OpenCV 2.1 Ferns 테스트

OpenCV 2.1의 find_obj_ferns.cpp를 살짝 바꿔서 capture 영상과 함께 사용할 수 있게 했다.
tracking-by-detection에 초당 3프레임 정도 나오는 듯하다.

출연:
내 손
갤S
유선전화기
USB 케이블

[소스코드 추가]

#include <cv.h>
#include <cvaux.h>
#include <highgui.h>

#include <algorithm>
#include <iostream>
#include <vector>

#pragma comment(lib, "cv210.lib")
#pragma comment(lib, "cvaux210.lib")
#pragma comment(lib, "highgui210.lib")
#pragma comment(lib, "cxcore210.lib")

using namespace cv;

int main(int argc, char** argv)
{
 const char* object_filename = argc > 1 ? argv[1] : "box.png";
 const char* scene_filename = argc > 2 ? argv[2] : "box_in_scene.png";
 int i;
   
 cvNamedWindow("Object", 1);
 cvNamedWindow("Image", 1);
 cvNamedWindow("Object Correspondence", 1);
   
 Mat object = imread( object_filename, CV_LOAD_IMAGE_GRAYSCALE );
 Mat image;
   
    double imgscale = 1;

//  Mat _image = imread( scene_filename, CV_LOAD_IMAGE_GRAYSCALE );
//  resize(_image, image, Size(), 1./imgscale, 1./imgscale, INTER_CUBIC);

    if( !object.data ) // || !image.data )
    {
        fprintf( stderr, "Can not load %s and/or %s\n"
                "Usage: find_obj [<object_filename> <scene_filename>]\n",
                object_filename, scene_filename );
        exit(-1);
    }

    Size patchSize(32, 32);
    LDetector ldetector(7, 20, 2, 2000, patchSize.width, 2);
    ldetector.setVerbose(true);
    PlanarObjectDetector detector;
   
    vector<Mat> objpyr, imgpyr;
    int blurKSize = 3;
    double sigma = 0;
    GaussianBlur(object, object, Size(blurKSize, blurKSize), sigma, sigma);
    //GaussianBlur(image, image, Size(blurKSize, blurKSize), sigma, sigma);
    buildPyramid(object, objpyr, ldetector.nOctaves-1);
    //buildPyramid(image, imgpyr, ldetector.nOctaves-1);
   
    vector<KeyPoint> objKeypoints, imgKeypoints;
 PatchGenerator gen(0,256,5,true,0.8,1.2,-CV_PI/2,CV_PI/2,-CV_PI/2,CV_PI/2);
   
    string model_filename = format("%s_model.xml.gz", object_filename);
    printf("Trying to load %s ...\n", model_filename.c_str());
    FileStorage fs(model_filename, FileStorage::READ);
    if( fs.isOpened() )
    {
        detector.read(fs.getFirstTopLevelNode());
        printf("Successfully loaded %s.\n", model_filename.c_str());
    }
    else
    {
        printf("The file not found and can not be read. Let's train the model.\n");
        printf("Step 1. Finding the robust keypoints ...\n");
        ldetector.setVerbose(true);
        ldetector.getMostStable2D(object, objKeypoints, 100, gen);
        printf("Done.\nStep 2. Training ferns-based planar object detector ...\n");
        detector.setVerbose(true);
   
        detector.train(objpyr, objKeypoints, patchSize.width, 100, 11, 10000, ldetector, gen);
        printf("Done.\nStep 3. Saving the model to %s ...\n", model_filename.c_str());
        if( fs.open(model_filename, FileStorage::WRITE) )
            detector.write(fs, "ferns_model");
    }
    printf("Now find the keypoints in the image, try recognize them and compute the homography matrix\n");
    fs.release();

 CvCapture* cap = cvCreateCameraCapture(CV_CAP_DSHOW+1); // (CV_CAP_ANY);
 IplImage* iCap = cvQueryFrame(cap);
 Mat _image = Mat(iCap->width, iCap->height, CV_32FC1);

 while(1)
 {
  iCap = cvQueryFrame(cap);
  cvtColor(iCap, _image, CV_BGR2GRAY);

  //Mat _image = imread( scene_filename, CV_LOAD_IMAGE_GRAYSCALE );
  resize(_image, image, Size(), 1./imgscale, 1./imgscale, INTER_CUBIC);
  GaussianBlur(image, image, Size(blurKSize, blurKSize), sigma, sigma);
  buildPyramid(image, imgpyr, ldetector.nOctaves-1);

    vector<Point2f> dst_corners;
    Mat correspond( object.rows + image.rows, std::max(object.cols, image.cols), CV_8UC3);
    correspond = Scalar(0.);
    Mat part(correspond, Rect(0, 0, object.cols, object.rows));
    cvtColor(object, part, CV_GRAY2BGR);
    part = Mat(correspond, Rect(0, object.rows, image.cols, image.rows));
    cvtColor(image, part, CV_GRAY2BGR);
 
    vector<int> pairs;
    Mat H;
   
    double t = (double)getTickCount();
    objKeypoints = detector.getModelPoints();
    ldetector(imgpyr, imgKeypoints, 300);
   
    std::cout << "Object keypoints: " << objKeypoints.size() << "\n";
    std::cout << "Image keypoints: " << imgKeypoints.size() << "\n";
    bool found = detector(imgpyr, imgKeypoints, H, dst_corners, &pairs);
    t = (double)getTickCount() - t;
    printf("%gms\n", t*1000/getTickFrequency());
   
    if( found )
    {
        for( i = 0; i < 4; i++ )
        {
            Point r1 = dst_corners[i%4];
            Point r2 = dst_corners[(i+1)%4];
            line( correspond, Point(r1.x, r1.y+object.rows),
                 Point(r2.x, r2.y+object.rows), Scalar(0,0,255) );
        }
    }
   
    for( i = 0; i < (int)pairs.size(); i += 2 )
    {
        line( correspond, objKeypoints[pairs[i]].pt,
             imgKeypoints[pairs[i+1]].pt + Point2f(0,object.rows),
             Scalar(0,255,0) );
    }
   
    imshow( "Object Correspondence", correspond );
    Mat objectColor;
    cvtColor(object, objectColor, CV_GRAY2BGR);
    for( i = 0; i < (int)objKeypoints.size(); i++ )
    {
        circle( objectColor, objKeypoints[i].pt, 2, Scalar(0,0,255), -1 );
        circle( objectColor, objKeypoints[i].pt, (1 << objKeypoints[i].octave)*15, Scalar(0,255,0), 1 );
    }
    Mat imageColor;
    cvtColor(image, imageColor, CV_GRAY2BGR);
    for( i = 0; i < (int)imgKeypoints.size(); i++ )
    {
        circle( imageColor, imgKeypoints[i].pt, 2, Scalar(0,0,255), -1 );
        circle( imageColor, imgKeypoints[i].pt, (1 << imgKeypoints[i].octave)*15, Scalar(0,255,0), 1 );
    }
    imwrite("correspond.png", correspond );
    imshow( "Object", objectColor );
    imshow( "Image", imageColor );

  if(cvWaitKey(1) == 27)
   break;
 }

    waitKey(0);
    return 0;
}