = 30;//45
double threshold = 43.75;
//For SBRISK SBRISK, hammingDistance = 85, Threshold = 100
// bool hamming=true;
// std::string feat_detector = "BRISK";
// std::string feat_descriptor = "BRISK";
// double threshold = 46.25;//46.25;//46.25 KNN
// // double threshold = 78.75;//Hamming
// double hammingDistance = 121.25;//Hamming
//For BRISK4 (4 octaves)
// bool hamming=true;
// std::string feat_detector = "BRISK4";
// std::string feat_descriptor = "BRISK4";
// //double threshold = 30; //KNN 51.25
// double threshold = 51.25;//Hamming
// double hammingDistance = 120;//Hamming
//Set the date and time
string myDate = "06082012";
string myTime = "1402";
//Set if you are matching or not matching
bool isMatching = false;
//Determine whether the MPS or CPS threshold is being used
bool isMax = true;
//Are we comparing camera pics to the Nao pics
bool usingCamera = true;
//Lighting parameter
lighting light = LIGHT_BOTH;
//NB *************************************
//CHOOSE THE DATASET TO USE
int dataset=1;
//dataset =1 for original robocup pics
//dataset = 2 for office pics and large hall pics
//dataset = 3 for varying illumination and for comparing the robocup dataset with the camera robocup dataset
//dataset = 4 Nao Street View 27-28 (directory indices) with Google Street view
// AND second dataset 19-20 with camera pic indices 25-26
//Create the non-matching upper and lower bounds
int upperBound = 0;
int lowerBound = 0;
//Matching parameters
int k_start =0;
int k_end = 0;
int s_start = 0;
int s_end = 0;
int index = 0;
//Counters used to ensure standard output for processing in Matlab
int tempDirCounterkk = 0;
int tempDirCounterss = 0;
//USED FOR LIGHTING AND FOR COMPARING NAO CAMERA PICS
int step = 4;
//If we are performing the matching routine
if(isMatching)
{
if (dataset==1)//The original dataset
{
//Counters used to ensure standard output for processing in Matlab
tempDirCounterkk = 1;
tempDirCounterss = 1;
k_start = 1;//Left dataset: 5; Right Dataset: 9; Both Dataset: 13
k_end = 4;//Left dataset: 8; Right Dataset 12; Both Dataset: 16
}
else if (dataset==2)//Additional datasets
{
//Counters used to ensure standard output for processing in Matlab
tempDirCounterkk = 5;
tempDirCounterss = 5;
k_start = 19; //Dataset 2: 17; Dataset 3: 19
k_end = 20; //Dataset 2: 18; Dataset 3: 20
}
else if (dataset==3) //Varying lighting
{
//Counters used to ensure standard output for processing in Matlab
tempDirCounterkk = 1;
tempDirCounterss = 1;
k_start = 1;//Left dataset: 5; Right Dataset: 9; Both Dataset: 13
k_end = 4;//Left dataset: 6; Right Dataset: 10; Both Dataset: 14
//Determines which dataset to use. Either lighting datasets or camera datasets
if(!usingCamera)
{
if(light == LIGHT_LEFT)
step = 4;
else if(light == LIGHT_RIGHT)
step = 8;
else if (light ==LIGHT_BOTH)
step=12;
}
else //Camera dataset
step=20;
//Camera: 20 for camera pics
}
else if (dataset==4) //Google Street View
{
//Street view 27,28,29,30
//In matching, we want to match 27-29 and 28-30
//Used for matching the correct images
step = 2;
//Counters used to ensure standard output for processing in Matlab
tempDirCounterkk = 1;
tempDirCounterss = 1;
if(!usingCamera)
{
k_start = 27;
k_end = 28;
}
else
{
k_start = 25;
k_end = 26;
}
}
}
else
{
if (dataset==1)
{
//Counters used to ensure standard output for processing in Matlab
tempDirCounterkk = 1;
tempDirCounterss = 3;
k_start = 1;//Left dataset: 5; Right Dataset: 9; Both Dataset: 13
k_end = 2;//Left dataset: 6; Right Dataset: 10; Both Dataset: 14
s_start = 3;//Left dataset: 7; Right Dataset: 11; Both Dataset: 15
s_end = 4;//Left dataset: 8; Right Dataset: 12; Both Dataset: 16
}
else if (dataset==2)
{
//Counters used to ensure standard output for processing in Matlab
tempDirCounterkk = 5;
tempDirCounterss = 6;
k_start = 19; //Dataset 2: 17; Dataset 3: 19
k_end = 19; //Dataset 2: 17; Dataset 3: 19
s_start = 20; //Dataset 2: 18; Dataset 3: 20
s_end = 20; //Dataset 2: 18; Dataset 3: 20
}
else if(dataset==3) //Varying lighting or comparing the Nao robocup pics with the Nikon Camera pics
{
//s_start; s_end
//Left Light off dataset: 5; Left Light off Dataste 8;
//Right Light off Dataset: 9; Right Light off Dataset: 12;
//Both Light off Dataset: 13; Both Light off Dataset: 16
//Camera Pics Dataset: 21; Camera Pics Dataset: 24
//Counters used to ensure standard output for processing in Matlab
//---------------------------------------------------------------
tempDirCounterkk = 1;
tempDirCounterss = 3;
//Set the datasets that you want to match
k_start = 1;
k_end = 4;
if(!usingCamera){
if(light == LIGHT_LEFT){
lowerBound = 5;
upperBound = 8;
}
else if(light == LIGHT_RIGHT){
lowerBound = 9;
upperBound = 12;
}
else if (light ==LIGHT_BOTH){
lowerBound = 13;
upperBound = 16;
}
}
else{
lowerBound = 21;
upperBound = 24;
}
s_start = lowerBound;
s_end = upperBound;
//REMEMBER: CHANGE s_start, s_end IN THE KK LOOP AS WELL
//---------------------------------------------------------------
}
else if(dataset==4)
{
//Run this twice
//1. First do 27 and 30
//2. Second do 28 and 29
//Counters used to ensure standard output for processing in Matlab
tempDirCounterkk = 1;
tempDirCounterss = 3;
if(!usingCamera)
{
lowerBound = 29;
upperBound = 30;
s_start = 31;
s_end = 32;
}
else
{
lowerBound = 25;
upperBound = 26;
s_start = 27;
s_end = 28;
}
k_start = lowerBound;
k_end = upperBound;
}
}
//Begin the matching procedure
for(int kk=k_start;kk<=k_end;kk++)
{
if(dataset==1){
//The counter must be reset for formatting purposes in Matlab
if(isMatching){
s_start = s_end = kk;
tempDirCounterss = kk;//kk for matching
}
else{
tempDirCounterss = 3;// 3 for non-matching
}
}
else if(dataset==2)
{
if(isMatching)
s_start = s_end = kk;
}
else if(dataset==3){ //This is for the varying illumination datasets
if(isMatching)
{
//Comparing datasets
s_start = s_end = kk+step;
//The counter must be reset for formatting purposes in Matlab
tempDirCounterss = kk;
}
else
{
//This ensures that no datasets are repeated
if (kk>=3){
tempDirCounterss = 5;
s_start = lowerBound;
s_end = lowerBound+1;
}
else{
tempDirCounterss = 3;
s_start = upperBound - 1;
s_end = upperBound;
}
}
}
else if (dataset==4)//This is for google street view
{
if(isMatching)
s_start = s_end = kk+ step;
else{
if (kk==lowerBound)
s_start = s_end = lowerBound + 3;
else if (kk==upperBound)
s_start = s_end = upperBound+1;
}
}
for (int ss = s_start;ss<=s_end;ss++)
{
//Create object for dataAnalysis
DataAnalysis dataAnalysis;
//Create the Feature extraction object
FeatureExtraction feature(angle, distance, usingKnnCriterion);
//The directory where the files are stored
std::string dir, dir1;
dir = to_string<int>(kk);
dir1 = to_string<int>(ss);
std::string tempDir = to_string<int>(tempDirCounterkk);
std::string tempDir1 =to_string<int>(tempDirCounterss);
//Names of the two image files
std::string name1;
std::string name2;
//For changing the threshold
double testThreshold = 0;
//Set the directory names and determine the number of images in each directory
int jpegCounter = dataAnalysis.getNumImagesInDirectory(&dir, terminal);
int jpegCounter1 = dataAnalysis.getNumImagesInDirectory(&dir1, terminal);
std::cout<<"The number of images in the directory is: "<<jpegCounter<<endl;
//Set the flag for1 setting up the BRISK lookup table for the extractor
cv::Ptr<cv::DescriptorExtractor> descriptorExtractor;
descriptorExtractor = feature.getExtractor(argc, feat_descriptor, hamming, descriptorExtractor);
//*************************************
//Get the current time for saving purposes
//*************************************
time_t rawtime;
struct tm * timeinfo;
//char filename [80];
time ( &rawtime );
timeinfo = localtime ( &rawtime );
//strftime (filename,80,"../../data/Matches/matchingData_%b_%d_%H%M%S.txt",timeinfo);
//strftime (filename,80,"../data/Matches/nonmatching_matching_Data__BRISK__BRISK_Hamming_070421012_1222.txt",timeinfo);
//puts (filename);
string filename = "../";
if (dataset==1)
{
filename.append("dataRobocup/Matches/");
filename.append("nonmatching_matching_Data__");
}else if (dataset==2)
{
filename.append("data2/Matches/");//Or data3
filename.append("dataset2_nonmatching_matching_Data__");//Or dataset3
}
else if (dataset==3 && !usingCamera)
{
filename.append("dataLighting/Matches/");
if(light==LIGHT_LEFT)
filename.append("dataLighting_left_light_off_nonmatching_matching_Data__");
else if(light==LIGHT_RIGHT)
filename.append("dataLighting_right_light_off_nonmatching_matching_Data__");
else if (light==LIGHT_BOTH)
filename.append("dataLighting_both_lights_off_nonmatching_matching_Data__");
}
else if (dataset==3 && usingCamera){
filename.append("dataCamera/Matches/");
filename.append("camera_nonmatching_matching_Data__");
}
else if (dataset==4){
if(usingCamera){
filename.append("dataCamera/Matches/");
filename.append("camera_large_hall_nonmatching_matching_Data__");
}
else{
filename.append("dataStreetView/Matches/");
filename.append("streetview_nonmatching_matching_Data__");
}
}
filename.append(feat_detector);
filename.append("_");
filename.append(feat_descriptor);
filename.append("_");
if(usingKnnCriterion)
filename.append("KNN_");
else
filename.append("Hamming_");
filename.append(myDate);
filename.append("_");
filename.append(myTime);
filename.append("_");
filename.append(to_string<double>(threshold));
filename.append("_");
if (!usingKnnCriterion)
filename.append(to_string<int>(hammingDistance));
if (isMax)
filename.append("_max");
else
filename.append("_consistent");
filename.append(".txt");
cout<<filename<<endl;
//*************************************
//Make sure that there are the same number of images in each frame (Not for non matches)
if (isMatching && !(dataset==3))
{
if(jpegCounter>jpegCounter1)
jpegCounter = jpegCounter1;
else
jpegCounter1 = jpegCounter;
}
//Remember that for non-matches, we can compare 1,1;2,2;3,3...etc
//Determine matches without repetition
for (int ii = 1;ii<=jpegCounter;ii++)
{
if(isMatching && !(dataset ==3))
index = ii;
else
index = jpegCounter1;
for (int jj = 1; jj<index;jj++)//jpegCounter1
{
//Choose the images to compare
name1 = to_string<int>(ii);
name2 = to_string<int>(jj);
cout<<"Image "<<ii<<", Image "<<jj<<endl;
// names of the two images
std::string fname1;
std::string fname2;
cv::Mat imgGray1Full;
cv::Mat imgGray2Full;
cv::Mat imgGray1;
cv::Mat imgGray2;
//Process the first image which would normally be stored on the robot
//****************************************************
while(imgGray1Full.empty()){
fname1 = dir+"/"+name1+".jpg";
imgGray1Full = cv::imread(fname1.c_str(), CV_LOAD_IMAGE_GRAYSCALE);
}
if (imgGray1Full.empty())
{
std::cout<<"image(s)"<<fname1<<" not found." << std::endl;
return 2;
}
//Only take the image from the horizon line
imgGray1 = imgGray1Full(cv::Rect(0, 0, imgGray1Full.cols, horizonLine));
//****************************************************
timespec ts, te, matchings, matchinge, detectors, detectore, extractors, extractore, verifys, verifye;
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts);
//Read in images
//*****************************************************************
while(imgGray2Full.empty()){
fname2 = dir1+"/"+name2+".jpg";
imgGray2Full = cv::imread(fname2.c_str(), CV_LOAD_IMAGE_GRAYSCALE);
}
if (imgGray2Full.empty())
{
std::cout<<"image(s)"<<fname2<<" not found." << std::endl;
return 2;
}
//We only need the keypoints above the horizon
imgGray2 = imgGray2Full(cv::Rect(0, 0, imgGray2Full.cols, horizonLine));
//*****************************************************************
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &te);
float imageProcessingTime = diff(ts,te).tv_nsec/1000000.0f;
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &detectors);
//MC: Generate a vector of keypoints
std::vector<cv::KeyPoint> keypoints, keypoints2;
// create the detector:
//*****************************************************************
cv::Ptr<cv::FeatureDetector> detector;
//Note: Testflag==1 which means that the threshold will equal testThreshold
detector = feature.getDetector(argc, feat_detector, detector, threshold, testThreshold,1);
//*****************************************************************
detector->detect(imgGray2,keypoints2);
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &detectore);
float detectionTime = diff(detectors,detectore).tv_nsec/1000000.0f;
//The descriptor stored on the robot
//*****************************************************************
detector->detect(imgGray1,keypoints);
//*****************************************************************
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &extractors);
// now the extractor:
//*****************************************************************
// get the descriptors
cv::Mat descriptors, descriptors2;
std::vector<cv::DMatch> indices;
// first image. Computes the descrkeypoints2iptor for each of the keypoints.
//Outputs a 64 bit vector describing the keypoints.
descriptorExtractor->compute(imgGray2,keypoints2,descriptors2);
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &extractore);
float extractionTime = diff(extractors,extractore).tv_nsec/1000000.0f;
//The descriptors stored on the robot
//*****************************************************************
descriptorExtractor->compute(imgGray1,keypoints,descriptors);
//*****************************************************************
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &matchings);
// matching
//*****************************************************************
std::vector<std::vector<cv::DMatch> > matches;
cv::Ptr<cv::DescriptorMatcher> descriptorMatcher;
if(hamming)
descriptorMatcher = new cv::BruteForceMatcher<cv::HammingSse>();
else
descriptorMatcher = new cv::BruteForceMatcher<cv::L2<float> >();
if(hamming){
if(usingKnnCriterion)
descriptorMatcher->knnMatch(descriptors,descriptors2,matches,2);
else
descriptorMatcher->radiusMatch(descriptors,descriptors2,matches,hammingDistance);
}
else{
if(usingKnnCriterion)
{
if(descriptors2.rows>0)
descriptorMatcher->knnMatch(descriptors,descriptors2,matches,2);
else
matches.clear();
}
else{
//Decreasing with the maxdistance value will drastically reduce the number of matches
if(descriptors2.rows>0)
descriptorMatcher->radiusMatch(descriptors,descriptors2,matches,hammingDistance);
else
matches.clear();
}
}
cv::Mat outimg;
//Write the data to a file
ofstream writeFile;
//Create the filename with the current time
writeFile.open(filename.c_str(), ios::app);//ios::app
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &verifys);
//Perform the matching verification
//*************************************************************************
// cout<<"The total number of keypoints in image 1 is: "<<keypoints.size()<<endl;
// cout<<"The total number of keypoints in image 2 is: "<<keypoints2.size()<<endl;
// cout<<"The total number of matches is: "<<matches.size()<<endl;
feature.performMatchingValidation(imgGray1,keypoints, keypoints2, matches, hamming);
//*************************************************************************
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &verifye);
float verifyTime = diff(verifys,verifye).tv_nsec/1000000.0f;
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &matchinge);
float matchingTime = diff(matchings,matchinge).tv_nsec/1000000.0f;
float overallTime = imageProcessingTime + detectionTime + extractionTime + matchingTime + verifyTime;
#if (DEBUG_MODE)
cout<<"****************************************"<<endl;
cout<<"The matching score for the image (condsidering all matches) is "<<feature.imageMatchingScore<<endl;
cout<<"The matching score for the image (condsidering best match only) is "<<feature.imageMatchingScoreBest<<endl;
cout<<"The total number of matches is "<<feature.totalNumMatches<<endl;
cout<<"The total number of best matches is "<<feature.totalNumBestMatches<<endl;
cout<<"****************************************"<<endl;
#endif
#if (DEBUG_TIMES)
std::cout<<"The times:"<<endl;
std::cout<<"Detection Time: "<<detectionTime<<" ms"<<endl;
std::cout<<"Extraction Time: "<<extractionTime<<" ms"<<endl;
std::cout<<"Matching Time: "<<matchingTime<<" ms"<<endl;
std::cout<<"Verify Matches Time: "<<verifyTime<<" ms"<<endl;
std::cout<<"Overall Time: "<<overallTime<<" ms"<<endl;
cv::waitKey(1000);
#endif
//threshold = atoi(argv[3]+5);
//Write all the information to a file
writeFile <<tempDirCounterkk<<", "<<tempDirCounterss<<", "<<threshold<<", "<<name1<<", "<<name2<<", "<<keypoints.size()<<", "<<keypoints2.size()<<", "<<feature.imageMatchingScoreBest<<", "<<feature.imageMatchingScore<<","<<feature.totalNumMatches<<", "<<feature.totalNumValidMatches<<", "<<feature.totalNumInvalidMatches<<", "<<feature.totalNumBestMatches<<", "<<detectionTime<<", "<<extractionTime<<", "<<matchingTime<<", "<<verifyTime<<", "<<overallTime<<"\n";
//close the file
writeFile.close();
#if (DISPLAY)
drawMatches(imgGray1, keypoints, imgGray2, keypoints2,matches,outimg,
cv::Scalar(0,255,0), cv::Scalar(0,0,255),
std::vector<std::vector<char> >(), cv::DrawMatchesFlags::DRAW_RICH_KEYPOINTS );
cv::namedWindow("Matches");
cv::imshow("Matches", outimg);
//cv::imshow("keypoints", imgRGB1);
//cv::imshow("keypoints2", imgRGB2);
cv::waitKey();
#endif
}//End of inner for loop (jj)
}//end of outer for loop (ii)
tempDirCounterss++;
}//End of ss loop
tempDirCounterkk++;
}//end of kk loop
return 0;
}
//Main function
int main(int argc, char ** argv) {
//The angle used for matching validation
double angle = 10;
//The distance threshold
double distance = 200;
//The horizon line
int horizonLine = 300;
//Determine if the KNN validation criterion is necessary
bool usingKnnCriterion = true;
//To determine whether we are finding keypoint properties or KNN Ratio
bool keypointProperties = true;
//Set if you are matching or not matchingg
bool isMatching = true;
//Set the date and time
string myDate = "11072012";
string myTime = "2300";
//Using the terminal
bool terminal = false;
//BRISK SURF2D
bool hamming=false;
std::string feat_detector = "BRISK";
int threshold = 43.75;
int hammingDistance = 40;//BRISK BRISK
//double radius = 0.15;//BRISK SURF
std::string feat_descriptor = "SURF";
//Set the arguments
// bool hamming=true;
// std::string feat_detector = "BRISK";
// int threshold = 55;
// int hammingDistance = 40;//BRISK BRISK
// //double radius = 0.15;//BRISK SURF
// std::string feat_descriptor = "U-BRISK";
//Create the Feature extraction object
FeatureExtraction feature(angle, distance, usingKnnCriterion);
//Create data analysis object
DataAnalysis dataAnalysis;
//*************************************
//Get the current time for saving purposes and set the file to save to
//*************************************
time_t rawtime;
struct tm * timeinfo;
char filename [80];
string testThresholdString = to_string<int>(threshold);
time ( &rawtime );
timeinfo = localtime ( &rawtime );
//Declare the files
string file,file1, file2, file3;
if(terminal)
{
file.append("../");
file1.append("../");
file2.append("../");
file3.append("../");
}
file = "../data/singleImage/singleImageData_keypoints_";
file.append(feat_detector);
file.append("_");
file.append(feat_descriptor);
file.append("_");
if(usingKnnCriterion)
file.append("KNN_");
else
file.append("Hamming_");
file.append(myDate);
file.append("_");
file.append(myTime);
file1 = "../data/singleImage/singleImageData_matches_";
file1.append(feat_detector);
file1.append("_");
file1.append(feat_descriptor);
file1.append("_");
if(usingKnnCriterion)
file1.append("KNN_");
else
file1.append("Hamming_");
file1.append(myDate);
file1.append("_");
file1.append(myTime);
file2 = "../data/singleImage/singleImageData_invalid_matches_KNN_";
file2.append(feat_detector);
file2.append("_");
file2.append(feat_descriptor);
file2.append("_");
if(usingKnnCriterion)
file2.append("KNN_");
else
file2.append("Hamming_");
file2.append(myDate);
file2.append("_");
file2.append(myTime);
file3 = "../data/singleImage/singleImageData_invalid_matches_AngleDistance_";
file3.append(feat_detector);
file3.append("_");
file3.append(feat_descriptor);
file3.append("_");
if(usingKnnCriterion)
file3.append("KNN_");
else
file3.append("Hamming_");
file3.append(myDate);
file3.append("_");
file3.append(myTime);
if(keypointProperties)
{
file.append("_keypointStats");
file1.append("_keypointStats");
file2.append("_keypointStats");
file3.append("_keypointStats");
}
else{
if(!isMatching)
{
file.append("_false");
file1.append("_false");
file2.append("_false");
file3.append("_false");
}
else
{
file.append("_identical");
file1.append("_identical");
file2.append("_identical");
file3.append("_identical");
}
}
file.append(".txt");
cout<<file<<endl;
file1.append(".txt");
cout<<file1<<endl;
file2.append(".txt");
cout<<file2<<endl;
file3.append(".txt");
cout<<file3<<endl;
//*************************************
//Matching parameters
int k_start =0;
int k_end = 0;
int s_start = 0;
int s_end = 0;
int index = 0;
if(isMatching)
{
k_start = 1;
k_end = 4;
}
else
{
k_start = 1;
k_end = 2;
s_start = 3;
s_end = 4;
}
for (int kk=k_start;kk<=k_end; kk++)
{
if(isMatching)
s_start = s_end = kk;
for (int ss=s_start;ss<=s_end;ss++)
{
std::string dir, dir1;
dir = to_string<int>(kk);
dir1 = to_string<int>(ss);
std::string tempDir = to_string<int>(kk);
std::string tempDir1 = to_string<int>(ss);
//Names of the two image files
std::string name1;
std::string name2;
//For changing the threshold
int testThreshold = 10;
//Set the directory names and determine the number of images in each directory
int jpegCounter = dataAnalysis.getNumImagesInDirectory(&dir, terminal);
int jpegCounter1 = dataAnalysis.getNumImagesInDirectory(&dir1, terminal);
std::cout<<"The number of images in the directory is: "<<jpegCounter<<endl;
// Declare the extractor. Only needs to be performed once since it computes lookup
//tables for each of the various patterns on initialisation
//*****************************************************************
cv::Ptr<cv::DescriptorExtractor> descriptorExtractor;
descriptorExtractor = feature.getExtractor(argc, feat_descriptor, hamming, descriptorExtractor);
//*****************************************************************
//Make sure that there are the same number of images in each frame
if (isMatching)
{
jpegCounter=5;
// if(jpegCounter>jpegCounter1)
// jpegCounter = jpegCounter1;
// else
// jpegCounter1 = jpegCounter;
}
//Remember that for non-matches, we can compare 1,1;2,2;3,3...etc
//Determine matches without repetition
for (int ii = 1;ii<=jpegCounter;ii++)
{
if(isMatching)
index = ii;
else
index = jpegCounter1;
for (int jj = 1; jj<=index;jj++)//if int jj = index; jj<=index;jj++ then matching images to themselves
{
//Choose the images to compare
name1 = to_string<int>(ii);
name2 = to_string<int>(jj);
cout<<"Image "<<ii<<", Image "<<jj<<endl;
//For changing the threshold
int testThreshold = 10;
//cout<<"Image in directory 1: "<<name1<<", Image in directory 2: "<<name2<<endl;
// names of the two images
std::string fname1;
std::string fname2;
cv::Mat imgGray1Full;
cv::Mat imgGray2Full;
cv::Mat imgGray1;
cv::Mat imgGray2;
//Process the first image
//****************************************************************
while(imgGray1Full.empty()){
fname1 = dir+"/"+name1+".jpg";
imgGray1Full = cv::imread(fname1.c_str(), CV_LOAD_IMAGE_GRAYSCALE);
}
if (imgGray1Full.empty())
{
std::cout<<"image(s)"<<fname1<<" not found." << std::endl;
return 2;
}
//Process only the image above the horizon line
imgGray1 = imgGray1Full(cv::Rect(0, 0, imgGray1Full.cols, horizonLine));
timespec ts, te, matchings, matchinge, detectors, detectore, extractors, extractore, verifys, verifye;
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts);
// if no arguments are passed:
//Read in images
//*****************************************************************
while(imgGray2Full.empty()){
fname2 = dir1+"/"+name2+".jpg";
imgGray2Full = cv::imread(fname2.c_str(), CV_LOAD_IMAGE_GRAYSCALE);
}
if (imgGray2Full.empty())
{
std::cout<<"image(s)"<<fname2<<" not found." << std::endl;
return 2;
}
//*****************************************************************
//We only need the keypoints above the horizon
imgGray2 = imgGray2Full(cv::Rect(0, 0, imgGray2Full.cols, horizonLine));
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &te);
float imageProcessingTime = diff(ts,te).tv_nsec/1000000.0f;
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &detectors);
//MC: Generate a vector of keypoints
std::vector<cv::KeyPoint> keypoints, keypoints2;
// create the detector:
//*****************************************************************
cv::Ptr<cv::FeatureDetector> detector;
detector = feature.getDetector(argc, feat_detector, detector, threshold, testThreshold,1);
//*****************************************************************
detector->detect(imgGray2,keypoints2);
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &detectore);
float detectionTime = diff(detectors,detectore).tv_nsec/1000000.0f;
// run the detector:
//*****************************************************************
detector->detect(imgGray1,keypoints);
//*****************************************************************
//*****************************************************************
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &extractors);
// get the descriptors
cv::Mat descriptors, descriptors2;
std::vector<cv::DMatch> indices;
// second image. Computes the descriptor for each of the keypoints.
//Outputs a 64 bit vector describing the keypoints.
//*****************************************************************
descriptorExtractor->compute(imgGray2,keypoints2,descriptors2);
//*****************************************************************
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &extractore);
float extractionTime = diff(extractors,extractore).tv_nsec/1000000.0f;
//Compute the descriptors for the stored image
//*****************************************************************
descriptorExtractor->compute(imgGray1,keypoints,descriptors);
//*****************************************************************
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &matchings);
// matching
//*****************************************************************
std::vector<std::vector<cv::DMatch> > matches;
cv::Ptr<cv::DescriptorMatcher> descriptorMatcher;
if(hamming){
//descriptorMatcher = new cv::BruteForceMatcher<cv::HammingSse>();
}
else
descriptorMatcher = new cv::BruteForceMatcher<cv::L2<float> >();
if(hamming)
//descriptorMatcher->radiusMatch(descriptors,descriptors2,matches,hammingDistance);
//The first parameter is the query descriptor. The second parameter is the train descriptor
descriptorMatcher->knnMatch(descriptors,descriptors2,matches,2);
else{
//Messing with the maxdistance value will drastically reduce the number of matches
descriptorMatcher->knnMatch(descriptors,descriptors2,matches,2);//radiusMatch radius
}
//For the above method, we could use KnnMatch. All values less than 0.21 max distance are selected
//*****************************************************************
//Image created for drawing
cv::Mat outimg;
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &verifys);
//Perform the matching verification
//*************************************************************************
feature.performMatchingValidation(imgGray1,keypoints, keypoints2, matches, hamming);
//*************************************************************************
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &verifye);
float verifyTime = diff(verifys,verifye).tv_nsec/1000000.0f;
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &matchinge);
float matchingTime = diff(matchings,matchinge).tv_nsec/1000000.0f;
float overallTime = imageProcessingTime + detectionTime + extractionTime + matchingTime + verifyTime;
//Write the data to a file
ofstream writeFile;
writeFile.open(file.c_str(), ios::app);
//1. Write all the keypoints to the first image file
//Swap hamming distance with radius
for(int im1 = 0;im1<keypoints.size();im1++){
//[image num, hammingdist, threshold, kp1x, kp1y, angle, size, octave]
writeFile <<kk<<", "<<ss<<", "<<name1<<", "<<hammingDistance<<", "<<threshold<<", "<<im1<<", "<<keypoints[im1].pt.x<<", "<<keypoints[im1].pt.y<<", "<<keypoints[im1].angle<<", "<<keypoints[im1].size<<", "<<keypoints[im1].octave<<", "<<keypoints[im1].response<<"\n";
}
//2. Write all the keypoints to the second image file
//Swap hamming distance with radius
for(int im2 = 0;im2<keypoints2.size();im2++){
//[image num, hammingdist, threshold, kp1x, kp1y, angle, size, octave]
writeFile <<kk<<", "<<ss<<", "<<name2<<", "<<hammingDistance<<", "<<threshold<<", "<<im2<<", "<<keypoints2[im2].pt.x<<", "<<keypoints2[im2].pt.y<<", "<<keypoints2[im2].angle<<", "<<keypoints2[im2].size<<", "<<keypoints2[im2].octave<<", "<<keypoints2[im2].response<<"\n";
}
//Close the file
writeFile.close();
writeFile.clear();
//Write the data to a file
writeFile.open(file1.c_str(), ios::app);
//Write all the matches data to a file
//image num left, image num right, queryIdx, trainIdx, keypoint1 x, kp1y, angle size, octave, response, kp2x, kp2y, angle, size, octave, neighbor num, distance
//Swap hamming distance with radius
for(size_t mIdx = 0;mIdx<matches.size();mIdx++){
for(size_t nIdx = 0;nIdx<matches[mIdx].size();nIdx++)
{
int q = matches[mIdx][nIdx].queryIdx;
int t = matches[mIdx][nIdx].trainIdx;
#if (DEBUG_MODE)
cout<<"neighbor index: "<<nIdx<<endl;
cout<<"Matches size: "<<matches[mIdx].size()<<endl;
cout<<"Keypoint Left to be erased row,col : "<<keypoints[q].pt.y<<", "<<keypoints[q].pt.x<<endl;
cout<<"Keypoint Right to be erased row,col : "<<keypoints2[t].pt.y<<", "<<keypoints2[t].pt.x<<endl;
cout<<"The match distance is: "<<matches[mIdx][nIdx].distance<<endl;
#endif
writeFile <<kk<<", "<<ss<<", "<<name1<<", "<<name2<<", "<<q<<", "<<t<<", "<<keypoints[q].pt.x<<", "<<keypoints[q].pt.y<<", "<<keypoints[q].angle<<", "<<keypoints[q].size<<", "<<keypoints[q].octave<<", "<<keypoints[q].response<<", "<<keypoints2[t].pt.x<<", "<<keypoints2[t].pt.y<<", "<<keypoints2[t].angle<<", "<<keypoints2[t].size<<", "<<keypoints2[t].octave<<", "<<keypoints2[t].response<<", "<<nIdx<<", "<<matches[mIdx][nIdx].distance<<"\n";
}
}
//Close the file
writeFile.close();
writeFile.clear();
//Write the invalid KNN matches data to a file
writeFile.open(file2.c_str(), ios::app);
for (size_t it = 0; it<feature.leftPointsKNN.size(); it++)
{
// cout<<"it index: "<<it<<endl;
// cout<<"Matches size: "<<feature.leftPointsKNN.size()<<endl;
// cout<<"Keypoint Left to be erased row,col : "<<feature.leftPointsKNN[it].pt.y<<", "<<feature.leftPointsKNN[it].pt.x<<endl;
// cout<<"Keypoint Right to be erased row,col : "<<feature.rightPointsKNN[it].pt.y<<", "<<feature.rightPointsKNN[it].pt.x<<endl;
// cout<<"The match distance is: "<<feature.keypointDistanceKNN[it]<<endl;
writeFile <<kk<<", "<<ss<<", "<<name1<<", "<<name2<<", "<<feature.leftPointsKNN[it].pt.x<<", "<<feature.leftPointsKNN[it].pt.y<<", "<<feature.leftPointsKNN[it].angle<<", "<<feature.leftPointsKNN[it].size<<", "<<feature.leftPointsKNN[it].octave<<", "<<feature.leftPointsKNN[it].response<<", "<<feature.rightPointsKNN[it].pt.x<<", "<<feature.rightPointsKNN[it].pt.y<<", "<<feature.rightPointsKNN[it].angle<<", "<<feature.rightPointsKNN[it].size<<", "<<feature.rightPointsKNN[it].octave<<", "<<feature.rightPointsKNN[it].response<<", "<<feature.neighborIndexKNN[it]<<", "<<feature.keypointDistanceKNN[it]<<"\n";
}
//Close the file
writeFile.close();
writeFile.clear();
//Write the invalid angle and distance matches data to a file
writeFile.open(file3.c_str(), ios::app);
for (int it = 0; it<feature.leftPointsAngle.size(); it++)
{
writeFile <<kk<<", "<<ss<<", "<<name1<<", "<<name2<<", "<<feature.leftPointsAngle[it].pt.x<<", "<<feature.leftPointsAngle[it].pt.y<<", "<<feature.leftPointsAngle[it].angle<<", "<<feature.leftPointsAngle[it].size<<", "<<feature.leftPointsAngle[it].octave<<", "<<feature.leftPointsAngle[it].response<<", "<<feature.rightPointsAngle[it].pt.x<<", "<<feature.rightPointsAngle[it].pt.y<<", "<<feature.rightPointsAngle[it].angle<<", "<<feature.rightPointsAngle[it].size<<", "<<feature.rightPointsAngle[it].octave<<", "<<feature.rightPointsAngle[it].response<<", "<<feature.keypointDistanceAngle[it]<<"\n";
}
//Close the file
writeFile.close();
writeFile.clear();
#if (DEBUG_MODE)
cout<<"****************************************"<<endl;
cout<<"The matching score for the image (condsidering all matches) is "<<feature.imageMatchingScore<<endl;
cout<<"The matching score for the image (condsidering best match only) is "<<feature.imageMatchingScoreBest<<endl;
cout<<"The total number of matches is "<<feature.totalNumMatches<<endl;
cout<<"The total number of valid matches is "<<feature.totalNumValidMatches<<endl;
cout<<"The total number of invalid matches is "<<feature.totalNumInvalidMatches<<endl;
cout<<"****************************************"<<endl;
#endif
#if (DEBUG_TIMES)
std::cout<<"The times:"<<endl;
std::cout<<"Detection Time: "<<detectionTime<<" ms"<<endl;
std::cout<<"Extraction Time: "<<extractionTime<<" ms"<<endl;
std::cout<<"Matching Time: "<<matchingTime<<" us"<<endl;
std::cout<<"Verify Matches Time: "<<verifyTime<<" us"<<endl;
std::cout<<"Overall Time: "<<overallTime<<" ms"<<endl;
#endif
#if (DEBUG_MODE)
cout<<"The total number of keypoints in image 1 is: "<<keypoints.size()<<endl;
cout<<"The total number of keypoints in image 2 is: "<<keypoints2.size()<<endl;
#endif
#if (DISPLAY)
drawMatches(imgGray1, keypoints, imgGray2, keypoints2,matches,outimg,
cv::Scalar(0,255,0), cv::Scalar(0,0,255),
std::vector<std::vector<char> >(), cv::DrawMatchesFlags::DRAW_RICH_KEYPOINTS );
//NOT_DRAW_SINGLE_POINTS
//Note: leftpoints correspond to keypoints - Image 1. rightpoints correspond to keypoints2 - Image 2.
int colourChanger = 0;
// for (int k = 0; k<feature.rightPoints.size(); k++)
// {
//
// circle(imgGray2,cv::Point(feature.rightPoints[k].x, feature.rightPoints[k].y), 5, cv::Scalar(colourChanger, 100, 255), 4, 8, 0);
//
//#if(DEBUG_MATCHES)
// cout<<"Incorrect coord Left row,col : "<<feature.rightPoints[k].y<<", "<<feature.rightPoints[k].x<<endl;
//#endif
// colourChanger = colourChanger+30;
// }
// colourChanger = 0;
// for (int k = 0; k<feature.leftPoints.size(); k++)
// {
// circle(imgGray1,cv::Point(feature.leftPoints[k].x, feature.leftPoints[k].y), 5, cv::Scalar(colourChanger, 100, 255), 4, 8, 0);
//#if(DEBUG_MATCHES)
// cout<<"Incorrect coord Right row,col : "<<feature.leftPoints[k].y<<", "<<feature.leftPoints[k].x<<endl;
//#endif
// colourChanger = colourChanger+30;
// }
for (int k = 0; k<matches.size(); k++)
{
for (int j=0;j<matches[k].size();j++)
{
int qi = matches[k][j].queryIdx;//Reference Point. Assumed to be for image left
int ti = matches[k][j].trainIdx;//Assumed to be image right
//The points in the left image
int kp1x = (*(keypoints.begin()+qi)).pt.x;
int kp1y = (*(keypoints.begin()+qi)).pt.y;
//The points in the right image
int kp2x = (*(keypoints2.begin()+ti)).pt.x;
int kp2y = (*(keypoints2.begin()+ti)).pt.y;
circle(outimg,cv::Point(kp1x, kp1y), 5, cv::Scalar(colourChanger, 100, 255), 4, 8, 0);
circle(outimg,cv::Point(640+ kp2x, kp2y), 5, cv::Scalar(colourChanger, 100, 255), 4, 8, 0);
#if(DEBUG_MODE)
cout<<"Correct coord Left row,col : "<<kp1y<<", "<<kp1x<<endl;
cout<<"Correct coord Right row,col : "<<kp2y<<", "<<kp2x<<endl;
#endif
}
colourChanger = colourChanger+30;
}
cv::namedWindow("Matches");
cv::imshow("Matches", outimg);
//imgRGB1 is right. imgRGB2 is left
#if(DEBUG_MODE)
cv::imshow("keypoints", imgGray1);
cv::imshow("keypoints2", imgGray2);
#endif
cv::waitKey();
#endif
}//end of jj loop
}//end of ii loop
}//End of ss loop
}//End of the kk loop
return 0;
}
