opencv内demo（find_obj.cpp）理解

#include "opencv2/objdetect/objdetect.hpp"
#include "opencv2/features2d/features2d.hpp"
#include "opencv2/highgui/highgui.hpp"
#include "opencv2/calib3d/calib3d.hpp"
#include "opencv2/nonfree/nonfree.hpp"
#include "opencv2/imgproc/imgproc_c.h"
#include "opencv2/legacy/legacy.hpp"
#include "opencv2/legacy/compat.hpp"

#include <iostream>
#include <vector>
#include <stdio.h>

using namespace std;
static void help()
{
    printf(
        "This program demonstrated the use of the SURF Detector and Descriptor using\n"
        "either FLANN (fast approx nearst neighbor classification) or brute force matching\n"
        "on planar objects.\n"
        "Usage:\n"
        "./find_obj <object_filename> <scene_filename>, default is box.png  and box_in_scene.png\n\n");
    return;
}

// define whether to use approximate nearest-neighbor search
#define USE_FLANN

#ifdef USE_FLANN
static void
flannFindPairs(const CvSeq*, const CvSeq* objectDescriptors,
const CvSeq*, const CvSeq* imageDescriptors, vector<int>& ptpairs)
//函数flannFindPairs有5个形参1.参照物的keypoint 2.keypoint的描述符 3.图片的keypoint 4.图片keypoint的描述符 5.int型容器
//用于找到两幅图像之间匹配的点对，并把匹配的点对存储在 ptpairs 向量中，其中物体(object)图像的特征点
{
    int length = (int)(objectDescriptors->elem_size / sizeof(float));
    //CvSeq可动态增长序列
    cv::Mat m_object(objectDescriptors->total, length, CV_32F);
    cv::Mat m_image(imageDescriptors->total, length, CV_32F);


    // copy descriptors
    CvSeqReader obj_reader;
    float* obj_ptr = m_object.ptr<float>(0);
    cvStartReadSeq(objectDescriptors, &obj_reader);//reader来读取seq内部数据的  
    for (int i = 0; i < objectDescriptors->total; i++)
    {
        const float* descriptor = (const float*)obj_reader.ptr;
        CV_NEXT_SEQ_ELEM(obj_reader.seq->elem_size, obj_reader);
        memcpy(obj_ptr, descriptor, length*sizeof(float));
        //memcpy内存拷贝函数
        obj_ptr += length;
    }
    CvSeqReader img_reader;
    float* img_ptr = m_image.ptr<float>(0);
    cvStartReadSeq(imageDescriptors, &img_reader);
    for (int i = 0; i < imageDescriptors->total; i++)
    {
        const float* descriptor = (const float*)img_reader.ptr;
        CV_NEXT_SEQ_ELEM(img_reader.seq->elem_size, img_reader);
        memcpy(img_ptr, descriptor, length*sizeof(float));
        img_ptr += length;
    }

    // find nearest neighbors using FLANN
    //FLANN（近似）最近邻开源库
    cv::Mat m_indices(objectDescriptors->total, 2, CV_32S);
    cv::Mat m_dists(objectDescriptors->total, 2, CV_32F);
    cv::flann::Index flann_index(m_image, cv::flann::KDTreeIndexParams(4));  // using 4 randomized kdtrees
    flann_index.knnSearch(m_object, m_indices, m_dists, 2, cv::flann::SearchParams(64)); // maximum number of leafs checked

    int* indices_ptr = m_indices.ptr<int>(0);
    float* dists_ptr = m_dists.ptr<float>(0);
    for (int i = 0; i<m_indices.rows; ++i) {
        if (dists_ptr[2 * i]<0.6*dists_ptr[2 * i + 1]) {
            ptpairs.push_back(i);
            ptpairs.push_back(indices_ptr[2 * i]);
        }
    }
}
#else

static double
compareSURFDescriptors(const float* d1, const float* d2, double best, int length)
{
    double total_cost = 0;
    assert(length % 4 == 0);
    for (int i = 0; i < length; i += 4)
    {
        double t0 = d1[i] - d2[i];
        double t1 = d1[i + 1] - d2[i + 1];
        double t2 = d1[i + 2] - d2[i + 2];
        double t3 = d1[i + 3] - d2[i + 3];
        total_cost += t0*t0 + t1*t1 + t2*t2 + t3*t3;
        if (total_cost > best)
            break;
    }
    return total_cost;
}

static int
naiveNearestNeighbor(const float* vec, int laplacian,
const CvSeq* model_keypoints,
const CvSeq* model_descriptors)
{
    int length = (int)(model_descriptors->elem_size / sizeof(float));
    int i, neighbor = -1;
    double d, dist1 = 1e6, dist2 = 1e6;
    CvSeqReader reader, kreader;
    cvStartReadSeq(model_keypoints, &kreader, 0);
    cvStartReadSeq(model_descriptors, &reader, 0);

    for (i = 0; i < model_descriptors->total; i++)
    {
        const CvSURFPoint* kp = (const CvSURFPoint*)kreader.ptr;
        const float* mvec = (const float*)reader.ptr;
        CV_NEXT_SEQ_ELEM(kreader.seq->elem_size, kreader);
        CV_NEXT_SEQ_ELEM(reader.seq->elem_size, reader);
        if (laplacian != kp->laplacian)
            continue;
        d = compareSURFDescriptors(vec, mvec, dist2, length);
        if (d < dist1)
        {
            dist2 = dist1;
            dist1 = d;
            neighbor = i;
        }
        else if (d < dist2)
            dist2 = d;
    }
    if (dist1 < 0.6*dist2)
        return neighbor;
    return -1;
}

static void
findPairs(const CvSeq* objectKeypoints, const CvSeq* objectDescriptors,
const CvSeq* imageKeypoints, const CvSeq* imageDescriptors, vector<int>& ptpairs)
{
    int i;
    CvSeqReader reader, kreader;
    cvStartReadSeq(objectKeypoints, &kreader);
    cvStartReadSeq(objectDescriptors, &reader);
    ptpairs.clear();

    for (i = 0; i < objectDescriptors->total; i++)
    {
        const CvSURFPoint* kp = (const CvSURFPoint*)kreader.ptr;
        const float* descriptor = (const float*)reader.ptr;
        CV_NEXT_SEQ_ELEM(kreader.seq->elem_size, kreader);
        CV_NEXT_SEQ_ELEM(reader.seq->elem_size, reader);
        int nearest_neighbor = naiveNearestNeighbor(descriptor, kp->laplacian, imageKeypoints, imageDescriptors);
        if (nearest_neighbor >= 0)
        {
            ptpairs.push_back(i);
            ptpairs.push_back(nearest_neighbor);
        }
    }
}
#endif

/* a rough implementation for object location */
static int
locatePlanarObject(const CvSeq* objectKeypoints, const CvSeq* objectDescriptors,
const CvSeq* imageKeypoints, const CvSeq* imageDescriptors,
const CvPoint src_corners[4], CvPoint dst_corners[4])
//函数locatePlanarObject有6个形参1.参照物的keypoint 2.keypoint的描述符 3.图片的keypoint 4.图片keypoint的描述符 5.src点 6.dst点
{
    double h[9];
    CvMat _h = cvMat(3, 3, CV_64F, h);
    vector<int> ptpairs;
    vector<CvPoint2D32f> pt1, pt2;
    CvMat _pt1, _pt2;
    int i, n;

#ifdef USE_FLANN
    flannFindPairs(objectKeypoints, objectDescriptors, imageKeypoints, imageDescriptors, ptpairs);
#else
    findPairs(objectKeypoints, objectDescriptors, imageKeypoints, imageDescriptors, ptpairs);
#endif

    n = (int)(ptpairs.size() / 2);
    if (n < 4)
        return 0;

    pt1.resize(n);
    pt2.resize(n);
    for (i = 0; i < n; i++)
    {
        pt1[i] = ((CvSURFPoint*)cvGetSeqElem(objectKeypoints, ptpairs[i * 2]))->pt;
        pt2[i] = ((CvSURFPoint*)cvGetSeqElem(imageKeypoints, ptpairs[i * 2 + 1]))->pt;
    }

    _pt1 = cvMat(1, n, CV_32FC2, &pt1[0]);
    _pt2 = cvMat(1, n, CV_32FC2, &pt2[0]);
    if (!cvFindHomography(&_pt1, &_pt2, &_h, CV_RANSAC, 5))
        //在两个平面之间寻找单映射变换矩阵
        return 0;

    for (i = 0; i < 4; i++)
    {
        double x = src_corners[i].x, y = src_corners[i].y;
        double Z = 1. / (h[6] * x + h[7] * y + h[8]);
        double X = (h[0] * x + h[1] * y + h[2])*Z;
        double Y = (h[3] * x + h[4] * y + h[5])*Z;
        dst_corners[i] = cvPoint(cvRound(X), cvRound(Y));
    }

    return 1;
}

int main(int argc, char** argv)
{
    const char* object_filename = argc == 3 ? argv[1] : "box.png";
    const char* scene_filename = argc == 3 ? argv[2] : "box_in_scene.png";
    //所在文件夹下的默认图片

    cv::initModule_nonfree();
    help();

    IplImage* object = cvLoadImage(object_filename, CV_LOAD_IMAGE_GRAYSCALE);
    IplImage* image = cvLoadImage(scene_filename, CV_LOAD_IMAGE_GRAYSCALE);
    //将默认图片加载到本地变量中

    if (!object || !image)
    {
        fprintf(stderr, "Can not load %s and/or %s\n",
            object_filename, scene_filename);
        exit(-1);
    }
    //如果本地变量内容为空则输出无法加载，并退出

    CvMemStorage* storage = cvCreateMemStorage(0);
    //创建一个内存的存储器

    cvNamedWindow("Object", 1);
    cvNamedWindow("Object Correspond", 1);
    //新建两个窗口并命名

    static CvScalar colors[] =
    {
        { { 0, 0, 255 } },
        { { 0, 128, 255 } },
        { { 0, 255, 255 } },
        { { 0, 255, 0 } },
        { { 255, 128, 0 } },
        { { 255, 255, 0 } },
        { { 255, 0, 0 } },
        { { 255, 0, 255 } },
        { { 255, 255, 255 } }
    };
    //设置颜色值

    IplImage* object_color = cvCreateImage(cvGetSize(object), 8, 3);
    cvCvtColor(object, object_color, CV_GRAY2BGR);
    //将object中的图片转化成灰度图并保存在object_color中

    CvSeq* objectKeypoints = 0, *objectDescriptors = 0;
    CvSeq* imageKeypoints = 0, *imageDescriptors = 0;
    int i;
    CvSURFParams params = cvSURFParams(800, 1);
    //定义SURF算法中要用到的参数(阙值)
    
    double tt = (double)cvGetTickCount();
    //程序启动 记录时间

    cvExtractSURF(object, 0, &objectKeypoints, &objectDescriptors, storage, params);
    //调用cvExtractSURF函数
    //参数1：输入灰度图
    //参数2：mask 标志位，指定我们识别特征点的区域
    //参数3：keypoints 向量的关键点
    //参数4：描述符（对特征点的属性进行描述）        参数5：储存空间        参数6：上面定义的参数

    printf("Object Descriptors: %d\n", objectDescriptors->total);
    //输出Object图片中的描述符 个数

    cvExtractSURF(image, 0, &imageKeypoints, &imageDescriptors, storage, params);
    printf("Image Descriptors: %d\n", imageDescriptors->total);
    //输出Image图片中的描述符 个数

    tt = (double)cvGetTickCount() - tt;
    printf("Extraction time = %gms\n", tt / (cvGetTickFrequency()*1000.));
    //计算程序运行时间，并输出

    CvPoint src_corners[4] = { { 0, 0 }, { object->width, 0 }, { object->width, object->height }, { 0, object->height } };
    CvPoint dst_corners[4];
    IplImage* correspond = cvCreateImage(cvSize(image->width, object->height + image->height), 8, 1);
    //create一个image宽度，高度为object+image的图片（通道为1）
    cvSetImageROI(correspond, cvRect(0, 0, object->width, object->height));
    //在图片correspond内set感兴趣区域
    cvCopy(object, correspond);
    //将object copy给该区域
    cvSetImageROI(correspond, cvRect(0, object->height, correspond->width, correspond->height));
    cvCopy(image, correspond);
    cvResetImageROI(correspond);
    //同上

#ifdef USE_FLANN
    printf("Using approximate nearest neighbor search\n");
#endif

    if (locatePlanarObject(objectKeypoints, objectDescriptors, imageKeypoints,
        imageDescriptors, src_corners, dst_corners))
    {
        for (i = 0; i < 4; i++)
        {
            CvPoint r1 = dst_corners[i % 4];
            CvPoint r2 = dst_corners[(i + 1) % 4];
            cvLine(correspond, cvPoint(r1.x, r1.y + object->height),
                cvPoint(r2.x, r2.y + object->height), colors[8]);
            //在correspond image上画出书的轮廓图
        }
    }
    vector<int> ptpairs;
#ifdef USE_FLANN
    flannFindPairs(objectKeypoints, objectDescriptors, imageKeypoints, imageDescriptors, ptpairs);
#else
    findPairs(objectKeypoints, objectDescriptors, imageKeypoints, imageDescriptors, ptpairs);
#endif
    for (i = 0; i < (int)ptpairs.size(); i += 2)
    {
        CvSURFPoint* r1 = (CvSURFPoint*)cvGetSeqElem(objectKeypoints, ptpairs[i]);
        CvSURFPoint* r2 = (CvSURFPoint*)cvGetSeqElem(imageKeypoints, ptpairs[i + 1]);
        cvLine(correspond, cvPointFrom32f(r1->pt),
            cvPoint(cvRound(r2->pt.x), cvRound(r2->pt.y + object->height)), colors[8]);
        //在correspond上画出匹配的关键点
    }

    cvShowImage("Object Correspond", correspond);
    //在窗口中显示correspond
    for (i = 0; i < objectKeypoints->total; i++)
    {
        //object中特征点的个数
        CvSURFPoint* r = (CvSURFPoint*)cvGetSeqElem(objectKeypoints, i);
        //返回objectKeypoints的索引，并将其强制转化为CvSURFPoint类型
        CvPoint center;//圆心
        int radius;
        center.x = cvRound(r->pt.x);//圆心的x
        center.y = cvRound(r->pt.y);//圆心的y
        radius = cvRound(r->size*1.2 / 9. * 2);
        cvCircle(object_color, center, radius, colors[0], 1, 8, 0);
        //在object_color上画圆，圆心center，半径radius
    }
    cvShowImage("Object", object_color);//将object_color输出

    cvWaitKey(0);

    cvDestroyWindow("Object");
    cvDestroyWindow("Object Correspond");

    return 0;
}