最近由于项目需要，需要在Ubuntu14.04中使用QT中使用OpenFace,配置了好长时间才配置好的，将配置过程记录下来，让后人少走点弯路。
安装OpenFace
OpenFace的官网：https://github.com/TadasBaltrusaitis/OpenFace
按照上面的操作安装OpenFace就可以了，注意：一定要严格按照上面的步骤来，否则很容易出错。安装完之后，就可以在QT中使用OpenFace了。
在QT中使用OpenFace
我在QT中需要使用OpenFace计算人脸的角度，用到的OpenFace自带的示例代码如下：
FaceLandmarkVid.cpp(路径：OpenFace/exe/FaceLandmarkVid/FaceLandmarkVid.cpp)
//////////////////////////////////////////////
// Copyright (C) 2016, Carnegie Mellon University and University of Cambridge,
// all rights reserved.
//
// THIS SOFTWARE IS PROVIDED “AS IS” FOR ACADEMIC USE ONLY AND ANY EXPRESS
// OR IMPLIED WARRANTIES WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
// THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
// PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDERS OR CONTRIBUTORS
// BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY.
// OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
// SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
// HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
// STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
// Notwithstanding the license granted herein, Licensee acknowledges that certain components
// of the Software may be covered by so-called “open source” software licenses (“Open Source
// Components”), which means any software licenses approved as open source licenses by the
// Open Source Initiative or any substantially similar licenses, including without limitation any
// license that, as a condition of distribution of the software licensed under such license,
// requires that the distributor make the software available in source code format. Licensor shall
// provide a list of Open Source Components for a particular version of the Software upon
// Licensee’s request. Licensee will comply with the applicable terms of such licenses and to
// the extent required by the licenses covering Open Source Components, the terms of such
// licenses will apply in lieu of the terms of this Agreement. To the extent the terms of the
// licenses applicable to Open Source Components prohibit any of the restrictions in this
// License Agreement with respect to such Open Source Component, such restrictions will not
// apply to such Open Source Component. To the extent the terms of the licenses applicable to
// Open Source Components require Licensor to make an offer to provide source code or
// related information in connection with the Software, such offer is hereby made. Any request
// for source code or related information should be directed to cl-face-tracker-distribution@lists.cam.ac.uk
// Licensee acknowledges receipt of notices for the Open Source Components for the initial
// delivery of the Software.
// * Any publications arising from the use of this software, including but
// not limited to academic journal and conference publications, technical
// reports and manuals, must cite at least one of the following works:
//
// OpenFace: an open source facial behavior analysis toolkit
// Tadas Baltrušaitis, Peter Robinson, and Louis-Philippe Morency
// in IEEE Winter Conference on Applications of Computer Vision, 2016
//
// Rendering of Eyes for Eye-Shape Registration and Gaze Estimation
// Erroll Wood, Tadas Baltrušaitis, Xucong Zhang, Yusuke Sugano, Peter Robinson, and Andreas Bulling
// in IEEE International. Conference on Computer Vision (ICCV),2015
//
// Cross-dataset learning and person-speci?c normalisation for automatic Action Unit detection
// Tadas Baltrušaitis, Marwa Mahmoud, and Peter Robinson
// in Facial Expression Recognition and Analysis Challenge,
// IEEE International Conference on Automatic Face and Gesture Recognition, 2015
//
// Constrained Local Neural Fields for robust facial landmark detection in the wild.
// Tadas Baltrušaitis, Peter Robinson, and Louis-Philippe Morency.
// in IEEE Int. Conference on Computer Vision Workshops, 300 Faces in-the-Wild Challenge, 2013.
//
//////////////////////////////////////////////
// FaceTrackingVid.cpp : Defines the entry point for the console application for tracking faces in videos.
// Libraries for landmark detection (includes CLNF and CLM modules)
#include "LandmarkCoreIncludes.h"
#include "GazeEstimation.h"
#include <fstream>
#include <sstream>
// OpenCV includes
#include <opencv2/videoio/videoio.hpp>// Video write
#include <opencv2/videoio/videoio_c.h>// Video write
#include <opencv2/imgproc.hpp>
#include <opencv2/highgui/highgui.hpp>
// Boost includes
#include <filesystem.hpp>
#include <filesystem/fstream.hpp>
#define INFO_STREAM( stream ) \
std::cout << stream << std::endl
#define WARN_STREAM( stream ) \
std::cout << "Warning: " << stream << std::endl
#define ERROR_STREAM( stream ) \
std::cout << "Error: " << stream << std::endl
static void printErrorAndAbort( const std::string & error )
{
std::cout << error << std::endl;
abort();
}
#define FATAL_STREAM( stream ) \
printErrorAndAbort( std::string( "Fatal error: " ) + stream )
using namespace std;
vector<string> get_arguments(int argc, char **argv)
{
vector<string> arguments;
for(int i = 0; i < argc; ++i)
{
arguments.push_back(string(argv[i]));
}
return arguments;
}
// Some globals for tracking timing information for visualisation
double fps_tracker = -1.0;
int64 t0 = 0;
// Visualising the results
void visualise_tracking(cv::Mat& captured_image, cv::Mat_<float>& depth_image, const LandmarkDetector::CLNF& face_model, const LandmarkDetector::FaceModelParameters& det_parameters, cv::Point3f gazeDirection0, cv::Point3f gazeDirection1, int frame_count, double fx, double fy, double cx, double cy)
{
// Drawing the facial landmarks on the face and the bounding box around it if tracking is successful and initialised
double detection_certainty = face_model.detection_certainty;
bool detection_success = face_model.detection_success;
double visualisation_boundary = 0.2;
// Only draw if the reliability is reasonable, the value is slightly ad-hoc
if (detection_certainty < visualisation_boundary)
{
LandmarkDetector::Draw(captured_image, face_model);
double vis_certainty = detection_certainty;
if (vis_certainty > 1)
vis_certainty = 1;
if (vis_certainty < -1)
vis_certainty = -1;
vis_certainty = (vis_certainty + 1) / (visualisation_boundary + 1);
// A rough heuristic for box around the face width
int thickness = (int)std::ceil(2.0* ((double)captured_image.cols) / 640.0);
cv::Vec6d pose_estimate_to_draw = LandmarkDetector::GetCorrectedPoseWorld(face_model, fx, fy, cx, cy);
// Draw it in reddish if uncertain, blueish if certain
LandmarkDetector::DrawBox(captured_image, pose_estimate_to_draw, cv::Scalar((1 - vis_certainty)*255.0, 0, vis_certainty * 255), thickness, fx, fy, cx, cy);
if (det_parameters.track_gaze && detection_success && face_model.eye_model)
{
FaceAnalysis::DrawGaze(captured_image, face_model, gazeDirection0, gazeDirection1, fx, fy, cx, cy);
}
}
// Work out the framerate
if (frame_count % 10 == 0)
{
double t1 = cv::getTickCount();
fps_tracker = 10.0 / (double(t1 - t0) / cv::getTickFrequency());
t0 = t1;
}
// Write out the framerate on the image before displaying it
char fpsC[255];
std::sprintf(fpsC, "%d", (int)fps_tracker);
string fpsSt("FPS:");
fpsSt += fpsC;
cv::putText(captured_image, fpsSt, cv::Point(10, 20), CV_FONT_HERSHEY_SIMPLEX, 0.5, CV_RGB(255, 0, 0));
if (!det_parameters.quiet_mode)
{
cv::namedWindow("tracking_result", 1);
cv::imshow("tracking_result", captured_image);
if (!depth_image.empty())
{
// Division needed for visualisation purposes
imshow("depth", depth_image / 2000.0);
}
}
}
int main (int argc, char **argv)
{
vector<string> arguments = get_arguments(argc, argv);
// Some initial parameters that can be overriden from command line
vector<string> files, depth_directories, output_video_files, out_dummy;
// By default try webcam 0
int device = 0;
LandmarkDetector::FaceModelParameters det_parameters(arguments);
// Get the input output file parameters
// Indicates that rotation should be with respect to world or camera coordinates
bool u;
LandmarkDetector::get_video_input_output_params(files, depth_directories, out_dummy, output_video_files, u, arguments);
// The modules that are being used for tracking
LandmarkDetector::CLNF clnf_model(det_parameters.model_location);
// Grab camera parameters, if they are not defined (approximate values will be used)
float fx = 0, fy = 0, cx = 0, cy = 0;
// Get camera parameters
LandmarkDetector::get_camera_params(device, fx, fy, cx, cy, arguments);
// If cx (optical axis centre) is undefined will use the image size/2 as an estimate
bool cx_undefined = false;
bool fx_undefined = false;
if (cx == 0 || cy == 0)
{
cx_undefined = true;
}
if (fx == 0 || fy == 0)
{
fx_undefined = true;
}
// If multiple video files are tracked, use this to indicate if we are done
bool done = false;
int f_n = -1;
det_parameters.track_gaze = true;
while(!done) // this is not a for loop as we might also be reading from a webcam
{
string current_file;
// We might specify multiple video files as arguments
if(files.size() > 0)
{
f_n++;
current_file = files[f_n];
}
else
{
// If we want to write out from webcam
f_n = 0;
}
bool use_depth = !depth_directories.empty();
// Do some grabbing
cv::VideoCapture video_capture;
if( current_file.size() > 0 )
{
if (!boost::filesystem::exists(current_file))
{
FATAL_STREAM("File does not exist");
}
current_file = boost::filesystem::path(current_file).generic_string();
INFO_STREAM( "Attempting to read from file: " << current_file );
video_capture = cv::VideoCapture( current_file );
}
else
{
INFO_STREAM( "Attempting to capture from device: " << device );
video_capture = cv::VideoCapture( device );
// Read a first frame often empty in camera
cv::Mat captured_image;
video_capture >> captured_image;
}
if( !video_capture.isOpened() ) FATAL_STREAM( "Failed to open video source" );
else INFO_STREAM( "Device or file opened");
cv::Mat captured_image;
video_capture >> captured_image;
// If optical centers are not defined just use center of image
if (cx_undefined)
{
cx = captured_image.cols / 2.0f;
cy = captured_image.rows / 2.0f;
}
// Use a rough guess-timate of focal length
if (fx_undefined)
{
fx = 500 * (captured_image.cols / 640.0);
fy = 500 * (captured_image.rows / 480.0);
fx = (fx + fy) / 2.0;
fy = fx;
}
int frame_count = 0;
// saving the videos
cv::VideoWriter writerFace;
if (!output_video_files.empty())
{
writerFace = cv::VideoWriter(output_video_files[f_n], CV_FOURCC('D', 'I', 'V', 'X'), 30, captured_image.size(), true);
}
// Use for timestamping if using a webcam
int64 t_initial = cv::getTickCount();
INFO_STREAM( "Starting tracking");
while(!captured_image.empty())
{
// Reading the images
cv::Mat_<float> depth_image;
cv::Mat_<uchar> grayscale_image;
if(captured_image.channels() == 3)
{
cv::cvtColor(captured_image, grayscale_image, CV_BGR2GRAY);
}
else
{
grayscale_image = captured_image.clone();
}
// Get depth image
if(use_depth)
{
char* dst = new char[100];
std::stringstream sstream;
sstream << depth_directories[f_n] << "\\depth%05d.png";
sprintf(dst, sstream.str().c_str(), frame_count + 1);
// Reading in 16-bit png image representing depth
cv::Mat_<short> depth_image_16_bit = cv::imread(string(dst), -1);
// Convert to a floating point depth image
if(!depth_image_16_bit.empty())
{
depth_image_16_bit.convertTo(depth_image, CV_32F);
}
else
{
WARN_STREAM( "Can't find depth image" );
}
}
// The actual facial landmark detection / tracking
bool detection_success = LandmarkDetector::DetectLandmarksInVideo(grayscale_image, depth_image, clnf_model, det_parameters);
// Visualising the results
// Drawing the facial landmarks on the face and the bounding box around it if tracking is successful and initialised
double detection_certainty = clnf_model.detection_certainty;
// Gaze tracking, absolute gaze direction
cv::Point3f gazeDirection0(0, 0, -1);
cv::Point3f gazeDirection1(0, 0, -1);
if (det_parameters.track_gaze && detection_success && clnf_model.eye_model)
{
FaceAnalysis::EstimateGaze(clnf_model, gazeDirection0, fx, fy, cx, cy, true);
FaceAnalysis::EstimateGaze(clnf_model, gazeDirection1, fx, fy, cx, cy, false);
}
visualise_tracking(captured_image, depth_image, clnf_model, det_parameters, gazeDirection0, gazeDirection1, frame_count, fx, fy, cx, cy);
// output the tracked video
if (!output_video_files.empty())
{
writerFace << captured_image;
}
video_capture >> captured_image;
// detect key presses
char character_press = cv::waitKey(1);
// restart the tracker
if(character_press == 'r')
{
clnf_model.Reset();
}
// quit the application
else if(character_press=='q')
{
return(0);
}
// Update the frame count
frame_count++;
}
frame_count = 0;
// Reset the model, for the next video
clnf_model.Reset();
// break out of the loop if done with all the files (or using a webcam)
if(f_n == files.size() -1 || files.empty())
{
done = true;
}
}
return 0;
}
上面的代码需要注意：
将LandmarkDetector::FaceModelParameters det_parameters(arguments);修改为LandmarkDetector::FaceModelParameters det_parameters；
具体原因我也不太清楚，我修改了上诉代码后，就能用了。
在QT中新建立一个工程，然后将上面代码复制到项目中，要特别注意配置文件的设置：
我将我的配置文件贴出来，供大家参考：
QT += core
QT -= gui
TARGET = OpenFace
CONFIG += console
CONFIG -= app_bundle
CONFIG += c++11
TEMPLATE = app
SOURCES += main.cpp
INCLUDEPATH+=/home/qq/Document/usr/local/OpenCV_3.1/so/include \
/home/qq/Document/Work/OpenFace/lib/local/LandmarkDetector/include/ \
/home/qq/Document/Work/OpenFace/lib/local/FaceAnalyser/include/\
/home/qq/Document/usr/local/Boost/include/ \
/home/qq/Document/usr/local/Boost/include/boost \
/home/qq/Document/Work/OpenFace/lib/3rdParty/dlib/include \
/home/qq/Document/usr/local/tbb/include/ \
/usr/local/include\
/usr/include/boost \
/home/qq/Document/usr/local/CBLAS/include
LIBS += -L/home/qq/Document/Work/OpenFace/Build/lib/local/FaceAnalyser \
-lFaceAnalyser \
LIBS += -L/home/qq/Document/Work/OpenFace/Build/lib/local/LandmarkDetector \
-lLandmarkDetector \
LIBS += -L/home/qq/Document/Work/OpenFace/Build/lib/3rdParty/dlib \
-ldlib \
LIBS += -L/home/qq/Document/usr/local/OpenCV_3.1/so/lib \
-lopencv_calib3d \
-lopencv_core \
-lopencv_cudaarithm \
-lopencv_cudabgsegm \
-lopencv_cudacodec \
-lopencv_cudafeatures2d \
-lopencv_cudafilters \
-lopencv_cudaimgproc \
-lopencv_cudalegacy \
-lopencv_cudaobjdetect \
-lopencv_cudaoptflow \
-lopencv_cudastereo \
-lopencv_cudawarping \
-lopencv_cudev \
-lopencv_features2d \
-lopencv_flann \
-lopencv_highgui \
-lopencv_imgcodecs \
-lopencv_imgproc \
-lopencv_ml \
-lopencv_objdetect \
-lopencv_photo \
-lopencv_shape \
-lopencv_stitching \
-lopencv_superres \
-lopencv_videoio \
-lopencv_video \
-lopencv_videostab
LIBS += -L/home/qq/Document/usr/local/Boost/lib/ \
-lboost_filesystem \
-lboost_system
LIBS += -L/home/qq/Document/usr/local/tbb/lib/ \
-ltbb \
-ltbbmalloc
LIBS +=/home/qq/Document/usr/local/CBLAS/lib/cblas_LINUX.a
LIBS +=/home/qq/Document/usr/local/CBLAS/lib/libblas.a
LIBS += -L/etc/alternatives \
-llapack \
需要注意的问题
1.C++代码在QT Creater出错
关于这个问题，需要在配置文件.pro中添加：
CONFIG += c++11
2.需要将模型文件夹放在与可执行文件同一个目录中
也就是OpenFace中的model，classifiers,和AU_predictors文件夹
运行结果
运行成功！