public static void ReprojectionError(
Image <Arthmetic, double> estReal, List <PointF> img,
Image <Arthmetic, double> K, Image <Arthmetic, double> R, Image <Arthmetic, double> t,
out double mean, out double median, out List <double> errors)
{
var P = ComputeMatrix.Camera(K, R, t);
var estImg = P.Multiply(estReal);
errors = new List <double>();
for (int i = 0; i < img.Count; ++i)
{
var estPoint = new Image <Arthmetic, double>(new double[, , ]
{
{ { estImg[0, i] / estImg[2, i] } }, { { estImg[1, i] / estImg[2, i] } },
});
var realPoint = new Image <Arthmetic, double>(new double[, , ]
{
{ { img[i].X } }, { { img[i].Y } },
});
errors.Add(estPoint.Sub(realPoint).Norm);
}
mean = errors.Sum() / errors.Count;
median = errors[errors.Count / 2];
}
private Image <Arthmetic, double> ComputeF()
{
var F = ComputeMatrix.F(new VectorOfPointF(pts1_n.ToArray()), new VectorOfPointF(pts2_n.ToArray()));
// F is normalized - lets denormalize it
F = N2.T().Multiply(F).Multiply(N1);
return(F);
}
private void ErrorOfRComputation_Click(object sender, RoutedEventArgs e)
{
Func <double> testFunc = () =>
{
var F = ComputeF();
var E = ComputeMatrix.E(F, K);
FindTransformation.DecomposeToRTAndTriangulate(pts1, pts2, K, E, out var R, out var t, out var pts3d);
return(FindRError(R));
};
public double Cost(Image <Arthmetic, double> F, double fx, double fy, double px, double py, double w, double h)
{
Image <Arthmetic, double> E = ComputeMatrix.E(F, ComputeMatrix.K(fx, fy, px, py));
Svd svd = new Svd(E);
double s1 = svd.S[0, 0];
double s2 = svd.S[1, 0];
double errS = s2 == 0 ? 1.0 : (s1 - s2) / s2;
return(errS);
}
private void EigenRatioForKnownK_Click(object sender, RoutedEventArgs e)
{
PlotFunctionForErrors(() =>
{
var F = ComputeF();
var E = ComputeMatrix.E(F, K);
var svd = new Svd(E);
double error = svd.S[1, 0] / svd.S[0, 0];
return(error);
}, "s2 / s1");
}
public static Matrix TransfromBack3dPoints(Matrix R, Matrix t, Matrix pts23, double scale)
{
var backprojected = Utils.PutRTo4x4(R.T()).Multiply(pts23);
var C = ComputeMatrix.Center(t, R);
for (int i = 0; i < backprojected.Cols; ++i)
{
backprojected[0, i] = backprojected[0, i] * scale + C[0, 0];
backprojected[1, i] = backprojected[1, i] * scale + C[1, 0];
backprojected[2, i] = backprojected[2, i] * scale + C[2, 0];
}
return(backprojected);
}
public static int TriangulateChieral(
List <PointF> left, List <PointF> right, Image <Arthmetic, double> K,
Image <Arthmetic, double> R, Image <Arthmetic, double> t,
out Image <Arthmetic, double> pts3d)
{
// init 3d point matrix
pts3d = new Image <Arthmetic, double>(left.Count, 4);
// init projection matrices
var P1 = ComputeMatrix.Camera(K);
var P2 = ComputeMatrix.Camera(K, R, t);
// Transform points lists into matrices
var img1 = new Image <Arthmetic, double>(left.Count, 2);
var img2 = new Image <Arthmetic, double>(left.Count, 2);
for (int i = 0; i < left.Count; ++i)
{
img1[0, i] = left[i].X;
img1[1, i] = left[i].Y;
img2[0, i] = right[i].X;
img2[1, i] = right[i].Y;
}
CvInvoke.TriangulatePoints(P1, P2, img1, img2, pts3d);
// Scale points, so that W = 1
for (int i = 0; i < left.Count; ++i)
{
pts3d[0, i] = pts3d[0, i] / pts3d[3, i];
pts3d[1, i] = pts3d[1, i] / pts3d[3, i];
pts3d[2, i] = pts3d[2, i] / pts3d[3, i];
pts3d[3, i] = pts3d[3, i] / pts3d[3, i];
}
// compute points in front of camera (TODO: how does it work?)
var AX1 = P1.Multiply(pts3d);
var BX1 = P2.Multiply(pts3d);
int num = 0;
for (int i = 0; i < left.Count; i++)
{
if (AX1[2, i] * pts3d[3, i] > 0 && BX1[2, i] * pts3d[3, i] > 0)
{
num++;
}
}
return(num);
}
public static Image <Arthmetic, double> K(List <Image <Arthmetic, double> > Fs, double width, double height)
{
double fi = (width + height) / 2;
var minimizer = new BfgsMinimizer(1e-6, 1e-6, 1e-6);
var result = minimizer.FindMinimum(
new ObjFunc()
{
Fs = Fs, Width = width, Height = height
},
new DenseVector(new double[] { fi, fi, width / 2, height / 2 })
);
var p = result.MinimizingPoint;
return(ComputeMatrix.K(p[0], p[1], p[2], p[3]));
}
public static double Cost(List <Image <Arthmetic, double> > Fs, double fx, double fy, double px, double py, double w, double h)
{
double errS = 0;
foreach (var F in Fs)
{
Image <Arthmetic, double> E = ComputeMatrix.E(F, ComputeMatrix.K(fx, fy, px, py));
Svd svd = new Svd(E);
double s1 = svd.S[0, 0];
double s2 = svd.S[1, 0];
errS += s2 == 0 ? 1 : (s1 - s2) / s2;
}
return(errS);
}
public static Image <Arthmetic, double> K(List <Mat> mats, Feature2D detector, Feature2D descriptor, DistanceType distanceType, double maxDistance)
{
List <Image <Arthmetic, double> > Fs = new List <Image <Arthmetic, double> >();
for (int i = 0; i < mats.Count - 1; i += 2)
{
var match = MatchImagePair.Match(mats[i], mats[i + 1], detector, descriptor, distanceType, maxDistance);
var F = ComputeMatrix.F(match.LeftPoints, match.RightPoints);
if (F == null)
{
continue;
}
Fs.Add(F);
}
return(K(Fs, mats[0].Width, mats[0].Height));
}
public void ProcessImages(Mat left, Mat right, Feature2D detector, Feature2D descriptor, DistanceType distanceType, double takeBest)
{
var match = MatchImagePair.Match(left, right, detector, descriptor, distanceType, 20.0);
DrawFeatures(left, right, match, takeBest);
var lps = match.LeftPointsList.Take((int)(match.LeftPoints.Size * takeBest));
var rps = match.RightPointsList.Take((int)(match.RightPoints.Size * takeBest));
var F = ComputeMatrix.F(new VectorOfPointF(lps.ToArray()), new VectorOfPointF(rps.ToArray()));
var K = EstimateCameraFromImagePair.K(F, left.Width, right.Height);
var E = ComputeMatrix.E(F, K);
FindTransformation.DecomposeToRTAndTriangulate(lps.ToList(), rps.ToList(), K, E,
out Image <Arthmetic, double> R, out Image <Arthmetic, double> t, out Image <Arthmetic, double> X);
PrintMatricesInfo(E, K, R, t);
}
public static void DecomposeToRT(Image <Arthmetic, double> E,
out Image <Arthmetic, double>[] Rs,
out Image <Arthmetic, double>[] ts)
{
var svd = new Svd(E);
Image <Arthmetic, double> W = new Image <Arthmetic, double>(new double[, , ] {
{ { 0 }, { -1 }, { 0 } },
{ { 1 }, { 0 }, { 0 } },
{ { 0 }, { 0 }, { 1 } },
});
var R1 = svd.U.Multiply(W.T()).Multiply(svd.VT);
double det1 = CvInvoke.Determinant(R1);
if (det1 < 0)
{
R1 = R1.Mul(-1);
}
var R2 = svd.U.Multiply(W).Multiply(svd.VT);
double det2 = CvInvoke.Determinant(R2);
if (det2 < 0)
{
R2 = R2.Mul(-1);
}
double ss = (svd.S[0, 0] + svd.S[1, 0]) / 2;
Image <Arthmetic, double> Z = new Image <Arthmetic, double>(new double[, , ] {
{ { 0 }, { -ss }, { 0 } },
{ { ss }, { 0 }, { 0 } },
{ { 0 }, { 0 }, { 0 } },
});
var t1 = svd.U.Multiply(Z).Multiply(svd.U.T());
t1 = ComputeMatrix.CrossProductToVector(t1);
var t2 = t1.Mul(-1);
Rs = new Image <Arthmetic, double>[] { R1, R2 };
ts = new Image <Arthmetic, double>[] { t1, t2 };
}
public static bool FindTwoViewsMatrices(List <PointF> left, List <PointF> right, Matrix K, out Matrix F, out Matrix E, out Matrix R, out Matrix t, out Matrix X)
{
var lps_n = new List <PointF>(left);
var rps_n = new List <PointF>(right);
NormalizePoints2d(lps_n, out var NL);
NormalizePoints2d(rps_n, out var NR);
F = ComputeMatrix.F(new VectorOfPointF(lps_n.ToArray()), new VectorOfPointF(rps_n.ToArray()));
if (F == null)
{
E = R = t = X = null;
return(false);
}
F = NR.T().Multiply(F).Multiply(NL);
E = ComputeMatrix.E(F, K);
DecomposeToRTAndTriangulate(left, right, K, E, out R, out t, out X);
return(true);
}
private void ResetPoints()
{
P1 = ComputeMatrix.Camera(K);
P2 = ComputeMatrix.Camera(K, R12, C2);
P3 = ComputeMatrix.Camera(K, R13, C3);
ptsReal = new List <Image <Arthmetic, double> >();
pts1Ref = new List <PointF>();
pts2Ref = new List <PointF>();
pts3Ref = new List <PointF>();
Random rand = new Random(564073);
for (int i = 0; i < pointsCount; ++i)
{
var real = new Image <Arthmetic, double>(new double[, , ] {
{ { rand.Next(100, 200) } },
{ { rand.Next(100, 200) } },
{ { rand.Next(50, 100) } },
{ { 1 } },
});
ptsReal.Add(real);
var i1 = P1.Multiply(real).ToPointF();
pts1Ref.Add(i1);
var i2 = P2.Multiply(real).ToPointF();
pts2Ref.Add(i2);
var i3 = P3.Multiply(real).ToPointF();
pts3Ref.Add(i3);
}
double rangeLx = pts1Ref.Max((x) => x.X) - pts1Ref.Min((x) => x.X);
double rangeLy = pts1Ref.Max((x) => x.Y) - pts1Ref.Min((x) => x.Y);
double rangeRx = pts2Ref.Max((x) => x.X) - pts2Ref.Min((x) => x.X);
double rangeRy = pts2Ref.Max((x) => x.Y) - pts2Ref.Min((x) => x.Y);
pixelRange = 0.25 * (rangeLx + rangeLy + rangeRx + rangeRy);
}
public void ProcessImages(Mat left, Mat middle, Mat right, Feature2D detector, Feature2D descriptor, DistanceType distance)
{
double maxDistance = 20.0;
var match12 = MatchImagePair.Match(left, middle, detector, descriptor, distance, maxDistance);
var match23 = MatchImagePair.Match(middle, right, detector, descriptor, distance, maxDistance);
var match13 = MatchImagePair.Match(left, right, detector, descriptor, distance, maxDistance);
TripletMatch tmatch = new TripletMatch();
List <MDMatch> m12 = new List <MDMatch>();
List <MDMatch> m23 = new List <MDMatch>();
var left1 = match12.LeftPoints;
var right1 = match12.RightPoints;
var left2 = match23.LeftPoints;
var left2_X = MatchClosePoints.SortByX(match23.LeftPoints);
var right2 = match23.RightPoints;
var left3 = match13.LeftPoints;
var right3 = match13.RightPoints;
var right3_X = MatchClosePoints.SortByX(match13.LeftPoints);
for (int idx12 = 0; idx12 < left1.Size; ++idx12)
{
var p1 = left1[idx12];
var p2 = right1[idx12];
int idx23 = IndexOf_X(left2_X, p2);
if (idx23 != -1)
{
var p3 = right2[idx23];
int idx13 = IndexOf_X(right3_X, p1);
if (idx13 != -1)
{
if (AreEqual(left1[idx12], left3[idx13]))
{
tmatch.Left.Add(p1);
tmatch.Middle.Add(p2);
tmatch.Right.Add(p3);
m12.Add(match12.Matches[idx12]);
m23.Add(match23.Matches[idx23]);
}
}
}
}
match12.Matches = new VectorOfDMatch(m12.ToArray());
match23.Matches = new VectorOfDMatch(m23.ToArray());
MatchDrawer.DrawFeatures(left, right, match12, 1.0, bottomView);
MatchDrawer.DrawFeatures(left, right, match23, 1.0, upperView);
var F12 = ComputeMatrix.F(new VectorOfPointF(tmatch.Left.ToArray()), new VectorOfPointF(tmatch.Middle.ToArray()));
var F23 = ComputeMatrix.F(new VectorOfPointF(tmatch.Middle.ToArray()), new VectorOfPointF(tmatch.Right.ToArray()));
var F13 = ComputeMatrix.F(new VectorOfPointF(tmatch.Left.ToArray()), new VectorOfPointF(tmatch.Right.ToArray()));
if (F12 == null || F23 == null || F13 == null)
{
info.Text = "Too few matches";
return;
}
var Fs = new List <Image <Arthmetic, double> > {
F12, F23, F13
};
var K = EstimateCameraFromImageSequence.K(Fs, left.Width, right.Height);
var Es = new List <Image <Arthmetic, double> >
{
ComputeMatrix.E(F12, K),
ComputeMatrix.E(F23, K),
ComputeMatrix.E(F13, K)
};
FindTransformation.DecomposeToRTAndTriangulate(tmatch.Left, tmatch.Middle, K, Es[0],
out Image <Arthmetic, double> R12, out Image <Arthmetic, double> t12, out Image <Arthmetic, double> X12);
FindTransformation.DecomposeToRTAndTriangulate(tmatch.Middle, tmatch.Right, K, Es[1],
out Image <Arthmetic, double> R23, out Image <Arthmetic, double> t23, out Image <Arthmetic, double> X23);
FindTransformation.DecomposeToRTAndTriangulate(tmatch.Left, tmatch.Right, K, Es[2],
out Image <Arthmetic, double> R13, out Image <Arthmetic, double> t13, out Image <Arthmetic, double> X13);
var Rs = new List <Image <Arthmetic, double> >
{
RotationConverter.MatrixToEulerXYZ(R12),
RotationConverter.MatrixToEulerXYZ(R23),
RotationConverter.MatrixToEulerXYZ(R13)
};
var ts = new List <Image <Arthmetic, double> >
{
t12,
t23,
t13
};
PrintMatricesInfo(Es, K, Rs, ts);
}
public static OdometerFrame GetOdometerFrame3(
Mat left, Mat middle, Mat right, double lastScale, out double thisScale,
Feature2D detector, Feature2D descriptor, DistanceType distanceType, double maxDistance,
Image <Arthmetic, double> K, double takeBest = 1.0)
{
thisScale = lastScale;
var match12 = MatchImagePair.Match(left, middle, detector, descriptor, distanceType, maxDistance);
var match23 = MatchImagePair.Match(middle, right, detector, descriptor, distanceType, maxDistance);
var match13 = MatchImagePair.Match(left, right, detector, descriptor, distanceType, maxDistance);
var left1 = match12.LeftPoints;
var right1 = match12.RightPoints;
var left2 = match23.LeftPoints;
var left2_X = MatchClosePoints.SortByX(match23.LeftPoints);
var right2 = match23.RightPoints;
var left3 = match13.LeftPoints;
var right3 = match13.RightPoints;
var right3_X = MatchClosePoints.SortByX(match13.LeftPoints);
TripletMatch tmatch = new TripletMatch();
List <MDMatch> m12 = new List <MDMatch>();
List <MDMatch> m23 = new List <MDMatch>();
for (int idx12 = 0; idx12 < left1.Size; ++idx12)
{
var p1 = left1[idx12];
var p2 = right1[idx12];
int idx23 = IndexOf_X(left2_X, p2);
if (idx23 != -1)
{
var p3 = right2[idx23];
int idx13 = IndexOf_X(right3_X, p1);
if (idx13 != -1)
{
if (AreEqual(left1[idx12], left3[idx13], maxDistance))
{
tmatch.Left.Add(p1);
tmatch.Middle.Add(p2);
tmatch.Right.Add(p3);
m12.Add(match12.Matches[idx12]);
m23.Add(match23.Matches[idx23]);
}
}
}
}
match12.Matches = new VectorOfDMatch(m12.ToArray());
match23.Matches = new VectorOfDMatch(m23.ToArray());
var F12 = ComputeMatrix.F(new VectorOfPointF(tmatch.Left.ToArray()), new VectorOfPointF(tmatch.Middle.ToArray()));
// var F23 = ComputeMatrix.F(new VectorOfPointF(tmatch.Middle.ToArray()), new VectorOfPointF(tmatch.Right.ToArray()));
var F13 = ComputeMatrix.F(new VectorOfPointF(tmatch.Left.ToArray()), new VectorOfPointF(tmatch.Right.ToArray()));
if (F12 == null || F13 == null)
{
return(null);
}
var Es = new List <Image <Arthmetic, double> >
{
ComputeMatrix.E(F12, K),
// ComputeMatrix.E(F23, K),
ComputeMatrix.E(F13, K)
};
FindTransformation.DecomposeToRTAndTriangulate(tmatch.Left, tmatch.Middle, K, Es[0],
out Image <Arthmetic, double> R12, out Image <Arthmetic, double> t12, out Image <Arthmetic, double> X12);
// FindTransformation.DecomposeToRT(Es[1], out Image<Arthmetic, double> R23, out Image<Arthmetic, double> t23);
FindTransformation.DecomposeToRTAndTriangulate(tmatch.Left, tmatch.Right, K, Es[1],
out Image <Arthmetic, double> R13, out Image <Arthmetic, double> t13, out Image <Arthmetic, double> X13);
var Rs = new List <Image <Arthmetic, double> >
{
R12,
R13
};
var ts = new List <Image <Arthmetic, double> >
{
t12,
t13
};
var cc = ComputeCameraCenter3(K, Rs, ts, tmatch);
OdometerFrame odometerFrame = new OdometerFrame();
odometerFrame.Rotation = RotationConverter.MatrixToEulerXYZ(Rs[0]);
odometerFrame.RotationMatrix = Rs[0];
odometerFrame.MatK = K;
odometerFrame.Match = match12;
// Image<Arthmetic, double> C = ComputeCameraCenter(R, t, K, match);
// odometerFrame.Translation = R.Multiply(C);
// odometerFrame.Translation = R.T().Multiply(odometerFrame.Translation);
odometerFrame.Translation = ts[0].Mul(lastScale / ts[0].Norm);
odometerFrame.Center = lastScale * cc.C12;
thisScale = cc.Ratio3To2;
return(odometerFrame);
}
public static void FindBestScale(Matrix R12, Matrix t12, Matrix R23, Matrix t23, Matrix K,
List <PointF> pts1, List <PointF> pts2, List <PointF> pts3, int minSampleSize,
out double scale, out double confidence, out List <int> inliers)
{
FindTransformation.TriangulateChieral(pts1, pts2, K, R12, t12, out var est3d_12);
FindTransformation.TriangulateChieral(pts2, pts3, K, R23, t23, out var est3d_23);
// Find best scale, so that both sets will be closest
RansacScaleEstimation ransacModel = new RansacScaleEstimation(est3d_12, est3d_23, R12, ComputeMatrix.Center(t12, R12));
int sampleSize = Math.Max(minSampleSize, (int)(0.05 * pts1.Count));
int minGoodPoints = (int)(0.4 * pts1.Count);
int maxIterations = 100;
double meanRefPointSize = GetMeanSize(est3d_12);
double threshold = meanRefPointSize * meanRefPointSize * 0.08;
var result = RANSAC.ProcessMostInliers(ransacModel, maxIterations, sampleSize, minGoodPoints, threshold, 1.0);
scale = (double)result.BestModel;
inliers = result.Inliers;
var backprojected23to12 = TransfromBack3dPoints(R12, t12, est3d_23, scale);
Image <Arthmetic, double> inliersOnly12 = new Image <Arthmetic, double>(result.Inliers.Count, 4);
Image <Arthmetic, double> inliersOnly23 = new Image <Arthmetic, double>(result.Inliers.Count, 4);
for (int i = 0; i < result.Inliers.Count; ++i)
{
int k = result.Inliers[i];
for (int j = 0; j < 4; ++j)
{
inliersOnly12[j, i] = est3d_12[j, k];
inliersOnly23[j, i] = backprojected23to12[j, k];
}
}
// Find errors
var distances = inliersOnly12.Sub(inliersOnly23);
double meanDistance = distances.Norm / distances.Cols;
double meanSize = GetMeanSize(inliersOnly12, inliersOnly23);
double relativeMeanDistance = meanDistance / meanSize;
double error = result.BestError;
double relativeError = error / (meanSize * meanSize);
Errors.TraingulationError(inliersOnly12, inliersOnly23, out double mean, out double median, out List <double> errs);
confidence = (double)inliers.Count / (double)pts1.Count;
}
