Q3D-Calibration/include/GaussFit.h

72 lines
1.6 KiB
C
Raw Normal View History

#pragma once
#ifndef utils_h
#define utils_h
#include "LevenbergMarquardt.h"
#include <Eigen/Dense>
#include <cmath>
#include <iostream>
double Gaussian(double x, double* p) {
return p[0] * exp(-(x - p[1]) * (x - p[1]) / (2 * p[2] * p[2]));
}
double* GaussianJacobian(double x, double* p) {
double* resJ = new double[3];
resJ[0] = -Gaussian(x, p) / p[0];
resJ[1] = -(x - p[1]) * Gaussian(x, p) / (p[2] * p[2]);
resJ[2] = -(x - p[1]) * (x - p[1]) * Gaussian(x, p) / (p[2] * p[2] * p[2]);
return resJ;
}
class GaussFit {
public:
GaussFit();
~GaussFit();
public:
void addData(double x, double y);
double* fit();
public:
double* parma = new double[3];
std::vector<Eigen::Vector2d> data;
};
GaussFit::GaussFit() {}
GaussFit::~GaussFit() {}
void GaussFit::addData(double x, double y) { data.push_back(Eigen::Vector2d(x, y)); }
double* GaussFit::fit() {
int n = data.size();
double x, y, sigma;
Eigen::VectorXd mY(n);
Eigen::MatrixXd mX(n, 3);
Eigen::MatrixXd mB(3, 1);
for (int i = 0; i < n; i++) {
Eigen::Vector2d& point = data.at(i);
x = point(0), y = point(1);
mY(i, 0) = log(y);
mX(i, 0) = 1, mX(i, 1) = x, mX(i, 2) = x * x;
}
mB = (mX.transpose() * mX).inverse() * mX.transpose() * mY;
sigma = -1 / mB(2, 0);
parma[1] = mB(1, 0) * sigma / 2;
parma[0] = exp(mB(0, 0) + parma[1] * parma[1] / sigma);
parma[2] = sqrt(sigma);
LevenbergMarquardt LM(3, parma, Gaussian, GaussianJacobian);
LM.data = data;
parma = LM.solve();
return parma;
}
#endif