烟雾环境下对RGB图SIFT特征提取，匹配及深度图有效值的影响

烟雾环境下对RGB图SIFT特征提取，匹配及深度图有效值的影响

1 环境配置

1.1 opencv_contrib

        已安装opencv3.4.7，如果需要SIFT，SURF等算法的时候，需要配置OpenCV扩展模块(opencv_contrib),版本需与原opencv版本一致。

        OpenCV扩展模块(opencv_contrib)下载地址:
https://github.com/opencv/opencv_contrib/releases

编译过程:

cd <opencv_build_directory>
sudo make clean
cmake -DOPENCV_EXTRA_MODULES_PATH=<opencv_contrib>/modules  -DOPENCV_ENABLE_NONFREE=ON <opencv_main>
make
sudo make install

1.2 MathGL-数据可视化

sduo apt-get install mathgl

或者源码安装，参考链接：

http://mathgl.sourceforge.net/
http://mathgl.sourceforge.net/doc_en/

1.3 实验数据

        实验室410采集的图片：

1.3.1 Left

1.3.2 Right

1.3.3 depth

1.4 实验环境

        Ubuntu 18.04 + CLion

2 准备工作

2.1 对大批量图片文件名的导出

cd 到当前目录
ls left/ > left.txt
ls right/ > right.txt
ls depth/ > depth.txt

2.2 SIFT特征

        contrastThreshold默认=0.04,对比阈值，过滤半均匀（低对比度）区域的弱特征，阈值越大，检测到的特征越少。

        edgeThreshold默认=10, 边缘阈值，过滤掉边缘状特征，阈值越大，过滤掉的特征越少，更多的特征被保留。

        考察以上两个阈值改变对SIFT特征提取与匹配的影响。

2.3 深度图

        烟雾对深度图的影响从深度值无效点的比例、整张图的平均深度来观察。

3 SIFT特征提取与匹配

        首先提取图像的sift特征点，再对相邻两帧的特征点做匹配，形成特征点对，最后计算图像变换的单应矩阵,RANSAC剔除误匹配点。

3.1 C++代码

// C++
#include<iostream>
#include<fstream>
#include<vector>
// OpenCV
#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/features2d/features2d.hpp>
#include <opencv2/xfeatures2d/nonfree.hpp>
#include <opencv2/calib3d/calib3d.hpp>
//MathGL
#include <mgl2/mgl.h>

using namespace std;
using namespace cv;
vector<int> *siftFeature(double contrastThreshold, double edgeThreshold, const std::string &txtfilename,
                         const std::string &foldername) {
   
    ifstream fin;
    fin.open(txtfilename, ios::in);
    if (!fin.is_open()) {
   
        cerr << "文件打开失败！" << endl;
        return 0;
    }
    std::string filename_1;
    getline(fin, filename_1);

    std::string filename_2;
    
    //检测结果
    vector<int> *num_feature = new vector<int>[2];

    //检测到的特征点数量
    vector<int> num_keyPoints;

    //匹配的特征点对
    vector<int> num_matches;
    //读取待提取和匹配的两张图
    Mat srcImg1 = imread(foldername + filename_1);
    Mat srcImg2;

    //定义SIFT特征检测类对象
    cv::Ptr<cv::xfeatures2d::SIFT> sift = cv::xfeatures2d::SIFT::create(1000, 3,
                                                                        contrastThreshold, edgeThreshold,
                                                                        1.6);
                                                                           

   //定义KeyPoint变量
    vector<KeyPoint> keyPoints1;
    vector<KeyPoint> keyPoints2;
    //特征点检测
    sift->detect(srcImg1, keyPoints1);

    //绘制特征点(关键点)
    Mat feature_pic1, feature_pic2;
    drawKeypoints(srcImg1, keyPoints1, feature_pic1, Scalar::all(-1), DrawMatchesFlags::DEFAULT);
    / drawKeypoints(srcImg1, keyPoints1, feature_pic1, Scalar(0, 255, 0), DrawMatchesFlags::DRAW_RICH_KEYPOINTS);

    //计算特征点描述符 / 特征向量提取
    Mat description1;
    sift->compute(srcImg1, keyPoints1, description1);
    Mat description2;
    
    //计数
    int count = 0;

    num_keyPoints.push_back(keyPoints1.size());
    while (getline(fin, filename_2)) {
   

        srcImg2 = imread(foldername + filename_2);

        //特征点检测
        sift->detect(srcImg2, keyPoints2);
         cout << "--------------------" << "特征提取" << "--------------------" << endl;
        count++;
        cout << "********************" << count << "********************" << endl;

        cout << "SITF特征检测提取到的特征点数:" << keyPoints1.size() << endl;

        count++;
        cout << "********************" << count << "********************" << endl;
         num_keyPoints.push_back(keyPoints2.size());
        cout << "SITF特征检测提取到的特征点数:" << keyPoints2.size() << endl;

        //绘制特征点(关键点)
        drawKeypoints(srcImg2, keyPoints2, feature_pic2, Scalar::all(-1), DrawMatchesFlags::DEFAULT);

        // drawKeypoints(srcImg2, keyPoints2, feature_pic2, Scalar(0, 255, 0), DrawMatchesFlags::DRAW_RICH_KEYPOINTS);

        //显示结果
        // imshow("feature1", feature_pic1);
        //imshow("feature2", feature_pic2);
        //计算特征点描述符 / 特征向量提取
        sift->compute(srcImg2, keyPoints2, description2);

        //进行BFMatch暴力匹配
        //BruteForceMatcher<L2<float>>matcher;    //实例化暴力匹配器
        FlannBasedMatcher matcher;  //实例化FLANN匹配器  //快速最近邻逼近搜索
        vector<DMatch> matches;   //定义匹配结果变量
        matcher.match(description1, description2, matches);  //实现描述符之间的匹配

        double max_dist = 0;
        double min_dist = 10000;
        for (int i = 0; i < matches.size(); i++) {
   
            double dist = matches[i].distance;
            if (dist < min_dist)
                min_dist = dist;
            if (dist > max_dist)
                max_dist = dist;
        }
        cout << "最大距离：" << max_dist << endl;
        cout << "最小距离：" << min_dist << endl;
        
        //筛选出较好的匹配点
        vector<DMatch> good_matches;
        double dThreshold = 0.5;    //匹配的阈值，越大匹配的点数越多
        for (int i = 0; i < matches.size(); i++) {
   
            if (matches[i].distance < dThreshold * max_dist) {
   
                good_matches.push_back(matches[i]);
            }
        }
        /