0

我一直试图让一些特征检测和匹配算法工作（我寻求手），但它们给我的结果很差。我找到了一个用于特征检测和匹配的模板代码，它使用单应矩阵和透视变换在场景图片（即相机输入）中定位匹配的模板对象。我一直在寻找参数校准或可能有帮助的东西，但我找不到太多信息，我在代码中使用的信息来自互联网上的不同站点。我通常会得到错误的结果，算法“认为”他们找到了足够好的匹配来绘制找到的对象。不幸的是，我几乎没有得到好的结果，要做到这一点，我通常需要一个同质的背景。我使用高斯平滑或中值滤波器作为预处理。我希望您对此有一些经验，并且可以帮助解决我可能不太了解的问题，参数化错误或不同的方法。谢谢！

这是一些图片： 未经预处理， 模板图像

这是我的代码（我写了评论）：

    Mat object = Mat(ImageObject);

    //surfThreshold=500
    int minHessian = surfThreshold;

    //surfUpright = false
    //surfExtended = false
    //surfNOctaves = 4
    //surfNOctaveLayers = 2
    //with SIFT i use SiftFeatureDetector etc.
    SurfFeatureDetector detector(minHessian, surfNOctaves, surfNOctaveLayers, surfExtended, surfUpright);

    //keypoints on template picture
    vector<KeyPoint> kp_object;
    detector.detect(object, kp_object);

    //with SIFT i use SiftDescriptorExtractor etc.
    SurfDescriptorExtractor extractor;

    //compute template descriptors
    Mat des_object;
    extractor.compute(object, kp_object, des_object);

    //I tried with bruteforce matcher too, didnt help much
    FlannBasedMatcher matcher;

    namedWindow("matches");

    vector<Point2f> obj_corners(4);

    //4 corners of the template image
    obj_corners[0] = cvPoint(0, 0);
    obj_corners[1] = cvPoint(object.cols, 0);
    obj_corners[2] = cvPoint(object.cols, object.rows);
    obj_corners[3] = cvPoint(0, object.rows);

    Mat des_image, img_matches;
    vector<KeyPoint> kp_image;
    vector<vector<DMatch > > matches;
    vector<DMatch > good_matches;
    vector<Point2f> obj;
    vector<Point2f> scene;
    vector<Point2f> scene_corners(4);
    Mat H;
    Mat image = Mat(ImageScene);

    //compute the scene image's keypoints and descriptors
    detector.detect(image, kp_image);
    extractor.compute(image, kp_image, des_image);

    //matching the descriptors with the matcher (2 best matches)
    matcher.knnMatch(des_object, des_image, matches, 2);

    //surfMatchDistance = 0.8
    for (int i = 0; i < min(des_image.rows - 1, (int)matches.size()); i++)
    {
        //we dont need the bad matches, so we keep only the good matches, when a match is within a specific threshold
        if ((matches[i][0].distance < surfMatchDistance*(matches[i][3].distance)) && ((int)matches[i].size() <= 2 && (int)matches[i].size()>0))
        {
            good_matches.push_back(matches[i][0]);
        }
    }

    //drawing out the remaining good matches
    drawMatches(object, kp_object, image, kp_image, good_matches, img_matches, Scalar::all(-1), Scalar::all(-1), vector<char>(), DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS);

    //if we have "enough" good matches, then we found the template object in the scene
    //goodMatchesCount = 7
    if (good_matches.size() >= goodMatchesCount)
    {
        for (int i = 0; i < good_matches.size(); i++)
        {
            //the good matches' keypoints
            obj.push_back(kp_object[good_matches[i].queryIdx].pt);
            scene.push_back(kp_image[good_matches[i].trainIdx].pt);
        }

        //we only keep the good points, so with the ransac classifier we disqualify the wrongs
        H = findHomography(obj, scene, CV_RANSAC, ransacThresold);

        //trying to find the object on the scene image, we compute its position and size with perspective transformation
        perspectiveTransform(obj_corners, scene_corners, H);

        //give the found object a frame on the scene image
        line(img_matches, scene_corners[0] + Point2f(object.cols, 0), scene_corners[1] + Point2f(object.cols, 0), Scalar(0, 255, 0), 4);
        line(img_matches, scene_corners[1] + Point2f(object.cols, 0), scene_corners[2] + Point2f(object.cols, 0), Scalar(0, 255, 0), 4);
        line(img_matches, scene_corners[2] + Point2f(object.cols, 0), scene_corners[3] + Point2f(object.cols, 0), Scalar(0, 255, 0), 4);
        line(img_matches, scene_corners[3] + Point2f(object.cols, 0), scene_corners[0] + Point2f(object.cols, 0), Scalar(0, 255, 0), 4);
    }

    imshow("matches", img_matches);
4

0 回答 0