2D、3D直线鲁棒性拟合(Ransac + SVD)与求近似交点

一、2D直线求交点

用一般式表示直线方程：ax + by + c = 0

 1 #include<opencv2/opencv.hpp>
 2 #include <iostream>
 3 using namespace std;
 4 /**************************************************************************************
 5 
 6 //function: 2D平面上，求解两条直线交点
 7 
 8 //input:    line1: 直线方程:a0 * x + b0 * y + c0 = 0
 9             line2: 直线方程:a1 * x + b1 * y + c1 = 0
10 
11 //output:   point: 交点
12 
13 //return:   -1: line1平行于line2
14             0: 有交点值返回
15 
16 **************************************************************************************/
17 
18 int CalculateIntersectionPoint(const cv::Vec3d& line1, const cv::Vec3d& line2, cv::Point2d& point)
19 {
20     // ...
21     double a0 = line1[0];
22     double b0 = line1[1];
23     double c0 = line1[2];
24     
25     double a1 = line2[0];
26     double b1 = line2[1];
27     double c1 = line2[2];
28 
29     double D = a0 * b1 - a1 * b0; // ...
30     if (abs(D) < 1e-5)
31         return -1;
32     point.x = (b0 * c1 - b1 * c0) / D;
33     point.y = (a1 * c0 - a0 * c1) / D;
34     return 0;
35 }
36 
37 int main()
38 {
39     cv::Vec3d line1(0, 1, -4);
40     cv::Vec3d line2(1, -2, 2);
41     cv::Point2d point;
42     if (CalculateIntersectionPoint(line1, line2, point) == 0)
43     {
44         cout << point.x << "," << point.y;
45     }
46     return 1;
47 }

二、3D直线求交“近似”点，因为有交点的概率几乎为0。

 

 

 

 

 

 

 

 

 

 

 

1、直线拟合

      首先3D直线方程一般有三种：1、用两个平面交线表示；2、用两个点表示；3、用一个点(x0 y0 z0) 和一个向量(a b c)表示，

即：(x - x0) / a = (y - y0) / b = (z - z0) / c  , 本博客选用第三种。

      对于3D点的直线拟合有两种方式1：SVD分解法 2：先用ransac选择内点集合，然后再用SVD分解法拟合内点集合

如图是以下Matlab代码拟合的结果：

 1 clc;
 2 clear all;
 3 close all;
 4 % 随机生成一组点，这写点距离直线l比较近，l的过点[1,1,1]，方向向量为[1,2,3]
 5 lineData=bsxfun(@plus, [1,1,1], (-1:.1:1).'*[1,2,3]);
 6 Noise=rand(size(lineData))*.1;
 7 lineData=lineData+Noise;
 8 scatter3(lineData(:,1), lineData(:,2), lineData(:,3),'filled')
 9 hold on;
10 
11 % 拟合的直线必过所有坐标的算数平均值
12 xyz0=mean(lineData,1);
13 
14 % 协方差矩阵奇异变换，与拟合平面不同的是
15 % 所得直线的方向实际上与最大奇异值对应的奇异向量相同
16 centeredLine=bsxfun(@minus,lineData,xyz0); % 去除质心
17 [U,S,V]=svd(centeredLine);% SVD
18 direction=V(:,1);
19 
20 % 画图
21 t=-8:0.1:8;
22 xx=xyz0(1)+direction(1)*t;
23 yy=xyz0(2)+direction(2)*t;
24 zz=xyz0(3)+direction(3)*t;
25 plot3(xx,yy,zz)

 

 

          下图是基于Ransac框架下的空间直线拟合，左边有杂点，通过随机抽样抽出符合条件的点，即：内点，下图右边是对内点采用SVD分解拟合的效果。Ransac也没什么好说的，精髓在于：1、内点集比例要高，2、累计误差足够小就行。代码如下：

test2_Fit3dLineRansac.m （这个是主函数，读取数据、执行Ransac + SVD流程）

 1 clc;
 2 clear all;
 3 close all;
 4 
 5 %% <2> 实测数据测试
 6 % 竖直 
 7 data = [];
 8 % % % left
 9 data = [data; importdata('TestData2/竖直方向/Point3D_V_Left20200311163646.csv')];
10 data = [data; importdata('TestData2/竖直方向/Point3D_V_Left20200311163759.csv')];
11 data = [data; importdata('TestData2/竖直方向/Point3D_V_Left20200311163840.csv')];
12 data = [data; importdata('TestData2/竖直方向/Point3D_V_Left20200311163927.csv')];
13 data = [data; importdata('TestData2/竖直方向/Point3D_V_Left20200311164009.csv')];
14 data = [data; importdata('TestData2/竖直方向/Point3D_V_Left20200311164045.csv')];
15 
16 % % right
17 % data = [data; importdata('TestData2/竖直方向/Point3D_V_Right20200311163646.csv')];
18 % data = [data; importdata('TestData2/竖直方向/Point3D_V_Right20200311163759.csv')];
19 % data = [data; importdata('TestData2/竖直方向/Point3D_V_Right20200311163840.csv')];
20 % data = [data; importdata('TestData2/竖直方向/Point3D_V_Right20200311163927.csv')];
21 % data = [data; importdata('TestData2/竖直方向/Point3D_V_Right20200311164009.csv')];
22 % data = [data; importdata('TestData2/竖直方向/Point3D_V_Right20200311164045.csv')];
23 
24 % 水平
25 % data = [data; importdata('TestData2/水平方向/Point3D_H20200311163646.csv')];
26 % data = [data; importdata('TestData2/水平方向/Point3D_H20200311163759.csv')];
27 % data = [data; importdata('TestData2/水平方向/Point3D_H20200311163840.csv')];
28 % data = [data; importdata('TestData2/水平方向/Point3D_H20200311163927.csv')];
29 % data = [data; importdata('TestData2/水平方向/Point3D_H20200311164009.csv')];
30 % data = [data; importdata('TestData2/水平方向/Point3D_H20200311164045.csv')];
31 
32 data = data(:,1:3); % 取出前三维
33 figure(1);
34 subplot(121); plot3(data(:,1), data(:,2), data(:,3), 'k.'); 
35 % ransac挑选内点
36 [diatance_vec, itr_num, paraN,point,inLier_Index, outLier_Index] = Fit3DLineRansac(data, 0.0000005, 50, 0.01);
37 data_line=data(inLier_Index,:);
38 
39 subplot(122); plot3(data_line(:,1),data_line(:,2),data_line(:,3),'g*'); grid on; hold on;
40 
41 % 拟合内点
42 [direction, xyz] = Fit3DLine(data_line); 
43 t=-0.2:0.001:0.2;
44 x=xyz(1)+direction(1)*t;
45 y=xyz(2)+direction(2)*t;
46 z=xyz(3)+direction(3)*t;
47 plot3(x,y,z, 'm.'); hold on;
48 
49 % X=[point(1)-20*paraN(1) point(1)+20*paraN(1)];
50 % Y=[point(2)-20*paraN(2) point(2)+20*paraN(2)];
51 % Z=[point(3)-20*paraN(3) point(3)+20*paraN(3)];
52 %  plot3(X,Y,Z,'b');grid on; 
53 % 内点比例、SSE、SSE平均值
54 SSE = sum(diatance_vec);
55 SSE_mean = mean(diatance_vec);
56 radio = length(data_line) / length(data);
57 sprintf('内点比例 = %f, SSE = %f,SSE_mean = %f ', radio, SSE, SSE_mean)
58 
59 a = (direction/norm(direction))'
View Code

Fit3DLineRansac.m(Ransac 选出内点集)

 1 function [diatance, i, paraN,point,inLier_Index, outLier_Index] = Fit3DLineRansac(data, Threshold, Max_Iter, delta)
 2 % Threshold=0.3;
 3 % Max_Iter=5000;
 4 
 5 %输出参数分别为拟合的直线的方向向量，经过的点，所有的内点，以及内点的索引号
 6 num_points=size(data,1);
 7 pretotal=0;
 8 
 9 for i=1:Max_Iter
10     idx = randperm(num_points,2);% [1, num_points]区间中随机选两个数
11     sample=data(idx,:);%随机选择两个点
12     x=sample(:,1); y=sample(:,2); z=sample(:,3);
13     %判断距离是否过近 排除距离过近的情况
14     n12=[x(1)-x(2),y(1)-y(2),z(1)-z(2)];
15     dist12=norm(n12); 
16     if dist12 < delta % 0.01 有待商榷????????<1>取点尽量分开 <2> 不重复取点
17         continue
18     end
19     %拟合直线
20     %计算点到直线的距离
21     p_p0=data-[x(1) y(1) z(1)];   % 所有data 对应的方向向量集合
22     n12=n12/norm(n12);            % sample对应的归一化方向向量
23     nAll=repmat(n12,num_points,1);% n12拷贝num_points次
24     % 设点p(x(1) y(1) z(1))
25     % 每一个样本点和p作叉积，叉积公式：a×b = |a|*|b|*sin(theta)
26     % 由于n12归一化了，即:|a| = 1，所以 a×b 就是垂线段，即：点到直线距离
27     cross_value=cross(p_p0,nAll);
28     Distance=sum(cross_value.*cross_value,2);% 所有样本点到直线距 离的平方 的集合
29     inLierIndex=find(Distance < Threshold);
30     outLierIndex=find(Distance >= Threshold);
31     
32     total=size(inLierIndex,1);
33     if total > pretotal
34         pretotal=total;          % 更新内点数量
35         paraN=n12;               % 更新方向向量
36         point=[x(1) y(1) z(1)];  % 画图用
37         inLier_Index = inLierIndex;% 更新内点索引
38         outLier_Index = outLierIndex;
39 %         data_line=data(inLier_Index,:); % 不合理
40         diatance = Distance(inLierIndex);
41     end
42 end
View Code

Fit3DLine.m （利用SVD分解法对上述内点集进行拟合）

 1 function [direction, xyz0] = Fit3DLine(lineData)
 2 %3DLINEFIT 三维点直线拟合
 3 %   direction:方向向量，
 4 %   xyz0:所有样本点质心
 5 xyz0 = mean(lineData,1);
 6 % 协方差矩阵奇异变换，与拟合平面不同的是
 7 % 所得直线的方向实际上与最大奇异值对应的奇异向量相同
 8 centeredLine=bsxfun(@minus,lineData,xyz0); % 每一个样本减去质心
 9 [U,S,V]=svd(centeredLine);% SVD分解
10 direction=V(:,1);
11 end
View Code

2、直线求交点

     由于工程上，两根空间直线大概率没有严格意义的交点，那我们只能求解近似交点。 如图，3D直线L1、L2没有严格意义交点，于是我们可以找到L1、L2的公共垂线，求得垂线分别与L1、L2的交点为a、b。对a、b求平均值就是“近似交点”。

 

       近似交点的数学问题推导如下图：

 

      上述的推导用Matlab实现了。

Intersection3DPoint.m(3D直线求交点)：

 1 function [direction2,p0, p1] = Intersection3DPoint(direction0,xyz0, direction1, xyz1)
 2 %求两条不共线、不平行且不相交的直线的近似交点
 3 
 4 % 功能：用3D直线的方向向量和一点表示直线方程
 5 % direction: 3D直线方向向量； 
 6 % xyz: 3D直线上一点
 7 % (注：xyz为方向向量，由定义可知，每一个分量不能为0，否则下面矩阵A不可逆；
 8 %  故：当xyz中存在0的时候，可以认为给定一个极小量，使其不为零)
 9 
10 % 直线1： (x - x0)/a0 = (y - y0)/b0 = (z - z0)/c0
11 a0 = direction0(1, 1);
12 b0 = direction0(1, 2);
13 c0 = direction0(1, 3);
14 x0 = xyz0(1, 1);
15 y0 = xyz0(1, 2);
16 z0 = xyz0(1, 3);
17 
18 % 直线2： (x - x1)/a1 = (y - y1)/b1 = (z - z1)/c1
19 a1 = direction1(1, 1);
20 b1 = direction1(1, 2);
21 c1 = direction1(1, 3);
22 x1 = xyz1(1, 1);
23 y1 = xyz1(1, 2);
24 z1 = xyz1(1, 3);
25 
26 % 求公垂线方向向量
27 direction2 = cross(direction0, direction1);
28 a2 = direction2(1, 1);
29 b2 = direction2(1, 2);
30 c2 = direction2(1, 3);
31 
32 % ...
33 % ...
34 % ...
35 % 求解方程Ax = b, （x = [β;α]）
36 A = [a1 * b2 - a2 * b1    a2 * b0 - a0 * b2; ...
37      b1 * c2 - b2 * c1    b2 * c0 - b0 * c2];
38 b = [a2 * (y1 - y0) + b2 * (x0 - x1);b2 * (z1 - z0) + c2 * (y0 - y1)];
39 
40 x = A\b; 
41 
42 alpha = x(2, 1); % α
43 beta = x(1, 1);
44 
45 p0 = [alpha * a0 + x0 alpha * b0 + y0 alpha * c0 + z0];
46 p1 = [beta * a1 + x1 beta * b1 + y1 beta * c1 + z1];
47 end
View Code

      mainV1.m(3D直线SVD拟合 + 直线求交点 测试比较理想数据)

 1 %% note: 确保严谨，代码后续加入：1、判定是否同一根线； 2、是否平行; 3、是否真的相交 4、方向向量不能为0
 2 clc;
 3 clear all;
 4 close all;
 5 %% <1> 数据读取、处理
 6 % 竖直 
 7 points3d_v = [];
 8 points3d_v = [points3d_v; importdata('TestData1/竖直方向/Point3D20200226164534.csv')];
 9 points3d_v = [points3d_v; importdata('TestData1/竖直方向/Point3D20200226165026.csv')];
10 points3d_v = [points3d_v; importdata('TestData1/竖直方向/Point3D20200226165100.csv')];
11 points3d_v = points3d_v(:,1:3); % 取出前三维
12 % 水平
13 points3d_h = [];
14 points3d_h = [points3d_h; importdata('TestData1/水平方向/Point3D20200226163545.csv')];
15 points3d_h = [points3d_h; importdata('TestData1/水平方向/Point3D20200226163621.csv')];
16 points3d_h = [points3d_h; importdata('TestData1/水平方向/Point3D20200226163654.csv')];
17 points3d_h = points3d_h(:,1:3); % 取出前三维
18 % 显示
19 figure;
20 plot3(points3d_v(:,1), points3d_v(:,2), points3d_v(:,3), 'g.'); hold on;
21 plot3(points3d_h(:,1), points3d_h(:,2), points3d_h(:,3), 'b.'); hold on;
22 
23 %% <2> 拟合
24 
25 % direction: 方向向量 
26 % xyz：3D直线上一点
27 
28 % 3D直线L1拟合(横向)
29 [direction_v, xyz_v] = Fit3DLine(points3d_v); 
30 tv=-1:0.1:1;
31 xv=xyz_v(1)+direction_v(1)*tv;
32 yv=xyz_v(2)+direction_v(2)*tv;
33 zv=xyz_v(3)+direction_v(3)*tv;
34 plot3(xv,yv,zv, 'r'); hold on;
35 
36 % 3D直线L2拟合(纵向)
37 [direction_h, xyz_h] = Fit3DLine(points3d_h);
38 th=-1:0.1:1;
39 xh=xyz_h(1)+direction_h(1)*th;
40 yh=xyz_h(2)+direction_h(2)*th;
41 zh=xyz_h(3)+direction_h(3)*th;
42 plot3(xh,yh,zh, 'm'); hold on;
43 
44 %% <3> 求公垂线L 
45 % 数学问题描述：
46 % 空间中，找一根线L；直线满足以下两个条件   
47 % 1、垂直于直线L1、L2（已经L1与L2不平行、不相交、不重合且L1、L2方向向量中不能有0分量）
48 % 2、与向量L1 和 L2都有交点
49 
50 [direction_, p0, p1] = Intersection3DPoint(direction_v', xyz_v, direction_h', xyz_h)
51 t=-1:0.1:1;
52 x_=p0(1)+direction_(1)*t;
53 y_=p0(2)+direction_(2)*t;
54 z_=p0(3)+direction_(3)*t;
55 plot3(x_,y_,z_, 'c'); hold on; % 画线
56 
57 % 画点
58 plot3(p0(1,1), p0(1,2), p0(1,3), 'go'); hold on; 
59 plot3(p1(1,1), p1(1,2), p1(1,3), 'go'); hold on;
View Code

      mainV2.m(3D直线SVD拟合 + 直线求交点 测试离群点较多理想数据)

 1 %% note: 确保严谨，代码后续加入：1、判定是否同一根线； 2、是否平行; 3、是否真的相交 4、方向向量不能为0
 2 
 3 % 针对离群点：1、聚类(点云库有api，opencv也有)； 2、ransac(注意停止迭代的概率公式补上去)
 4 
 5 clc;
 6 clear all;
 7 close all;
 8 %% <1> 数据读取、处理
 9 % 竖直 
10 points3d_v = [];
11 % left
12 points3d_v = [points3d_v; importdata('TestData2/竖直方向/Point3D_V_Left20200311163646.csv')];
13 points3d_v = [points3d_v; importdata('TestData2/竖直方向/Point3D_V_Left20200311163759.csv')];
14 points3d_v = [points3d_v; importdata('TestData2/竖直方向/Point3D_V_Left20200311163840.csv')];
15 points3d_v = [points3d_v; importdata('TestData2/竖直方向/Point3D_V_Left20200311163927.csv')];
16 points3d_v = [points3d_v; importdata('TestData2/竖直方向/Point3D_V_Left20200311164009.csv')];
17 points3d_v = [points3d_v; importdata('TestData2/竖直方向/Point3D_V_Left20200311164045.csv')];
18 % right
19 % points3d_v = [points3d_v; importdata('TestData2/竖直方向/Point3D_V_Right20200311163646.csv')];
20 % points3d_v = [points3d_v; importdata('TestData2/竖直方向/Point3D_V_Right20200311163759.csv')];
21 % points3d_v = [points3d_v; importdata('TestData2/竖直方向/Point3D_V_Right20200311163840.csv')];
22 % points3d_v = [points3d_v; importdata('TestData2/竖直方向/Point3D_V_Right20200311163927.csv')];
23 % points3d_v = [points3d_v; importdata('TestData2/竖直方向/Point3D_V_Right20200311164009.csv')];
24 % points3d_v = [points3d_v; importdata('TestData2/竖直方向/Point3D_V_Right20200311164045.csv')];
25 
26 points3d_v = points3d_v(:,1:3); % 取出前三维
27 % 水平
28 points3d_h = [];
29 points3d_h = [points3d_h; importdata('TestData2/水平方向/Point3D_H20200311163646.csv')];
30 points3d_h = [points3d_h; importdata('TestData2/水平方向/Point3D_H20200311163759.csv')];
31 points3d_h = [points3d_h; importdata('TestData2/水平方向/Point3D_H20200311163840.csv')];
32 points3d_h = [points3d_h; importdata('TestData2/水平方向/Point3D_H20200311163927.csv')];
33 points3d_h = [points3d_h; importdata('TestData2/水平方向/Point3D_H20200311164009.csv')];
34 points3d_h = [points3d_h; importdata('TestData2/水平方向/Point3D_H20200311164045.csv')];
35 points3d_h = points3d_h(:,1:3); % 取出前三维
36 
37 % 显示
38 figure;
39 plot3(points3d_v(:,1), points3d_v(:,2), points3d_v(:,3), 'g.'); hold on;
40 plot3(points3d_h(:,1), points3d_h(:,2), points3d_h(:,3), 'b.'); hold on;
41 
42 %% <2> 拟合
43 
44 % direction: 方向向量 
45 % xyz：3D直线上一点
46 
47 % 3D直线L1拟合(横向)
48 [direction_v, xyz_v] = Fit3DLine(points3d_v); 
49 tv=-1:0.1:1;
50 xv=xyz_v(1)+direction_v(1)*tv;
51 yv=xyz_v(2)+direction_v(2)*tv;
52 zv=xyz_v(3)+direction_v(3)*tv;
53 plot3(xv,yv,zv, 'r'); hold on;
54 
55 % 3D直线L2拟合(纵向)
56 [direction_h, xyz_h] = Fit3DLine(points3d_h);
57 th=-1:0.1:1;
58 xh=xyz_h(1)+direction_h(1)*th;
59 yh=xyz_h(2)+direction_h(2)*th;
60 zh=xyz_h(3)+direction_h(3)*th;
61 plot3(xh,yh,zh, 'm'); hold on;
62 
63 %% <3> 求公垂线L 
64 % 数学问题描述：
65 % 空间中，找一根线L；直线满足以下两个条件   
66 % 1、垂直于直线L1、L2（已经L1与L2不平行、不相交、不重合且L1、L2方向向量中不能有0分量）
67 % 2、与向量L1 和 L2都有交点
68 
69 [direction_, p0, p1] = Intersection3DPoint(direction_v', xyz_v, direction_h', xyz_h)
70 t=-1:0.1:1;
71 x_=p0(1)+direction_(1)*t;
72 y_=p0(2)+direction_(2)*t;
73 z_=p0(3)+direction_(3)*t;
74 plot3(x_,y_,z_, 'c'); hold on; % 画线
75 
76 % 画点
77 plot3(p0(1,1), p0(1,2), p0(1,3), 'go'); hold on; 
78 plot3(p1(1,1), p1(1,2), p1(1,3), 'go'); hold on;
View Code

    整个Matlab代码工程放进百度云了。C++版本后续上传。

 链接: https://pan.baidu.com/s/1htim99dlfz56UNblh6e_Pw 提取码: bmfj 复制这段内容后打开百度网盘手机App，操作更方便哦

 

以下是3D直线的拟合算法（ransac选点 + SVD拟合内点集，注：依赖Eigen337）

       文件：FitSpaceLineRansacTest.cpp(拟合测试)

 1 // FitSpaceLineRansacTest.cpp : 此文件包含 "main" 函数。程序执行将在此处开始并结束。
 2 //
 3 
 4 #include"FitSpaceLineRansac.h"
 5 
 6 int main()
 7 {
 8     Eigen::MatrixXd pointsMatrix(12, 3); // 3D点数量最少四个点
 9     pointsMatrix << 
10                     // 内点
11                     0, 0, 0,
12                     1, 1, 1,
13                     2, 2, 2,
14                     3, 3, 3,
15                     4, 4, 4,
16                     5, 5, 5,
17                     6, 6, 6,
18                     7, 7, 7,
19                     //外点
20                     3, 4, 5, 
21                     4, 5, 6,
22                     5, 4, 3,
23                     4, 4, 7;
24 
25     FitSpaceLineRansac fitLine(pointsMatrix, 0.0000005, 50, 0.01);
26     fitLine.compute();
27     cout << fitLine.getModel() << endl;
28 
29     // TODO
30     return -1;
31 }

           这里是输出为：0.57735 0.57735 0.57735     3.5     3.5     3.5 ，前三个为方向向量，后三个为一个点(3.5     3.5     3.5)

 

            文件：FitSpaceLineRansac.h(拟合函数头文件)

 1 #pragma once
 2 #ifndef FITSPACELINERANSAC_H
 3 #define FITSPACELINERANSAC_H
 4 
 5 #include<iostream>
 6 #include<vector>
 7 #include<ctime>
 8 using namespace std;
 9 
10 #include"Eigen/Core"
11 #include"Eigen/Geometry"
12 
13 /*******************************************************************************************************
14 ** 功能描述：鲁棒性空间3D点直线拟合
15 ** 1、基于RANSAC算法的3D样本点集筛选
16 ** 2、基于SVD分解的直线拟合
17 ** 3、本模块依赖Eigen算法库，使用前需配置Eigen
18 ********************************************************************************************************
19 ** 模块名：基于RANSAC和SVD分解的3D点直线拟合
20 ** 输  入：_pointsMatrix               3D点集
21 **         _threshold                  若点到直线距离为d；如果d < sqrt(threshold), 则该点为内点，反之为外点
22 **           _maxIterNum                 最大迭代次数
23 **           _delta                      随机选取两点距离必须大于 delta
24 ** 输  出：_V                          直线方向向量
25 **        (_centerX _centerY _centerZ) 3D样本点集质心坐标           
26 ** 全局变量：
27 ** 调用模块： Eigen/Core, Eigen/Geometry.
28 ** 作  者：   shiruiyu
29 ** 日  期：   2020-03-24
30 ** 修  改：
31 ** 日  期：
32 ** 版  本：   1.0
33 ********************************************************************************************************
34 ** 备 注: 当输入3D点数量小于4时，成员函数getModel() 输出为0向量
35 *******************************************************************************************************/
36 class FitSpaceLineRansac
37 {
38 public:
39     FitSpaceLineRansac(const Eigen::MatrixXd& pointsMatrix, const double& threshold, const int& maxIterNum, const double&delta);
40     ~FitSpaceLineRansac();
41     // 执行拟合流程
42     void compute(); 
43     // 获取拟合结果
44     Eigen::Matrix<double, 1, 6>  getModel() const; 
45 
46 private:
47     // 获取随机数
48     int randIndex(const int& min, const int& max);
49     // 基于RANSAC算法的3D样本点集筛选
50     int selectInliersRansac(const Eigen::MatrixXd& pointsMatrix, const double& threshold, const int& maxIterNum, const double&delta, vector<int>& inLierVec);
51     // 基于SVD分解的直线拟合
52     void spacialLineFitting(const Eigen::MatrixXd& pointsMatrix, Eigen::Matrix3d& V, double& centerX, double& centerY, double& centerZ);
53 
54 private:
55     // 输入
56     Eigen::MatrixXd _pointsMatrix;       // 3D点样本集
57     double          _threshold;          // 若点到直线距离为d； d < sqrt(_threshold),则该点为内点
58     int             _maxIterNum;         // 最大迭代次数
59     double          _delta;              // 随机选取两点距离必须大于 _delta
60     
61     vector<int>     _inLierVec;          // 内点集合inlierPointsMatrix在_pointsMatrix中的索引集
62     Eigen::MatrixXd _inlierPointsMatrix; // 内点集合
63 
64     // 输出直线方程参数，直线方程:(x - x0)/a = (y - y0)/b = (z - z0)/c
65     Eigen::Matrix3d _V;                  // V.col(0) = [a b c]
66     double          _centerX;             // x0  
67     double          _centerY;            // y0
68     double          _centerZ;            // z0
69 };
70 
71 #endif // !FITSPACELINERANSAC_H

          文件：FitSpaceLineRansac.cpp(拟合函数具体实现)

  1 #include "FitSpaceLineRansac.h"
  2 
  3 FitSpaceLineRansac::FitSpaceLineRansac(const Eigen::MatrixXd& pointsMatrix, const double& threshold, const int& maxIterNum, const double&delta) :
  4                                        _pointsMatrix(pointsMatrix),
  5                                        _threshold(threshold),
  6                                        _maxIterNum(maxIterNum),
  7                                        _delta(delta)
  8 {}
  9 
 10 FitSpaceLineRansac::~FitSpaceLineRansac()
 11 {}
 12 
 13 // ********************************************************************* //
 14 // 功能：执行拟合流程
 15 // 输入：空          
 16 // 输出：空           
 17 // 备注：无            
 18 // ********************************************************************* //
 19 void FitSpaceLineRansac::compute()
 20 {
 21     // <1> 选内点集
 22     if (this->selectInliersRansac(this->_pointsMatrix, this->_threshold, this->_maxIterNum, this->_delta, this->_inLierVec) != 0)
 23         return;
 24     
 25     // <2> 深拷贝数据
 26     this->_inlierPointsMatrix.resize(this->_inLierVec.size(), 3);
 27     for (int i = 0; i < this->_inLierVec.size(); i++)
 28     {
 29         this->_inlierPointsMatrix.row(i) = this->_pointsMatrix.row(this->_inLierVec[i]);
 30     }
 31 
 32     // <3> 拟合
 33     this->_centerX = 0.0; 
 34     this->_centerY = 0.0; 
 35     this->_centerZ = 0.0;
 36     this->spacialLineFitting(this->_inlierPointsMatrix, this->_V, this->_centerX, this->_centerY, this->_centerZ); 
 37 }
 38 
 39 // ********************************************************************* //
 40 // 功能：获取拟合结果
 41 // 输入：空
 42 // 输出：直线方程参数 [a b c x0 y0 z0]          
 43 // 备注：直线方程 (x - x0)/a = (y - y0)/b = (z - z0)/c            
 44 // ********************************************************************* //
 45 Eigen::Matrix<double, 1, 6> FitSpaceLineRansac::getModel() const
 46 {
 47     Eigen::Matrix<double, 1, 6> model;
 48     model <<(this->_V)(0, 0), // a
 49             (this->_V)(1, 0), // b
 50             (this->_V)(2, 0), // c
 51              this->_centerX,  // x0
 52              this->_centerY,  // y0
 53              this->_centerZ;  // z0
 54     return model;
 55 }
 56 
 57 // ********************************************************************* //
 58 // 功能：获取随机值
 59 // 输入：min 区间下限、max 区间上限   
 60 // 输出：空
 61 // 返回：随机值，取值范围 [min, max)          
 62 // 备注：无            
 63 // ********************************************************************* //
 64 int FitSpaceLineRansac::randIndex(const int & min, const int & max)
 65 {
 66     return rand() % (max - min) + min;
 67 }
 68 
 69 // ********************************************************************* //
 70 // 功能：基于RANSAC算法的3D样本点集筛选
 71 // 输入：pointsMatrix 3D点集
 72 //       threshold    若点到直线距离为d；如果d < sqrt(threshold), 则该点为内点，反之为外点
 73 //         maxIterNum   最大迭代次数
 74 //         delta        随机选取两点距离必须大于 delta
 75 // 输出：inLierVec    内点集合在pointsMatrix中的索引集  
 76 // 返回：-1           异常
 77 //        0           正常
 78 // 备注：3D点集至少包含4个点            
 79 // ********************************************************************* //
 80 int FitSpaceLineRansac::selectInliersRansac(const Eigen::MatrixXd& pointsMatrix, const double& threshold, const int& maxIterNum, const double&delta, vector<int>& inLierVec)
 81 {
 82     if (pointsMatrix.data() == nullptr || pointsMatrix.rows() < 4 ||pointsMatrix.cols() != 3)
 83         return -1;
 84 
 85     srand((unsigned)time(nullptr)); // seed of time
 86     int preTotal = 0;               // 上一次迭代内点总数
 87     int total = 0;                  // 当前迭代内点总数
 88 
 89     for (int i = 0; i < maxIterNum; i++)
 90     {
 91         // <1> 随机取两点p0、p1；p0与p1连线记为L0
 92         Eigen::Matrix<double, 2, 3> sample;
 93         sample.row(0) = pointsMatrix.row(randIndex(0, pointsMatrix.rows())); // p0(x0 y0 z0)
 94         sample.row(1) = pointsMatrix.row(randIndex(0, pointsMatrix.rows())); // p1(x1 y1 z1)
 95 
 96         // <2> 确保p0 p1距离适当
 97         Eigen::Vector2d x, y, z;
 98         x = sample.col(0); 
 99         y = sample.col(1); 
100         z = sample.col(2);
101         Eigen::Vector3d v01;
102         v01 << x(0) - x(1), y(0) - y(1), z(0) - z(1);// 向量v01 = p0 - p1
103         if (v01.norm() < delta) // p0 p1太近？
104             continue;
105         v01.normalize(); // 归一化
106 
107         // <3> 筛选点
108         vector<int> idxVec;
109         for (int i = 0; i < pointsMatrix.rows(); i++)
110         {
111             // pi(xi yi zi)
112             double xi = pointsMatrix.row(i).x();
113             double yi = pointsMatrix.row(i).y();
114             double zi = pointsMatrix.row(i).z();
115 
116             // vi0 = pi - p0
117             Eigen::Vector3d vi0(xi - x(0), yi - y(0), zi - z(0));
118 
119             // vi0.cross(v01) : pi 到 直线L0 的距离
120             // norm : 平方根
121             if ((vi0.cross(v01)).norm() < sqrt(threshold))
122             {
123                 idxVec.push_back(i);
124             }
125         }
126         // <4> 更新
127         total = (int)idxVec.size();
128         if (total > preTotal)
129         {
130             preTotal = total;
131             inLierVec = idxVec; 
132         }
133 
134     }
135 
136     return 0;
137 }
138 
139 // ********************************************************************* //
140 // 功能：基于SVD分解的直线拟合
141 // 输入：pointsMatrix              3D点样本集
142 // 输出：V                         直线方向向量
143 //       (centerX centerY centerZ) 3D样本点集质心坐标           
144 // 备注：            
145 // ********************************************************************* //
146 void FitSpaceLineRansac::spacialLineFitting(const Eigen::MatrixXd& pointsMatrix, Eigen::Matrix3d& V, double& centerX, double& centerY, double& centerZ)
147 {
148     int r = (int)pointsMatrix.rows();
149     Eigen::MatrixXd sumMatrixXd(1, 3);
150     sumMatrixXd = pointsMatrix.colwise().sum();//求x，y，z和
151     // <1>、求质心
152     centerX = sumMatrixXd(0, 0) / r;
153     centerY = sumMatrixXd(0, 1) / r;
154     centerZ = sumMatrixXd(0, 2) / r;
155     Eigen::MatrixXd centerdMatrix(r, 3);
156     // <2>、去质心
157     for (int i = 0; i < r; i++)
158     {
159         centerdMatrix.row(i)[0] = pointsMatrix.row(i)[0] - centerX;
160         centerdMatrix.row(i)[1] = pointsMatrix.row(i)[1] - centerY;
161         centerdMatrix.row(i)[2] = pointsMatrix.row(i)[2] - centerZ;
162     }
163     // <3>、SVD分解
164     Eigen::JacobiSVD<Eigen::MatrixXd> svd(centerdMatrix, Eigen::ComputeThinU | Eigen::ComputeThinV);
165     V = svd.matrixV();
166 }

       以下是C++求交点代码

 CrossPointTest.cpp(主函数，用于测试)

 1 #include"CrossPoint.h"
 2 
 3 
 4 int test2( Eigen::MatrixXd& pointsMatrix_H, 
 5            Eigen::MatrixXd& pointsMatrix_V,
 6            vector<double>& p0, 
 7            vector<double>& p1)
 8 {
 9     // <1>
10     Eigen::Matrix3d V_H;
11     double centerX_H = 0.0, centerY_H = 0.0, centerZ_H = 0.0;
12     spacialLineFitting(pointsMatrix_H, V_H, centerX_H, centerY_H, centerZ_H); // 拟合
13     //cout << V_H(0, 0) << V_H(1, 0)  << V_H(2, 0) << endl; 方向
14     vector<double> center_H; // 质心
15     center_H.reserve(3);
16     center_H.push_back(centerX_H);
17     center_H.push_back(centerY_H);
18     center_H.push_back(centerZ_H);
19     // <2>
20     Eigen::Matrix3d V_V;
21     double centerX_V = 0.0, centerY_V = 0.0, centerZ_V = 0.0;
22     spacialLineFitting(pointsMatrix_V, V_V, centerX_V, centerY_V, centerZ_V);
23     //cout << V_V(0, 0) << V_V(1, 0) << V_V(2, 0) << endl;
24     vector<double> center_V;
25     center_V.reserve(3);
26     center_V.push_back(centerX_V);
27     center_V.push_back(centerY_V);
28     center_V.push_back(centerZ_V);
29     // <3>
30     return Intersection3DPoint(V_V, center_V, V_H, center_H, p0, p1);
31 }
32 
33 int main()
34 {
35     cout << "p0、p1分别是公垂线L与直线L0、L1的交点" << endl;
36 
37     vector<double> p0, p1;//与两直线交点
38     p0.resize(3);
39     p1.resize(3);
40     
41     // 情况1:实测数据
42     /*if (test1(p0, p1) == 0)
43     {
44         cout << "p0(" << p0[0] << "," << p0[1] << "," << p0[2] << ")" << endl;
45         cout << "p1(" << p1[0] << "," << p1[1] << "," << p1[2] << ")" << endl;
46         cout << "************************" << endl;
47     }*/
48     
49 
50     // 情况2:方向向量有零分量
51     Eigen::MatrixXd pointsMatrix_H(3, 3); // 3个3D点，水平
52     pointsMatrix_H << 0, 0, 1, 2, 0, 1, 3, 0, 1; // (0, 0, 1), (2, 0, 1), (3, 0, 1)
53     Eigen::MatrixXd pointsMatrix_V(3, 3);
54     pointsMatrix_V << 0, 0, 4, 0, 2, 4, 0, 9, 4; // (0, 0, 4), (0, 2, 4), (0, 9, 4)
55 
56     if (test2(pointsMatrix_H, pointsMatrix_V, p0, p1) == 0)
57     {
58         cout << "p0(" << p0[0] << "," << p0[1] << "," << p0[2] << ")" << endl; //  理论：(0 0 4)
59         cout << "p1(" << p1[0] << "," << p1[1] << "," << p1[2] << ")" << endl; //  理论：(0 0 1)
60         cout << "************************" << endl;
61     }
62     
63 
64     // 情况3:两直线有交点
65     pointsMatrix_H << 0, 0, 0, 0, 0, 2, 0, 0, 7; // (0, 0, 0), (0, 0, 2), (0, 0, 7)
66     pointsMatrix_V << 0, 0, 3, 3, 0, 3, 4, 0, 3; // (0, 0, 3), (3, 0, 3), (4, 0, 3)
67 
68     if (test2(pointsMatrix_H, pointsMatrix_V, p0, p1) == 0)
69     {
70         cout << "p0(" << p0[0] << "," << p0[1] << "," << p0[2] << ")" << endl; // 理论：(0 0 3)
71         cout << "p1(" << p1[0] << "," << p1[1] << "," << p1[2] << ")" << endl; // 理论：(0 0 3)
72         cout << "************************" << endl;
73     }
74 
75     // 情况4:平行(重合)(既平行，又是零向量，所以要先判断是否平行)
76     pointsMatrix_H << 1.33, 0, 3.21, 1.33, 0, 3.31, 1.33, 0, 4.81; // (1.33, 0, 3.21), (1.33, 0, 3.31), (1.33, 0, 4.81)
77     pointsMatrix_V << 0, 2.3, 3.1, 0, 2.3, 3.2, 0, 2.3, 1.222; // (0, 2.3, 3.1), (0, 2.3, 3.2), (0, 2.3, 1.222)
78 
79     if (test2(pointsMatrix_H, pointsMatrix_V, p0, p1) == 0)
80     {
81         cout << "p0(" << p0[0] << "," << p0[1] << "," << p0[2] << ")" << endl; // 理论：由于平行，直接return
82         cout << "p1(" << p1[0] << "," << p1[1] << "," << p1[2] << ")" << endl; // 
83         cout << "************************" << endl;
84     }
85     
86 
87 
88 
89     system("pause");
90     return 0;
91 }

CrossPoint.h(空间直线拟合、直线求交点)

 1 #pragma once
 2 
 3 #ifndef CROSSPOINT_H
 4 #define CROSSPOINT_H
 5 
 6 #include<iostream>
 7 #include<fstream>
 8 #include<sstream>
 9 #include<vector>
10 #include"Eigen/Dense"
11 using namespace std;
12 
13 void spacialLineFitting(Eigen::MatrixXd& pointsMatrix, Eigen::Matrix3d& V, double& centerX, double& centerY, double& centerZ);
14 
15 void readPoint(string strPath, vector<vector<double>>& pointsVec);
16 
17 int Intersection3DPoint(Eigen::Matrix3d dir_V, vector<double> center_V, Eigen::Matrix3d dir_H, vector<double> center_H, vector<double>& p0, vector<double>& p1);
18 
19 #endif // !CROSSPOINT_H

CrossPoint.cpp

  1 #include"CrossPoint.h"
  2 
  3 // **************************************************************************************************************
  4 
  5 // 功能：空间直线拟合
  6 
  7 // 输入：pointsMatrix：3D点样本集
  8 
  9 // 输出：V：直线方向向量；(centerX centerY centerZ):3D样本点集质心
 10 
 11 // return: void
 12 
 13 // **************************************************************************************************************
 14 void spacialLineFitting(Eigen::MatrixXd& pointsMatrix, Eigen::Matrix3d& V, double& centerX, double& centerY, double& centerZ)
 15 {
 16     int r = pointsMatrix.rows();
 17     Eigen::MatrixXd sumMatrixXd(1, 3);
 18     sumMatrixXd = pointsMatrix.colwise().sum();//求x，y，z和
 19     // <1>、求质心
 20     centerX = sumMatrixXd(0, 0) / r;
 21     centerY = sumMatrixXd(0, 1) / r;
 22     centerZ = sumMatrixXd(0, 2) / r;
 23     Eigen::MatrixXd centerdMatrix(r, 3);
 24     // <2>、去质心
 25     for (int i = 0; i < r; i++)
 26     {
 27         centerdMatrix.row(i)[0] = pointsMatrix.row(i)[0] - centerX;
 28         centerdMatrix.row(i)[1] = pointsMatrix.row(i)[1] - centerY;
 29         centerdMatrix.row(i)[2] = pointsMatrix.row(i)[2] - centerZ;
 30     }
 31     // <3>、SVD分解
 32     Eigen::JacobiSVD<Eigen::MatrixXd> svd(centerdMatrix, Eigen::ComputeThinU | Eigen::ComputeThinV);
 33     V = svd.matrixV();
 34 }
 35 //字符转double
 36 template <class Type>
 37 Type stringToNum(string& str)
 38 {
 39     istringstream iss(str);
 40     Type num;
 41     iss >> num;
 42     return num;
 43 }
 44 //读取csv文件点
 45 void readPoint(string strPath, vector<vector<double>>& pointsVec)
 46 {
 47     std::ifstream _csvInput;
 48     _csvInput.open(strPath, std::ios::in);
 49     std::string _Oneline;
 50     std::vector<double> _lineOfstr;
 51     vector<vector<double>> verticalLinePoints;
 52     while (std::getline(_csvInput, _Oneline))  // 整行读取，换行符“\n”区分，遇到文件尾标志eof终止读取  
 53     {
 54         std::istringstream _Readstr(_Oneline); // 将整行字符串_Oneline读入到字符串流istringstream中
 55         std::string _partOfstr;
 56         double pointCorr;
 57         int i = 0;
 58         while (std::getline(_Readstr, _partOfstr, ',')) // 将字符串流_Readstr中的字符读入到_partOfstr字符串中，以逗号为分隔符
 59         {
 60             if (i < 3)
 61             {
 62                 pointCorr = stringToNum<double>(_partOfstr);
 63                 _lineOfstr.push_back(pointCorr); // 将刚刚读取的字符串添加到向量_lineOfstr中,
 64             }
 65             else if (i == 3)
 66             {
 67                 verticalLinePoints.push_back(_lineOfstr);
 68                 _lineOfstr.clear();
 69                 _Readstr.clear();
 70                 break;
 71             }
 72             ++i;
 73         }
 74     }
 75     pointsVec = verticalLinePoints;
 76 }
 77 
 78 
 79 // **************************************************************************************************************
 80 
 81 // 功能：计算公垂线L与两直线（L0、L1）交点
 82 
 83 // 输入：dir_V:竖向直线（L0）方向向量； center_V：竖向3D点质心
 84 //       dir_H:横向直线（L1）方向向量； center_H：横向3D点质心
 85 
 86 // 输出：p0: L0与公垂线（L）的交点； p1: L1与公垂线（L）的交点；
 87 
 88 // return: -1: L0、L1平行（重合）
 89 //          0: L0、L1不平行（不重合）
 90 
 91 // **************************************************************************************************************
 92 int Intersection3DPoint(Eigen::Matrix3d dir_V, vector<double> center_V, Eigen::Matrix3d dir_H, vector<double> center_H, vector<double>& p0, vector<double>& p1)
 93 {
 94     //直线1： (x - x0)/a0 = (y - y0)/b0 = (z - z0)/c0
 95     double a0 = dir_V(0, 0), b0 = dir_V(1, 0), c0 = dir_V(2, 0); // (a0 b0 c0) 是竖向直线L0的方向向量
 96     double x0 = center_V[0], y0 = center_V[1], z0 = center_V[2]; // (x0 y0 z0) 是竖向3D样本点的质心
 97     //直线2： (x - x1)/a1 = (y - y1)/b1 = (z - z1)/c1
 98     double a1 = dir_H(0, 0), b1 = dir_H(1, 0), c1 = dir_H(2, 0);
 99     double x1 = center_H[0], y1 = center_H[1], z1 = center_H[2];
100     Eigen::Matrix<double, 1, 3> dir_V_, dir_H_;
101 
102     // 方向向量零分量补偿
103     if (a0 == 0) // abs(a0 - 1e-5)
104         a0 += 1e-5;
105     if (b0 == 0) 
106         b0 += 1e-5;
107     if (c0 == 0)
108         c0 += 1e-5;
109 
110     if (a1 == 0)
111         a1 += 1e-5;
112     if (b1 == 0)
113         b1 += 1e-5;
114     if (c1 == 0)
115         c1 += 1e-5;
116 
117     //cout << a0 / a1 << endl;
118     //cout << b0 / b1 << endl;
119     //cout << c0 / c1 << endl;
120     // 判断平行(重合)
121     if (a0 / a1 == b0 / b1 == c0 / c1)
122         return -1;
123 
124     dir_V_(0, 0) = a0;
125     dir_V_(0, 1) = b0;
126     dir_V_(0, 2) = c0;
127     dir_H_(0, 0) = a1;
128     dir_H_(0, 1) = b1;
129     dir_H_(0, 2) = c1;
130     // <1>、求公垂线方向
131     Eigen::MatrixXd crossM = dir_H_.cross(dir_V_); // (a0 b0 c0) 与 (a1 b1 c1) 作叉积 -> 公垂向量(a2 b2 c2)
132     double a2 = crossM(0, 0);
133     double b2 = crossM(0, 1);
134     double c2 = crossM(0, 2);
135     // <2>、求交点p0、p1(推导见 readme.txt)
136     Eigen::Matrix<double, 2, 2> A;
137     A(0, 0) = a1 * b2 - a2 * b1;
138     A(0, 1) = a2 * b0 - a0 * b2;
139     A(1, 0) = b1 * c2 - b2 * c1;
140     A(1, 1) = b2 * c0 - b0 * c2;
141     Eigen::Matrix<double, 2, 1> B;
142     B(0, 0) = a2 * (y1 - y0) + b2 * (x0 - x1);
143     B(1, 0) = b2 * (z1 - z0) + c2 * (y0 - y1);
144     Eigen::Matrix<double, 2, 1> X;
145     X = A.inverse()*B; // 解方程 A*X = B
146     double alpha = X(1, 0);
147     double beta = X(0, 0);
148     /*p0.resize(3);
149     p1.resize(3);*/
150     p0[0] = alpha * a0 + x0;
151     p0[1] = alpha * b0 + y0;
152     p0[2] = alpha * c0 + z0;
153     p1[0] = beta * a1 + x1;
154     p1[1] = beta * b1 + y1;
155     p1[2] = beta * c1 + z1;
156     return 0;
157 }

 

posted @ 2020-04-23 17:40  佚名12  阅读(490)  评论(0)  编辑  收藏  举报