int TrainingSet::getTrainingSet(cv::vector<cv::Mat> &images, cv::vector<int> &labels, cv::vector<QString> &names){
QStringList directoriesList = getAvailableFaces();
labels.clear();
images.clear();
names.clear();
for(int i=0; i<directoriesList.size(); i++){
QString person = directoriesList.at(i);
QDir faceDirectory;
faceDirectory.setPath(mainDirectory.path() + "/" + person);
QStringList photosList = faceDirectory.entryList(QDir::Files);
foreach(QString photo, photosList){
images.push_back(cv::imread(faceDirectory.path().toStdString() + "/" + photo.toStdString(), CV_LOAD_IMAGE_GRAYSCALE));
labels.push_back(i);
}
names.push_back(person);
}
//ランダムに6点を抽出
void get_six_points(cv::vector<cv::Point2d>& calib_p, cv::vector<cv::Point3d>& calib_P, cv::vector<cv::Point2d>& src_p, cv::vector<cv::Point3d>& src_P)
{
int i=0;
srand(time(NULL)); /* 乱数の初期化 */
cv::Vector<int> exists;
while(i <= 6){
int maxValue = (int)src_p.size();
int v = rand() % maxValue;
bool e2=false;
for(int s=0; s<i; s++){
if(exists[s] == v) e2 = true;
}
if(!e2){
exists.push_back(v);
calib_P.push_back(src_P[v]);
calib_p.push_back(src_p[v]);
i++;
}
}
}
void loadImages(cv::vector<cv::Mat> &rgb,
cv::vector<cv::Mat> &depth,
const int &fileNum){
rgb.clear();
depth.clear();
for(int i = 0; i < fileNum; ++i){
stringstream rgbfilename, depthfilename;
rgbfilename << FLAGS_folder <<"/" << FLAGS_color << i << FLAGS_type;
depthfilename << FLAGS_folder << "/" << FLAGS_depth << i << FLAGS_type;
cout << "loading : " << rgbfilename.str() << " and " << depthfilename.str() << endl;
// load RGB image
cv::Mat tempRGB = cv::imread(rgbfilename.str(), 0);
rgb.push_back(tempRGB);
// load depth image
cv::Mat tempDepth = cv::imread(depthfilename.str(), CV_LOAD_IMAGE_ANYDEPTH);
tempDepth.convertTo(tempDepth, CV_8U, 255.0/1000.0);
// cv::Mat maxDist = cv::Mat::ones(tempDepth.rows, tempDepth.cols, CV_8U) * MAX_DEPTH;
// cv::Mat minDist = cv::Mat::ones(tempDepth.rows, tempDepth.cols, CV_8U) * MIN_DEPTH;
// cv::min(tempDepth, maxDist, tempDepth);
// tempDepth -= minDist;
cv::resize(tempDepth, tempDepth, cv::Size(), 2.0,2.0);
cv::Mat roiTempDepth;
cv::resize(tempDepth(cv::Rect(40, 43,498,498 / 4 * 3)), roiTempDepth, cv::Size(640, 480));
depth.push_back(roiTempDepth);
std::cout << "loaded" << std::endl;
cv::imshow("rgb",rgb[i]);
cv::imshow("depth",depth[i]);
cv::waitKey(100);
}
}
void loadImages(cv::vector<cv::Mat> &lefts, cv::vector<cv::Mat> &rights,
const int &fileNum) {
cv::namedWindow("Left", CV_WINDOW_AUTOSIZE | CV_WINDOW_FREERATIO);
cv::namedWindow("Right", CV_WINDOW_AUTOSIZE | CV_WINDOW_FREERATIO);
lefts.clear();
rights.clear();
for (int i = 0; i < fileNum; i++) {
std::stringstream lfile, rfile;
lfile << FLAGS_dir << "/left_" << i << FLAGS_suffix;
rfile << FLAGS_dir << "/right_" << i << FLAGS_suffix;
std::cout << "load: " << lfile.str() << ", " << rfile.str() << std::endl;
cv::Mat left = cv::imread(lfile.str(), 0);
cv::Mat right = cv::imread(rfile.str(), 0);
lefts.push_back(left);
rights.push_back(right);
cv::imshow("Left", left);
cv::imshow("Right", right);
cv::waitKey(100);
}
}
void SiftGPUWrapper::detect(const cv::Mat& image, cv::vector<cv::KeyPoint>& keypoints, std::vector<float>& descriptors, const Mat& mask) const {
if (error) {
keypoints.clear();
ROS_FATAL("SiftGPU cannot be used. Detection of keypoints failed");
}
//get image
cvMatToSiftGPU(image, data);
int num_features = 0;
SiftGPU::SiftKeypoint* keys = 0;
ROS_DEBUG("SIFTGPU: cols: %d, rows: %d", image.cols, image.rows);
if (siftgpu->RunSIFT(image.cols, image.rows, data, GL_LUMINANCE, GL_UNSIGNED_BYTE)) {
num_features = siftgpu->GetFeatureNum();
ROS_INFO("Number of features found: %i", num_features);
keys = new SiftGPU::SiftKeypoint[num_features];
descriptors.resize(128 * num_features);
//descriptors = new float[128 * num_features];
siftgpu->GetFeatureVector(&keys[0], &descriptors[0]);
} else {
ROS_WARN("SIFTGPU->RunSIFT() failed!");
}
//copy to opencv structure
keypoints.clear();
for (int i = 0; i < num_features; ++i) {
KeyPoint key(keys[i].x, keys[i].y, keys[i].s, keys[i].o);
keypoints.push_back(key);
}
// FILE *fp = fopen("bla.pgm", "w");
// WritePGM(fp, data, image.cols, image.rows);
// fclose(fp);
}
void paperRegistration::detectFigures(cv::vector<cv::vector<cv::Point>>& squares, cv::vector<cv::vector<cv::Point>>& triangles,
float minLength, float maxLength, int tresh_binary)
{
if (currentDeviceImg.empty())
return;
//cv::Mat image = currentDeviceImg;
//cv::Mat image = cv::imread("C:/Users/sophie/Desktop/meinz.png", CV_LOAD_IMAGE_GRAYSCALE);// cv::imread(path, CV_LOAD_IMAGE_GRAYSCALE);
//resize(image, image, cv::Size(500,700));
squares.clear();
triangles.clear();
cv::Mat gray;
cv::Mat element = getStructuringElement(cv::MORPH_RECT, cv::Size(7,7));
cv::vector<cv::vector<cv::Point> > contours;
//compute binary image
//use dilatation and erosion to improve edges
threshold(currentDeviceImg, gray, tresh_binary, 255, cv::THRESH_BINARY_INV);
dilate(gray, gray, element, cv::Point(-1,-1));
erode(gray, gray, element, cv::Point(-1,-1));
// find contours and store them all as a list
cv::findContours(gray, contours, CV_RETR_LIST, CV_CHAIN_APPROX_SIMPLE);
//test each contour
cv::vector<cv::Point> approx;
cv::vector<cv::vector<cv::Point> >::iterator iterEnd = contours.end();
for(cv::vector<cv::vector<cv::Point> >::iterator iter = contours.begin(); iter != iterEnd; ++iter)
{
// approximate contour with accuracy proportional
// to the contour perimeter
cv::approxPolyDP(*iter, approx, arcLength(*iter, true)*0.03, true);
//contours should be convex
if (isContourConvex(approx))
{
// square contours should have 4 vertices after approximation and
// relatively large length (to filter out noisy contours)
if( approx.size() == 4)
{
bool rectangular = true;
for( int j = 3; j < 6; j++ )
{
// if cosines of all angles are small
// (all angles are ~90 degree) then write
// vertices to result
if (fabs(90 - fabs(computeAngle(approx[j%4], approx[j-3], approx[j-2]))) > 7)
{
rectangular = false;
break;
}
}
if (!rectangular)
continue;
float side1 = computeLength(approx[0], approx[1]);
float side2 = computeLength(approx[1], approx[2]);
if (side1 > minLength && side1 < maxLength &&
side2 > minLength && side2 < maxLength)
squares.push_back(approx);
}
// triangle contours should have 3 vertices after approximation and
// relatively large length (to filter out noisy contours)
else if ( approx.size() == 3)
{
float side1 = computeLength(approx[0], approx[1]);
float side2 = computeLength(approx[1], approx[2]);
float side3 = computeLength(approx[2], approx[0]);
if (side1 > minLength && side1 < maxLength &&
side2 > minLength && side2 < maxLength &&
side3 > minLength && side3 < maxLength)
triangles.push_back(approx);
}
}
}
}