OpenCVSharp 4.5 定位已知物体（Features2D+Homography）

用 OpenCVSharp 4.5 跑一遍 OpenCV 官方教程

原 OpenCV 官方教程链接：OpenCV: Features2D + Homography to find a known object

核心函数：

•  cv::findHomography
• cv::perspectiveTransform

using System;
using OpenCvSharp;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using System.Threading.Tasks;
using OpenCvSharp.XFeatures2D;

namespace ConsoleApp1
{
    class tutorial48 : ITutorial
    {
        public void Run()
        {
            string input1 = @"I:\csharp\images\match1.png";
            string input2 = @"I:\csharp\images\match2.png";
            Mat img1 = Cv2.ImRead(input1, ImreadModes.Color);
            Mat img2 = Cv2.ImRead(input2, ImreadModes.Color);
            if (img1.Empty() || img2.Empty())
            {
                Console.WriteLine("无法打开图像文件");
                return;
            }
            //第一步：用 SURF Detector 检测关键点，然后计算特征描述向量
            int minHessian = 400;
            SURF detector = SURF.Create(minHessian);
            KeyPoint[] keypoints1, keypoints2;
            Mat descriptor1 = new Mat(), descriptor2 = new Mat();
            detector.DetectAndCompute(img1, null, out keypoints1, descriptor1);
            detector.DetectAndCompute(img2, null, out keypoints2, descriptor2);

            //第二步：用基于 FLANN 的方法匹配特征描述向量
            DescriptorMatcher matcher = DescriptorMatcher.Create("FlannBased");
            DMatch[][] matches = matcher.KnnMatch(descriptor1, descriptor2, 2);

            //用 lowe 比法则过滤匹配结果
            double ratio_thresh = 0.8;
            List<DMatch> good_matches = new List<DMatch>();
            for (int i = 0; i < matches.Length; i++)
            {
                if (matches[i][0].Distance < ratio_thresh * matches[i][1].Distance)
                {
                    good_matches.Add(matches[i][0]);
                }
            }
            //第三步：画出匹配线
            Mat image_matches = new Mat();
            Cv2.DrawMatches(img1, keypoints1, img2, keypoints2, good_matches, image_matches,
                null, null, null, DrawMatchesFlags.NotDrawSinglePoints);

            //第四步：锚定物体
            Point2d[] obj = new Point2d[good_matches.Count()], scene = new Point2d[good_matches.Count()];
            for (int i = 0; i < good_matches.Count(); i++)
            {
                obj[i] = keypoints1[good_matches[i].QueryIdx].Pt.ToPoint();
                scene[i] = keypoints2[good_matches[i].TrainIdx].Pt.ToPoint();
            }


            Mat H = Cv2.FindHomography(obj, scene, HomographyMethods.Ransac, 3, null);
            Point2f[] obj_corners = new Point2f[4];
            obj_corners[0] = new Point(0, 0);
            obj_corners[1] = new Point((float)img1.Cols, 0);
            obj_corners[2] = new Point((float)img1.Cols, (float)img1.Rows);
            obj_corners[3] = new Point(0, (float)img1.Rows);

            Point2f[] scene_corners = Cv2.PerspectiveTransform(obj_corners, H);

            //第五步：画出锚定物体
            Point p1 = new Point(scene_corners[0].X + img1.Cols, scene_corners[0].Y);
            Point p2 = new Point(scene_corners[1].X + img1.Cols, scene_corners[1].Y);
            Cv2.Line(image_matches, p1, p2, new Scalar(0, 255, 0), 4);

            p1 = new Point(scene_corners[1].X + img1.Cols, scene_corners[1].Y);
            p2 = new Point(scene_corners[2].X + img1.Cols, scene_corners[2].Y);
            Cv2.Line(image_matches, p1, p2, new Scalar(0, 255, 0), 4);

            p1 = new Point(scene_corners[2].X + img1.Cols, scene_corners[2].Y);
            p2 = new Point(scene_corners[3].X + img1.Cols, scene_corners[3].Y);
            Cv2.Line(image_matches, p1, p2, new Scalar(0, 255, 0), 4);

            p1 = new Point(scene_corners[3].X + img1.Cols, scene_corners[3].Y);
            p2 = new Point(scene_corners[0].X + img1.Cols, scene_corners[0].Y);
            Cv2.Line(image_matches, p1, p2, new Scalar(0, 255, 0), 4);

            Cv2.ImShow("Matches", image_matches);
            Cv2.WaitKey();
        }
    }
}