OpenCV：如何正确使用findHomography

Question

我一直在尝试使用OpenCV的findhomography和perspectivetransform来获取从一组点到另一组的转换矩阵。

我为每个数据集生成50个随机点，并在屏幕上绘制它们。然后我要求findHomography从两组点计算变换矩阵，并用它来变换原始集合的角。透视变换的结果将在屏幕上绘制。

问题是矩阵findhomography产生的不正确，即使忽略了异常值的问题。我想在功能匹配问题中使用此示例。

这是我的测试代码。

    //


    Mat outImg = new Mat(500,1000,CvType.CV_32FC3,new Scalar(0,0,0));

    //make obj test points
    List<Point> p1 = new ArrayList<Point>();
    for(int i=0;i<50;i++){
        p1.add(new OpenCVTemplateMatcher().generateRandom2DPoint(50,50,450,450));
    }
    //make scene test points
    List<Point> p2 = new ArrayList<Point>();
    for(int i=0;i<50;i++){
        p2.add(new OpenCVTemplateMatcher().generateRandom2DPoint(550,50,950,450));
    }

    System.out.println(p1.size());
    //draw the points
    for(Point p:p1){
        Core.circle(outImg, p, 1, new Scalar(255,0,255),2);
    }
    for(Point p:p2){
        Core.circle(outImg, p, 1, new Scalar(0,255,0),2);
    }

    //find bounding boxes on points and draw them
    MatOfPoint2f mp1 = new MatOfPoint2f(); mp1.fromList(p1);
    MatOfPoint2f mp2 = new MatOfPoint2f(); mp2.fromList(p2);
    RotatedRect r1 = Imgproc.minAreaRect(mp1);
    RotatedRect r2 = Imgproc.minAreaRect(mp2);
    Point[] v1 = new Point[4]; r1.points(v1);
    Point[] v2 = new Point[4]; r2.points(v2);

    Core.line(outImg, v1[0], v1[1], new Scalar(0, 255, 0),1);
    Core.line(outImg, v1[1], v1[2], new Scalar(0, 255, 0),1);
    Core.line(outImg, v1[2], v1[3], new Scalar(0, 255, 0),1);
    Core.line(outImg, v1[3], v1[0], new Scalar(0, 255, 0),1);

    Core.line(outImg, v2[0], v2[1], new Scalar(255, 255, 0),1);
    Core.line(outImg, v2[1], v2[2], new Scalar(255, 255, 0),1);
    Core.line(outImg, v2[2], v2[3], new Scalar(255, 255, 0),1);
    Core.line(outImg, v2[3], v2[0], new Scalar(255, 255, 0),1);

    //show the corners
    for(int i=0;i<4;i++){
        Core.circle(outImg, v1[i], 3, new Scalar(200,250,50),2);
        Core.circle(outImg, v2[i], 3, new Scalar(0,238,250),2);
    }

    MatOfPoint2f p1Corners = new MatOfPoint2f(); p1Corners.fromArray(v1);
    MatOfPoint2f p2Corners = new MatOfPoint2f(); p2Corners.fromArray(v2);
    //find transform as H
    //Mat H = Calib3d.findHomography(p1Corners, p2Corners,Calib3d.RANSAC, 5);
    Mat H = Calib3d.findHomography(mp1, mp2,0, 5);
    //H = findTransform(mp1,mp2);

    //find the transform of H from 1'(original)s corners
    Mat orig_corners = new Mat(4,1,CvType.CV_32FC2);
    Mat transformed_corners = new Mat(4,1,CvType.CV_32FC2);
    orig_corners.put(0, 0, new double[] {v1[0].x,v1[0].y});
    orig_corners.put(1, 0, new double[] {v1[1].x,v1[1].y});
    orig_corners.put(2, 0, new double[] {v1[2].x,v1[2].y});
    orig_corners.put(3, 0, new double[] {v1[3].x,v1[3].y});

    Core.perspectiveTransform(orig_corners,transformed_corners,H);

    Core.line(outImg, new Point(transformed_corners.get(0,0)), new Point(transformed_corners.get(1,0)), new Scalar(0, 255, 255),4);
    Core.line(outImg, new Point(transformed_corners.get(1,0)), new Point(transformed_corners.get(2,0)), new Scalar(0, 255, 255),4);
    Core.line(outImg, new Point(transformed_corners.get(2,0)), new Point(transformed_corners.get(3,0)), new Scalar(0, 255, 255),4);
    Core.line(outImg, new Point(transformed_corners.get(3,0)), new Point(transformed_corners.get(0,0)), new Scalar(0, 255, 255),4);

    Highgui.imwrite("test.jpg", outImg);

这是一个结果

我发现findhomography在每组中除了4个角以外的任意点数时根本不起作用，但问题变得微不足道。我用错了吗？如果我想编写自己的函数来查找2个矩阵之间的转换，我将如何去做呢？