ORB和SURF特征提取和效果对比

代码：

#include "StereoMatch.h"
#include <opencv2/opencv.hpp>
#include <opencv2/xfeatures2d.hpp>
using namespace cv;
using namespace cv::xfeatures2d;

#include <stdio.h>
using namespace std;

#include <Eigen/Eigen>
using namespace Eigen;
	#include<chrono>

void FLANN_Mh(Mat &imgL, Mat imgR)
{
    int keypoints_num = 400;
    vector<KeyPoint>keypointL,keypointR;
    Mat descritorL,descritorR;

    Ptr<SURF> detector = SURF::create(keypoints_num);   //创建SURF对象

    detector->detectAndCompute(imgL,Mat(),keypointL,descritorL);
    detector->detectAndCompute(imgR,Mat(),keypointR,descritorR);

    chrono::steady_clock::time_point t1=chrono::steady_clock::now();
    FlannBasedMatcher matcher;
    vector<DMatch> matches;      //匹配数据
    matcher.match(descritorL,descritorR,matches);   //描述匹配 将匹配结果存放至matches
    //flann匹配基于距离匹配 进行最优距离查找
    chrono::steady_clock::time_point t2=chrono::steady_clock::now();
    chrono::duration<double> time_used=chrono::duration_cast<chrono::duration<double>>(t2-t1);
    cout<<"FLANN方法的运行时间为："<<time_used.count()<<"秒。"<<endl;
    float maxdist = 3;
    float mindist = 50;
    //通过逐步收敛的办法查找到最大值最小值
    for(int i=0;i<descritorL.rows;i++)
    {
        float distance = matches[i].distance;
        distance*=100;
       // cout<<"distance: "<<distance<<endl;
        if(distance>maxdist)
            maxdist = distance;
        if(distance<mindist)
            mindist = distance;
    }
    cout<<"maxdist: "<<maxdist<<endl;
    cout<<"mindist: "<<mindist<<endl;

    vector<DMatch> goodmatches;   //定义最优的距离点集合
    for(int i=0;i<descritorL.rows;i++)
    {
        float distance = matches[i].distance;
        if(distance*100<=2*mindist)
        {
            //将查找到的最小距离添加到goodmatches中
            goodmatches.push_back(matches[i]);
        }
    }
    Mat matchimages;
    //绘制匹配点 点数大大减少
    drawMatches(imgL,keypointL,imgR,keypointR,goodmatches,matchimages);

    waitKey(0);
    while(1)
    {
        imshow("matchImages",matchimages);
        int c= waitKey(0);
        if (c==27)
            break;
    }

}


void ORB_Match(Mat &imgL, Mat imgR)
{
    int keypoints_num = 400;
    vector<KeyPoint>keypointL,keypointR;
    Mat descritorL,descritorR;

    Ptr<ORB> detector = ORB::create(keypoints_num);   //创建SURF对象

    detector->detectAndCompute(imgL,Mat(),keypointL,descritorL);
    detector->detectAndCompute(imgR,Mat(),keypointR,descritorR);


    chrono::steady_clock::time_point t1=chrono::steady_clock::now();

    BFMatcher matcher(NORM_HAMMING);  //hamming
    vector<DMatch> matches;      //匹配数据
    matcher.match(descritorL,descritorR,matches);   //描述匹配 将匹配结果存放至matches
    chrono::steady_clock::time_point t2=chrono::steady_clock::now();
    chrono::duration<double> time_used=chrono::duration_cast<chrono::duration<double>>(t2-t1);
    cout<<"BF方法的运行时间为："<<time_used.count()<<"秒。"<<endl;

    double min=10000;
    for(int i=0;i!=descritorL.rows;++i)
    {
        double dist=matches[i].distance;
        //cout<<dist<<endl;
        if(dist>=1)
            min=min<dist?min:dist;

    }
    cout<<min<<endl;
//select good matchers
    std::vector<DMatch> good_feature_match;
    for(int i=0;i!=descritorL.rows;++i)
    {
        double dist=matches[i].distance;
        if(dist<2.5*min&&dist>min)
            good_feature_match.push_back(matches[i]);
    }

    cout<<good_feature_match.size()<<endl;
    cv::Mat img_match,good_img_match;
    drawMatches(imgL,keypointL,imgR,keypointR,matches,img_match);
    drawMatches(imgL,keypointL,imgR,keypointR,matches,good_img_match);
    while(1)
    {
        imshow("BF匹配结果",img_match);
        imshow("处理后的BF匹配结果",good_img_match);

        int c= waitKey(0);
        if (c==27)
            break;
    }


}

int main()
{
    Mat imgL = imread("../picture/left/left0.jpg",-1);
    Mat imgR = imread("../picture/right/right0.jpg",-1);

    //ComputePyramid_brief(imgL);
   // ComputePyramid_brief(imgR);  //resize
    FLANN_Mh(imgL, imgR);
    ORB_Match(imgL, imgR);
    //ComputeStereoMatches(imgL , imgR);
    waitKey(0);



}

效果对比：

FLANN和BF匹配运行时间对比：

FLANN是类似最近邻的快速匹配库
它会根据数据本身选择最合适的算法来处理数据
比其他搜索算法快10倍
BF是暴力匹配

这里记录一个想法，在双目特征提取匹配过程中，其实不需要暴力匹配的，因为双目可以认为两个摄像头的距离保持不变，就是单目基础上平移了基线长度的距离的两个单目。那么共视的程度保持不变。那么左图摄像头的特征点，应该能在右图对应位置的某个范围内能沟找到该特征点。
这样则不用全局匹配。暂时不知道flann的原理