Objective: Take a point (or set of points) from a camera perspective view and translate it/them to the respective ground plane points.
Approach: Used findHomography to obtain a homography Mat. Plan to use perspectiveTransform().
Issue: Cannot interpret homography Mat, nor understand how to continue. It appears 5 of the 9 elements are null, which I take to mean that the homography is faulty, though have no way of interpreting the results, so a made a number of print statements.
import org.opencv.calib3d.*;
import org.opencv.core.Core;
import org.opencv.core.Mat;
import org.opencv.core.MatOfPoint2f;
import org.opencv.core.Point;
import org.opencv.highgui.Highgui;
import org.opencv.imgproc.*;
import java.util.List;
import java.util.LinkedList;
import java.util.ArrayList;
final public class HomographyTest {
public static void main(String[] args) {
System.loadLibrary(Core.NATIVE_LIBRARY_NAME); // needed by OpenCV
String rfileoutput = "/home/will/dev/Homog.jpg";
String ofileoutput = "/home/will/dev/HomogOutput.jpg";
Point SEShedCornerDst = new Point(49, 74);
Point CloseForsythiaDst = new Point(41, 41);
Point CornerHazelDst = new Point(111, 157);
Point FarForsythiaDst = new Point(175, 21);
Point FirstLiquidAmberDst = new Point(235, 164);
Point SecondLiquidAmberDst = new Point(282, 721);
Point ThirdLiquidAmberDst = new Point(317, 544);
Point SEShedCornerSrc = new Point(30, 231);
Point CloseForsythiaSrc = new Point(160, 290);
Point CornerHazelSrc = new Point(50, 125);
Point FarForsythiaSrc = new Point(628, 146);
Point FirstLiquidAmberSrc = new Point(299, 64);
Point SecondLiquidAmberSrc = new Point(146, 37);
Point ThirdLiquidAmberSrc = new Point(48,34);
Point [] srcArray = new Point[7];
srcArray[0] = SEShedCornerSrc;
srcArray[1] = CloseForsythiaSrc;
srcArray[2] = CornerHazelSrc;
srcArray[3] = FarForsythiaSrc;
srcArray[4] = FirstLiquidAmberSrc;
srcArray[5] = SecondLiquidAmberSrc;
srcArray[6] = ThirdLiquidAmberSrc;
Mat OutputMat = new Mat();
LinkedList<Point> dstArray = new LinkedList<Point>();
dstArray.add(SEShedCornerDst);
dstArray.add(CloseForsythiaDst);
dstArray.add(CornerHazelDst);
dstArray.add(FarForsythiaDst);
dstArray.add(FirstLiquidAmberDst);
dstArray.add(SecondLiquidAmberDst);
dstArray.add(ThirdLiquidAmberDst);
MatOfPoint2f dst = new MatOfPoint2f();
dst.fromList(dstArray);
MatOfPoint2f src = new MatOfPoint2f();
src.fromArray(srcArray);
Mat Homog;
Homog = Calib3d.findHomography(src, dst, Calib3d.RANSAC, 10, OutputMat);
System.out.println("Columns = " + Homog.cols());
System.out.println("Rows = " + Homog.rows());
System.out.println("Width = " + Homog.width());
System.out.println("Dims = " + Homog.dims());
for (int i=1; i<= Homog.cols();i++){
for (int j=1; j<=Homog.rows();j++){
System.out.println("Row, column " + i + "," + j + " = " + Homog.get(j, i));
}
System.out.println();
}
System.out.println(Homog.toString());
System.out.println(OutputMat.toString());
Highgui.imwrite(rfileoutput, Homog);
Highgui.imwrite(ofileoutput, OutputMat);
}
}
The output;
Columns = 3
Rows = 3
Width = 3
Dims = 2
Row, column 1,1 = [D@674f1c67
Row, column 1,2 = [D@7ad1e32d
Row, column 1,3 = null
Row, column 2,1 = [D@6999de59
Row, column 2,2 = [D@74d4db38
Row, column 2,3 = null
Row, column 3,1 = null
Row, column 3,2 = null
Row, column 3,3 = null
Mat [ 3*3*CV_64FC1, isCont=true, isSubmat=false, nativeObj=0x7f744016bb50, dataAddr=0x7f744016b940 ]
Mat [ 7*1*CV_8UC1, isCont=true, isSubmat=false, nativeObj=0x7f7440166fe0, dataAddr=0x7f744016b9b0 ]
Am I on the right track? If not, what should I do?
If so, what type is entered into the src Mat for perspectiveTransform, and how do I do that in the two channels it says is required?
I can't really interpret your code, especially your points but I don't think it matters because as you said your objective is to create a homography matrix from 2 sets of points and perspectively transform an image with the resulting homography matrix.
Iterating through the homography matrix
First of all the reason why you get null from your iteration through the homography matrix (which is a 3x3 matrix) is because Mat indices start just like arrays at 0 and end at "length - 1".
It should look something like this:
for (int i = 0; i < homography.cols(); i++){
for (int j = 0; j < homography.rows(); j++){
System.out.println("[" + i + "," + j + "] = " + Arrays.toString(homography.get(j, i)));
}
}
Find homography and apply (to image)
And to find the homography matrix between your src and dst planes you can use findHomography(). For more information on this function, it's parameters look in the API
// obtain your homography mat (picked your parameters.. you have to play to get the right results)
Mat homography = Calib3d.findHomography(src, dst, Calib3d.RANSAC, 10);
// image you want to transform
Mat image = Highgui.imread("image/img.jpg");
// outputMat will contain the perspectively changed image
Imgproc.warpPerspective(image, outputMat, homography, new Size(image.cols(), image.rows()));
Applying homography to single points (not an image)
Mat src = new Mat();
// push a mat object with your points to the src
src.push_back(new MatOfPoint2f(new Point(x,y)));
// dst will contain your transformed points
Core.perspectiveTransform(src, dst, homography)
Btw..
In Java it is common practice to start variable names with lowercase letters. I saw Homog and OutputMat. When I read something like that I expect them to be classes.