基于SIFT的视频跟踪

有些事现在不做 一辈子都不会做了

1 YEAR AGO

本文中采用的SIFT提取视频中的和待检测图像特征点，并利用特征点之间的映射关系找到视频中待检测物体的位置，绘制出绿色的边界并显示出来。

检测视频
视频地址

图片：待检测的图片（仙剑奇侠传5的游戏盒子）

程序源代码

#include 
#include 
#include 
#include 
#include 
#include 

using namespace cv;

int _sift(Mat &img_object, Mat &img_scene);

int main(int argc, char *argv[])
{
    //读取视频
    VideoCapture vc;
    vc.open("C:\\Users\\Arthur\\Desktop\\SRC\\video.mp4");
    double rate = vc.get(CV_CAP_PROP_FPS);//帧率
    int delay = 1000 / rate;
    Mat img = imread("C:\\Users\\Arthur\\Desktop\\SRC\\temp.jpg", CV_LOAD_IMAGE_COLOR);//模版图像
    VideoWriter vw;
    vw.open("C:\\Users\\Arthur\\Desktop\\SRC\\sift.avi",
         (int)vc.get(CV_CAP_PROP_FOURCC), // 也可设为CV_FOURCC_PROMPT，在运行时选取  
         (double)vc.get(CV_CAP_PROP_FPS), // 视频帧率  
        cv::Size((int)vc.get(CV_CAP_PROP_FRAME_WIDTH), (int)vc.get(CV_CAP_PROP_FRAME_HEIGHT)), // 视频大小  
         true); // 是否输出彩色视频  

    namedWindow("1");
    if (vc.isOpened())
    {

        while (1)
        {
            Mat frame;
            //原始图像每5帧图像取1帧进行处理
            for (int i = 0; i < 5; i++)
            {
                vc.read(frame);
            }

            if (frame.empty())
            {
                break;
            }
            _sift(img, frame);
            imshow("1", frame);
            vw << frame;
            waitKey(1);
        }
    }
    vc.release();
    return 0;
}

int _sift(Mat &img_object, Mat &img_scene)
{
    //Mat img_object = imread("1.jpg", CV_LOAD_IMAGE_COLOR);
    //Mat img_scene = imread("2.jpg", CV_LOAD_IMAGE_COLOR);
    double t = (double)getTickCount();
    if (!img_object.data || !img_scene.data)
    {
        std::cout << "Error reading images!" << std::endl;
        return -1;
    }

    //检测SIFT特征点
    int minHeassian = 400;
    SiftFeatureDetector detector(minHeassian);

    std::vector keypoints_object, keypoints_scene;

    detector.detect(img_object, keypoints_object);
    detector.detect(img_scene, keypoints_scene);

    //计算特征向量
    SiftDescriptorExtractor extractor;

    Mat descriptors_object, descriptors_scene;

    extractor.compute(img_object, keypoints_object, descriptors_object);
    extractor.compute(img_scene, keypoints_scene, descriptors_scene);

    //利用FLANN匹配算法匹配特征描述向量
    FlannBasedMatcher matcher;
    std::vector matches;
    matcher.match(descriptors_object, descriptors_scene, matches);

    double max_dist = 0; double min_dist = 100;

    //快速计算特征点之间的最大和最小距离
    for (int i = 0; i < descriptors_object.rows; i++)
    {
        double dist = matches[i].distance;
        if (dist < min_dist) min_dist = dist;
        if (dist > max_dist) max_dist = dist;
    }

    printf("---Max dist: %f \n", max_dist);
    printf("---Min dist: %f \n", min_dist);

    //只画出好的匹配点(匹配特征点之间距离小于3*min_dist)
    std::vector good_matches;

    for (int i = 0; i < descriptors_object.rows; i++)
    {
        if (matches[i].distance < 3 * min_dist)
            good_matches.push_back(matches[i]);
    }

    Mat img_matches;
    drawMatches(img_object, keypoints_object, img_scene, keypoints_scene,
        good_matches, img_matches, Scalar::all(-1), Scalar::all(-1),
        vector(), DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS);


    //定位物体/
    std::vector obj;
    std::vector scene;

    for (int i = 0; i < good_matches.size(); i++)
    {
        //从好的匹配中获取特征点
        obj.push_back(keypoints_object[good_matches[i].queryIdx].pt);
        scene.push_back(keypoints_scene[good_matches[i].trainIdx].pt);
    }

    //找出匹配特征点之间的变换
    Mat H = findHomography(obj, scene, CV_RANSAC);

    //得到image_1的角点(需要寻找的物体)
    std::vector obj_corners(4);
    obj_corners[0] = cvPoint(0, 0);
    obj_corners[1] = cvPoint(img_object.cols, 0);
    obj_corners[2] = cvPoint(img_object.cols, img_object.rows);
    obj_corners[3] = cvPoint(0, img_object.rows);
    std::vector scene_corners(4);

    //匹配四个角点
    perspectiveTransform(obj_corners, scene_corners, H);

    //画出匹配的物体两个匹配的图片
    //line(img_matches, scene_corners[0] + Point2f(img_object.cols, 0), scene_corners[1] + Point2f(img_object.cols, 0), Scalar(0,255,0), 4);
    //line(img_matches, scene_corners[1] + Point2f(img_object.cols, 0), scene_corners[2] + Point2f(img_object.cols, 0), Scalar(0,255,0), 4);
    //line(img_matches, scene_corners[2] + Point2f(img_object.cols, 0), scene_corners[3] + Point2f(img_object.cols, 0), Scalar(0,255,0), 4);
    //line(img_matches, scene_corners[3] + Point2f(img_object.cols, 0), scene_corners[0] + Point2f(img_object.cols, 0), Scalar(0,255,0), 4);

    //匹配之后画出匹配图形的轮廓
    line(img_scene, scene_corners[0], scene_corners[1], Scalar(0, 255, 0), 4);
    line(img_scene, scene_corners[1], scene_corners[2], Scalar(0, 255, 0), 4);
    line(img_scene, scene_corners[2], scene_corners[3], Scalar(0, 255, 0), 4);
    line(img_scene, scene_corners[3], scene_corners[0], Scalar(0, 255, 0), 4);

    //imshow("Good Matches & Object detection", img_matches);
    //imshow("识别图像", img_scene);
    t = 1000 * ((double)getTickCount() - t) / getTickFrequency();
    std::cout << "the time is :" << t << std::endl;

    //waitKey(0);
}

为了达到实时处理的效果，可以使用SURF替代SIFT，不过处理效果不如SIFT理想。
本文中的程序采用的是Opencv的2.4.8版本，在2.4.5版本以后可以结合VS2013和Image Watch插件，在调试的时候可以实时显示出图像，方便观测结果，再也不需要用各种imshow（）来显示结果。
图片：Image Watch直接显示内存中图片信息

SIFT原理解析可以查看SIFT算法详解 这篇博文