Cannot add 4 numbers to mat opencv - opencv

Question

I created a mat object in opencv , dimension Nx4 , in which I want to put N coordinates.
[Px Py 1 0]
[Py Px 0 1]
For this I wrote the following code :
vector<Point2f> features1 , features2;
Mat features_1;
for(int i = 0 , j = 0; i < feature1.size() ; ++i , j+=2)
{
features_1.at<Vec3d>(j) = {feature1[i].x , feature1[i].y , 1 , 0};
features_1.at<Vec3d>(j) = {feature1[i].y , -feature1[i].x , 0 , 1};
}
But at the first line of the loop I get the following error :
cv::Matx<_Tp, m, n>::Matx(_Tp, _Tp, _Tp, _Tp) [with _Tp = double; int m = 3; int n = 1]: Assertion channels >= 4' failed.
How can I solve this?

Answer 1

it's probably easier to push_back() single elements to a Mat, and later do a reshape().
this will make a 2*N x 4 Mat:
vector<Point2f> features1 , features2;
Mat features_1;
for(size_t i=0; i<feature1.size() ; ++i)
{
features_1.push_pack(feature1[i].x);
features_1.push_pack(feature1[i].y);
features_1.push_pack(1);
features_1.push_pack(0);
features_1.push_pack(feature2[i].y);
features_1.push_pack(feature2[i].x);
features_1.push_pack(0);
features_1.push_pack(1);
}
features_1 = features_1.reshape(1,4); // or reshape(1,features1.size()*2); //for 4xN*2

Answer 2

Where you wrote <Vec3d>, did you mean <Vec4d>?
Your code is not close to correct, but here is a minimal example of how your desired result can be achieved (if you don't have C++11, just use an old style iterator in the loop)
#include <iostream>
#include <vector>
using namespace std;
#include <opencv2/core/core.hpp>
using namespace cv;
int main()
{
vector<Point2f> feature1;
feature1.push_back(Point2f(1.0f, 2.0f));
feature1.push_back(Point2f(3.0f, 4.0f));
Mat features_1;
for(auto p : feature1)
{
features_1.push_back(Vec4d(p.x, p.y , 1 , 0));
features_1.push_back(Vec4d(p.y, -p.x , 0, 1));
}
// At this point the mat is (N, 1, CV_64FC4)
// Reshape if you want (N, 4, CV_64FC1);
features_1 = features_1.reshape(1,4);
cout << features_1;
}
This output of this is:
[1, 2, 1, 0;
2, -1, 0, 1;
3, 4, 1, 0;
4, -3, 0, 1]
However, if the features_1 is already allocated and you can't just use push_back, apart from releasing the mat, you could do this (assuming features_1.isContinuous() is true):
#include <iostream>
#include <vector>
using namespace std;
#include <opencv2/core/core.hpp>
using namespace cv;
int main()
{
vector<Point2f> feature1;
feature1.push_back(Point2f(1.0f, 2.0f));
feature1.push_back(Point2f(3.0f, 4.0f));
Mat features_1(2 * feature1.size(), 4, CV_64FC1);
Vec4d* ptr = features_1.ptr<Vec4d>(0);
for(auto p : feature1)
{
*ptr++ = Vec4d(p.x, p.y , 1 , 0);
*ptr++ = Vec4d(p.y, -p.x , 0, 1);
}
cout << features_1 << endl;
}
The above produces the same result as the previous version.
If there is any possibility in the above example that features_1.isContinuous() is false, you can use an iterator to scan features_1:
auto ptr = features_1.begin<double>();
for(auto p : feature1)
{
*ptr++ = p.x;
*ptr++ = p.y;
*ptr++ = 1;
*ptr++ = 0;
*ptr++ = p.y;
*ptr++ = -p.x;
*ptr++ = 0;
*ptr++ = 1;
}