mirror of
https://github.com/YutaItoh/3D-Eye-Tracker.git
synced 2025-11-04 15:39:41 +08:00
Eye center and position can be accessed with: cout << c_end.centre.x() << "," << c_end.centre.y() << "," << c_end.centre.z() << "," << filteredEye.centre[0] << "," << filteredEye.centre[1] << "," << filteredEye.centre[2] << std::endl;
449 lines
15 KiB
C++
449 lines
15 KiB
C++
/** @mainpage Eye position tracker documentation
|
|
|
|
@author Yuta Itoh <itoh@in.tum.de>, \n<a href="http://wwwnavab.in.tum.de/Main/YutaItoh">Homepage</a>.
|
|
|
|
**/
|
|
|
|
|
|
#include <iostream>
|
|
#include <fstream>
|
|
#include <iomanip>
|
|
#include <vector>
|
|
#include <string>
|
|
#include <sstream>
|
|
|
|
|
|
#include "ubitrack_util.h" // claibration file handlers
|
|
#include <boost/foreach.hpp>
|
|
#include <boost/filesystem.hpp>
|
|
#include <boost/filesystem/path.hpp>
|
|
#include <boost/filesystem/fstream.hpp>
|
|
#include <boost/thread.hpp>
|
|
|
|
#include "opencv2/opencv.hpp"
|
|
#include <opencv2/core/core.hpp>
|
|
#include <opencv2/imgproc/imgproc.hpp>
|
|
#include <opencv2/highgui/highgui.hpp>
|
|
#include <opencv2/calib3d/calib3d.hpp>
|
|
#include <opencv2/photo/photo.hpp>
|
|
|
|
|
|
#include "pupilFitter.h" // 2D pupil detector
|
|
#include "timer.h"
|
|
|
|
#include "eye_model_updater.h" // 3D model builder
|
|
#include "eye_cameras.h" // Camera interfaces
|
|
|
|
|
|
|
|
namespace {
|
|
|
|
enum InputMode { CAMERA, CAMERA_MONO, VIDEO, IMAGE };
|
|
|
|
}
|
|
|
|
|
|
int main(int argc, char *argv[]){
|
|
|
|
|
|
// Variables for FPS
|
|
eye_tracker::FrameRateCounter frame_rate_counter;
|
|
|
|
bool kVisualization = false;
|
|
kVisualization = true;
|
|
singleeyefitter::EyeModelFitter::Circle curr_circle;
|
|
|
|
InputMode input_mode =
|
|
//InputMode::VIDEO; // Set a video as a video source
|
|
// InputMode::CAMERA; // Set two cameras as video sources
|
|
InputMode::CAMERA_MONO; // Set a camera as video sources
|
|
// InputMode::IMAGE;// Set an image as a video source
|
|
|
|
|
|
////// Command line opitions /////////////
|
|
std::string kDir = "C:/Users/Yuta/Dropbox/work/Projects/20150427_Alex_EyeTracker/";
|
|
std::string media_file;
|
|
std::string media_file_stem;
|
|
//std::string kOutputDataDirectory(kDir + "out/"); // Data output directroy
|
|
if (argc > 2) {
|
|
boost::filesystem::path file_name = std::string(argv[2]);
|
|
kDir = std::string(argv[1]);
|
|
media_file_stem = file_name.stem().string();
|
|
media_file = kDir + file_name.string();
|
|
//kOutputDataDirectory = kDir + "./";
|
|
std::cout << "Load " << media_file << std::endl;
|
|
std::string media_file_ext = file_name.extension().string();
|
|
|
|
if (media_file_ext == ".avi" ||
|
|
media_file_ext == ".mp4" ||
|
|
media_file_ext == ".wmv") {
|
|
input_mode = InputMode::VIDEO;
|
|
}else{
|
|
input_mode = InputMode::IMAGE;
|
|
}
|
|
}
|
|
else {
|
|
if (input_mode == InputMode::IMAGE || input_mode == InputMode::VIDEO) {
|
|
switch (input_mode)
|
|
{
|
|
case InputMode::IMAGE:
|
|
media_file = kDir + "data3/test.png";
|
|
media_file_stem = "test";
|
|
break;
|
|
case InputMode::VIDEO:
|
|
media_file = kDir + "out/test.avi";
|
|
media_file_stem = "test";
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
///////////////
|
|
|
|
|
|
//// Camera intrinsic parameters
|
|
std::string calib_path="../../docs/cameraintrinsics_eye.txt";
|
|
eye_tracker::UbitrackTextReader<eye_tracker::Caib> ubitrack_calib_text_reader;
|
|
if (ubitrack_calib_text_reader.read(calib_path) == false){
|
|
std::cout << "Calibration file onpen error: " << calib_path << std::endl;
|
|
return -1;
|
|
}
|
|
cv::Mat K; // Camera intrinsic matrix in OpenCV format
|
|
cv::Vec<double, 8> distCoeffs; // (k1 k2 p1 p2 [k3 [k4 k5 k6]]) // k: radial, p: tangential
|
|
ubitrack_calib_text_reader.data_.get_parameters_opencv_default(K, distCoeffs);
|
|
|
|
// Focal distance used in the 3D eye model fitter
|
|
double focal_length = (K.at<double>(0,0)+K.at<double>(1,1))*0.5; // Required for the 3D model fitting
|
|
|
|
|
|
// Set mode parameters
|
|
size_t kCameraNums;
|
|
switch (input_mode)
|
|
{
|
|
case InputMode::IMAGE:
|
|
case InputMode::VIDEO:
|
|
case InputMode::CAMERA_MONO:
|
|
kCameraNums = 1;
|
|
break;
|
|
case InputMode::CAMERA:
|
|
kCameraNums = 2;
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
|
|
|
|
// Setup of classes that handle monocular/stereo camera setups
|
|
// We can encapslate them into a wrapper class in future update
|
|
std::vector<std::unique_ptr<eye_tracker::EyeCameraParent>> eyecams(kCameraNums); // Image sources
|
|
std::vector<std::unique_ptr<eye_tracker::CameraUndistorter>> camera_undistorters(kCameraNums); // Camera undistorters
|
|
std::vector<std::string> window_names(kCameraNums); // Window names
|
|
std::vector<cv::Mat> images(kCameraNums); // buffer images
|
|
std::vector<std::string> file_stems(kCameraNums); // Output file stem names
|
|
std::vector<int> camera_indices(kCameraNums); // Camera indices for Opencv capture
|
|
std::vector<std::unique_ptr<eye_tracker::EyeModelUpdater>> eye_model_updaters(kCameraNums); // 3D eye models
|
|
|
|
// Instantiate and initialize the class vectors
|
|
try{
|
|
switch (input_mode)
|
|
{
|
|
case InputMode::IMAGE:
|
|
eyecams[0] = std::make_unique<eye_tracker::EyeCamera>(media_file, false);
|
|
eye_model_updaters[0] = std::make_unique<eye_tracker::EyeModelUpdater>(focal_length, 5, 0.5);
|
|
camera_undistorters[0] = std::make_unique<eye_tracker::CameraUndistorter>(K, distCoeffs);
|
|
window_names = { "Video/Image" };
|
|
file_stems = { media_file_stem };
|
|
break;
|
|
case InputMode::VIDEO:
|
|
eyecams[0] = std::make_unique<eye_tracker::EyeCamera>(media_file, false);
|
|
eye_model_updaters[0] = std::make_unique<eye_tracker::EyeModelUpdater>(focal_length, 5, 0.5);
|
|
camera_undistorters[0] = std::make_unique<eye_tracker::CameraUndistorter>(K, distCoeffs);
|
|
window_names = { "Video/Image" };
|
|
file_stems = { media_file_stem };
|
|
break;
|
|
case InputMode::CAMERA:
|
|
camera_indices[0] = 0;
|
|
camera_indices[1] = 2;
|
|
#if 0
|
|
// OpenCV HighGUI frame grabber
|
|
eyecams[0] = std::make_unique<eye_tracker::EyeCamera>(camera_indices[0], false);
|
|
eyecams[1] = std::make_unique<eye_tracker::EyeCamera>(camera_indices[1], false);
|
|
#else
|
|
// DirectShow frame grabber
|
|
eyecams[0] = std::make_unique<eye_tracker::EyeCameraDS>("Pupil Cam1 ID1");
|
|
eyecams[1] = std::make_unique<eye_tracker::EyeCameraDS>("Pupil Cam2 ID2");
|
|
#endif
|
|
eye_model_updaters[0] = std::make_unique<eye_tracker::EyeModelUpdater>(focal_length, 5, 0.5);
|
|
eye_model_updaters[1] = std::make_unique<eye_tracker::EyeModelUpdater>(focal_length, 5, 0.5);
|
|
camera_undistorters[0] = std::make_unique<eye_tracker::CameraUndistorter>(K, distCoeffs);
|
|
camera_undistorters[1] = std::make_unique<eye_tracker::CameraUndistorter>(K, distCoeffs);
|
|
window_names = { "Cam0", "Cam1" };
|
|
file_stems = { "cam0", "cam1" };
|
|
break;
|
|
case InputMode::CAMERA_MONO:
|
|
eyecams[0] = std::make_unique<eye_tracker::EyeCameraDS>("Pupil Cam1 ID1"); //
|
|
eye_model_updaters[0] = std::make_unique<eye_tracker::EyeModelUpdater>(focal_length, 5, 0.5);
|
|
camera_undistorters[0] = std::make_unique<eye_tracker::CameraUndistorter>(K, distCoeffs);
|
|
window_names = { "Cam1" };
|
|
file_stems = { "cam1" };
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
}
|
|
catch (char *c){
|
|
std::cout << "Exception: ";
|
|
std::cout << c << std::endl;
|
|
return 0;
|
|
}
|
|
|
|
|
|
////////////////////////
|
|
// 2D pupil detector
|
|
PupilFitter pupilFitter;
|
|
pupilFitter.setDebug(false);
|
|
/////////////////////////
|
|
|
|
//std::getchar();
|
|
//For running a video
|
|
//VideoCapture inputVideo1("C:\\Documents\\Osaka\\Research\\Eye Tracking\\Benchmark Videos\\eyetracking4.avi"); // Open input
|
|
VideoWriter outputVideo1;
|
|
outputVideo1.open("C:\\Documents\\Osaka\\Research\\Eye Tracking\\Benchmark Videos\\outSaccade.avi",
|
|
CV_FOURCC('W', 'M', 'V', '2'),
|
|
20,
|
|
cv::Size(640,480),
|
|
true);
|
|
Mat frame1;
|
|
|
|
// Main loop
|
|
const char kTerminate = 27;//Escape 0x1b
|
|
bool is_run = true;
|
|
bool isSaccade = false;
|
|
bool isBlink = false;
|
|
int blinkCount = 0; //holds the number of blinks for this video
|
|
int saccadeCount = 0; //holds the number of saccades for this video
|
|
bool prevSaccade = false; //added if a saccade value was detected in the previous frame
|
|
vector<float> timeData; //vector holding timestamps in ms corresponding to gaze data for N frames
|
|
vector<float> xData; //corresponding x eye rotations for N frames
|
|
vector<float> yData; //corresponding y eye rotations for N frames
|
|
vector<float> intensityData; //holds average intensity of last N frames
|
|
vector<singleeyefitter::EyeModelFitter::Sphere> eyes; //holds a vector of spheres for the eye model filter
|
|
|
|
|
|
while (is_run) {
|
|
|
|
|
|
//inputVideo1 >> frame1;//for video
|
|
//if (frame1.empty()) {//for video
|
|
// break;
|
|
//}
|
|
|
|
//imshow("test", frame1);//for video
|
|
//waitKey(0);
|
|
|
|
|
|
// Fetch key input
|
|
char kKEY = 0;
|
|
if (kVisualization) {
|
|
kKEY = cv::waitKey(1);
|
|
}
|
|
switch (kKEY) {
|
|
case kTerminate:
|
|
is_run = false;
|
|
break;
|
|
}
|
|
|
|
|
|
|
|
// Fetch images
|
|
for (size_t cam = 0; cam < kCameraNums; cam++) {
|
|
|
|
eyecams[cam] -> fetchFrame(images[cam]);
|
|
|
|
}
|
|
// Process each camera images
|
|
for (size_t cam = 0; cam < kCameraNums; cam++) {
|
|
|
|
cv::Mat &img = images[cam];
|
|
//img = frame1; //for video
|
|
//imshow("test", img);
|
|
//waitKey(1);
|
|
|
|
if (img.empty()) {
|
|
//is_run = false;
|
|
break;
|
|
}
|
|
|
|
// Undistort a captured image
|
|
//camera_undistorters[cam]->undistort(img, img);
|
|
|
|
//cv::Mat img_rgb_debug = frame1.clone(); \\for video
|
|
cv::Mat img_rgb_debug = img.clone();
|
|
cv::Mat img_grey;
|
|
|
|
switch (kKEY) {
|
|
case 'r':
|
|
eye_model_updaters[cam]->reset();
|
|
break;
|
|
case 'p':
|
|
eye_model_updaters[cam]->add_fitter_max_count(10);
|
|
break;
|
|
case 'q':
|
|
is_run = false;
|
|
break;
|
|
case 'z':
|
|
eye_model_updaters[cam]->rm_oldest_observation();
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
|
|
|
|
|
|
const clock_t begin_time = clock();
|
|
|
|
// 2D ellipse detection
|
|
std::vector<cv::Point2f> inlier_pts;
|
|
cv::cvtColor(img, img_grey, CV_RGB2GRAY);
|
|
cv::RotatedRect rr_pf;
|
|
|
|
|
|
//imshow("test", img_grey);
|
|
|
|
bool is_pupil_found = pupilFitter.pupilAreaFitRR(img_grey, rr_pf, inlier_pts, 15, 0, 0, 20, 30, 250, 6);
|
|
|
|
//cout << "pupil fitter time: " << float(clock() - begin_time) / CLOCKS_PER_SEC << endl;
|
|
|
|
const clock_t begin_time2 = clock();
|
|
|
|
singleeyefitter::Ellipse2D<double> el = singleeyefitter::toEllipse<double>(eye_tracker::toImgCoordInv(rr_pf, img, 1.0));
|
|
|
|
//cout << "singleeyefitter time: " << float(clock() - begin_time2) / CLOCKS_PER_SEC << endl;
|
|
|
|
|
|
// 3D eye pose estimation
|
|
bool is_reliable = false;
|
|
bool is_added = false;
|
|
const bool force_add = false;
|
|
const double kReliabilityThreshold = 0.0;//0.96;
|
|
double ellipse_reliability = 0.0; /// Reliability of a detected 2D ellipse based on 3D eye model
|
|
if (is_pupil_found) {
|
|
if (eye_model_updaters[cam]->is_model_built()) {
|
|
ellipse_reliability = eye_model_updaters[cam]->compute_reliability(img, el, inlier_pts);
|
|
is_reliable = (ellipse_reliability > kReliabilityThreshold);
|
|
// is_reliable = true;
|
|
|
|
|
|
eye_model_updaters[cam]->rm_oldest_observation();
|
|
eye_model_updaters[cam]->add_observation(img_grey, el, inlier_pts, false);
|
|
eye_model_updaters[cam]->force_rebuild_model();
|
|
}
|
|
else {
|
|
cout << "oops" << endl;
|
|
is_added = eye_model_updaters[cam]->add_observation(img_grey, el, inlier_pts, force_add);
|
|
}
|
|
//TODO test in Unity to see how well this works
|
|
}
|
|
|
|
// Visualize results
|
|
if (cam == 0 && kVisualization) {
|
|
|
|
// 2D pupil
|
|
if (is_pupil_found) {
|
|
cv::ellipse(img_rgb_debug, rr_pf, cv::Vec3b(255, 128, 0), 1);
|
|
}
|
|
|
|
// 3D eye ball
|
|
if (eye_model_updaters[cam]->is_model_built()) {
|
|
cv::putText(img, "Reliability: " + std::to_string(ellipse_reliability), cv::Point(30, 440), cv::FONT_HERSHEY_SIMPLEX, 1.0, cv::Scalar(0, 128, 255), 1);
|
|
if (is_reliable) {
|
|
|
|
singleeyefitter::Sphere<double> medianCircle;
|
|
//bool useDriftCorrection = false;
|
|
//if (eyes.size() > 0) {
|
|
// medianCircle = eye_model_updaters[cam]->eyeModelFilter(curr_circle, eyes);
|
|
// useDriftCorrection = true;
|
|
//}
|
|
|
|
//std::cout << "after filter: " << curr_circle.radius << std::endl;
|
|
|
|
eye_model_updaters[cam]->render(img_rgb_debug, el, inlier_pts);
|
|
eye_model_updaters[cam]->set_fitter_max_count(130); //manually sets max count
|
|
|
|
//3D filtered eye model
|
|
medianCircle = eye_model_updaters[cam]->eyeModelFilter(eye_model_updaters[cam]->fitter().eye, eyes, 500);
|
|
if (medianCircle.radius < 10) {
|
|
medianCircle.radius = 10;
|
|
}
|
|
eye_model_updaters[cam]->setEye(medianCircle);
|
|
curr_circle = eye_model_updaters[cam]->unproject(img, el, inlier_pts);
|
|
// 3D pupil (relative to filtered eye model)
|
|
singleeyefitter::Ellipse2D<double> pupil_el(singleeyefitter::project(curr_circle, focal_length));
|
|
cv::RotatedRect rr_pupil = eye_tracker::toImgCoord(singleeyefitter::toRotatedRect(pupil_el), img, 1.0f);
|
|
singleeyefitter::EyeModelFitter::Sphere filteredEye(medianCircle.centre, medianCircle.radius);
|
|
|
|
cout << "radius was " << medianCircle.radius << endl;
|
|
cv::RotatedRect rr_eye = eye_tracker::toImgCoord(singleeyefitter::toRotatedRect(
|
|
singleeyefitter::project(filteredEye, focal_length)), img, 1.0f);
|
|
cv::ellipse(img_rgb_debug, rr_eye, cv::Vec3b(255, 222, 222), 2, CV_AA);
|
|
cv::circle(img_rgb_debug, rr_eye.center, 3, cv::Vec3b(255, 32, 32), 2); // Eyeball center projection
|
|
singleeyefitter::EyeModelFitter::Circle c_end = curr_circle;
|
|
c_end.centre = curr_circle.centre + (10.0)*curr_circle.normal;
|
|
|
|
cv::line(img_rgb_debug, rr_eye.center, rr_pupil.center, cv::Vec3b(25, 22, 222), 3, CV_AA);
|
|
|
|
//update time, xdata, and ydata vectors for input into saccade detector
|
|
dataAdd(curr_circle.centre(0), 5, xData);
|
|
dataAdd(curr_circle.centre(1), 5, yData);
|
|
dataAdd(clock(), 5, timeData);
|
|
float intensity = 0;
|
|
|
|
//to-Unity write
|
|
//std::ofstream myfile("C:\\Users\\O\\Documents\\Visual Studio 2013\\Projects\\EyeTrackerRealTime\\coordinates.txt");
|
|
//std::ofstream myfile;
|
|
//myfile.open("C:\\Documents\\Osaka\\Research\\Presence 2017\\testcoordinates.txt", std::ios_base::app);
|
|
//myfile << "" << c_end.centre.x() << "," << c_end.centre.y() << "," << c_end.centre.z()
|
|
// << "," << filteredEye.centre[0] << "," << filteredEye.centre[1] << "," << filteredEye.centre[2] <<
|
|
// std::endl;
|
|
//myfile.close();
|
|
|
|
}
|
|
}else{
|
|
eye_model_updaters[cam]->render_status(img_rgb_debug);
|
|
cv::putText(img, "Sample #: " + std::to_string(eye_model_updaters[cam]->fitter_count()) + "/" + std::to_string(eye_model_updaters[cam]->fitter_end_count()),
|
|
cv::Point(30, 440), cv::FONT_HERSHEY_SIMPLEX, 1.0, cv::Scalar(0, 128, 255), 2);
|
|
}
|
|
|
|
float confidence = 0;
|
|
//outputVideo1 << img_rgb_debug;
|
|
cv::imshow(window_names[cam], img_rgb_debug);
|
|
|
|
|
|
} // Visualization
|
|
|
|
} // For each cameras
|
|
|
|
// Compute FPS
|
|
frame_rate_counter.count();
|
|
// Print current frame data
|
|
static int ss = 0;
|
|
if (ss++ > 100) {
|
|
std::cout << "Frame #" << frame_rate_counter.frame_count() << ", FPS=" << frame_rate_counter.fps() << std::endl;
|
|
ss = 0;
|
|
}
|
|
|
|
singleeyefitter::EyeModelFitter::Circle curr_circle;
|
|
singleeyefitter::EyeModelFitter::Circle c_end = curr_circle;
|
|
c_end.centre = curr_circle.centre + (10.0)*curr_circle.normal; // Unit: mm
|
|
|
|
|
|
|
|
|
|
|
|
}// Main capture loop
|
|
outputVideo1.release();
|
|
return 0;
|
|
|
|
}
|