CDCRobustSZFitter Class Reference

Utility class implementing robust versions of linear sz trajectory line fits. More...

#include <CDCRobustSZFitter.h>

Public Member Functions
CDCTrajectorySZ	fitUsingSimplifiedTheilSen (const CDCSZObservations &szObservations) const
	Fit a linear sz trajectory to the reconstructed stereo segment.
CDCTrajectorySZ	fitTheilSen (const CDCSZObservations &szObservations) const
	Implements the original Theil-Sen line fit algorithm.
CDCTrajectorySZ	fitWeightedTheilSen (const CDCSZObservations &szObservations) const
	Implements the weighted Theil-Sen line fit algorithm.

Private Member Functions
double	getMedianZ0 (const CDCSZObservations &szObservations, double tanLambda) const
	Compute the median z0 intercept from the given observations and an estimated slope.

Detailed Description

Utility class implementing robust versions of linear sz trajectory line fits.

Definition at line 20 of file CDCRobustSZFitter.h.

Member Function Documentation

fitTheilSen()

CDCTrajectorySZ fitTheilSen

(

const CDCSZObservations &

szObservations

)

const

Implements the original Theil-Sen line fit algorithm.

Does not estimate the covariances of the fit parameters.

Definition at line 57 of file CDCRobustSZFitter.cc.

{
  std::vector<double> tanLambdas;
  tanLambdas.reserve(szObservations.size() * (szObservations.size() - 1) / 2);
  for (unsigned int i = 0; i < szObservations.size(); i++) {
    for (unsigned int j = i + 1; j < szObservations.size(); j++) {
      double s1 = szObservations.getS(i);
      double s2 = szObservations.getS(j);
      if (s1 == s2) continue;
      double z1 = szObservations.getZ(i);
      double z2 = szObservations.getZ(j);
      tanLambdas.push_back((z2 - z1) / (s2 - s1));
    }
  }
 
  const double tanLambda = median(std::move(tanLambdas));
  const double z0 = getMedianZ0(szObservations, tanLambda);
 
  CDCTrajectorySZ trajectorySZ(tanLambda, z0);
  trajectorySZ.setNDF(szObservations.size());
  return trajectorySZ;
}

fitUsingSimplifiedTheilSen()

CDCTrajectorySZ fitUsingSimplifiedTheilSen

(

const CDCSZObservations &

szObservations

)

const

Fit a linear sz trajectory to the reconstructed stereo segment.

It uses the normal fitting algorithm but does so multiple times: In every iteration, one hit is excluded from the observation set and the rest is fitted. In the end, the mean over the fitting parameters is built and returned.

Does not estimate the covariances of the fit parameters.

TODO:

• Use the median.
• Use RANSAC instead of Theil-Sen.
• Think about the parameters better.

Definition at line 22 of file CDCRobustSZFitter.cc.

{
  // This seems to be some other algorithm
 
 
  CDCTrajectorySZ trajectorySZ;
  CDCSZFitter szFitter;
 
  if (observationsSZ.size() > 4) {
    CDCSZObservations observationsSZFiltered;
 
    double meanTanLambda = 0;
    double meanStartZ = 0;
 
    for (unsigned int i = 0; i < observationsSZ.size(); i++) {
      for (unsigned int j = 0; j < observationsSZ.size(); j++) {
        if (i != j) {
          observationsSZFiltered.fill(observationsSZ.getS(j),
                                      observationsSZ.getZ(j),
                                      observationsSZ.getWeight(j));
        }
      } // for j
 
      szFitter.update(trajectorySZ, observationsSZFiltered);
      meanTanLambda += trajectorySZ.getTanLambda();
      meanStartZ += trajectorySZ.getZ0();
    } // for i
 
    return CDCTrajectorySZ(meanTanLambda / observationsSZ.size(),
                           meanStartZ / observationsSZ.size());
  } else {
    return CDCTrajectorySZ::basicAssumption();
  }
}

fitWeightedTheilSen()

CDCTrajectorySZ fitWeightedTheilSen

(

const CDCSZObservations &

szObservations

)

const

Implements the weighted Theil-Sen line fit algorithm.

Does not estimate the covariances of the fit parameters.

Definition at line 80 of file CDCRobustSZFitter.cc.

{
  std::vector<WithWeight<double> > weightedTanLambdas;
  Weight totalWeight = 0;
  weightedTanLambdas.reserve(szObservations.size() * (szObservations.size() - 1) / 2);
  for (unsigned int i = 0; i < szObservations.size(); i++) {
    for (unsigned int j = i + 1; j < szObservations.size(); j++) {
      double s1 = szObservations.getS(i);
      double s2 = szObservations.getS(j);
      if (s1 == s2) continue;
      double z1 = szObservations.getZ(i);
      double z2 = szObservations.getZ(j);
 
      double w1 = szObservations.getWeight(i);
      double w2 = szObservations.getWeight(j);
 
      // Longer legs receive proportionally longer weight.
      Weight weight = std::abs(s2 - s1) * hypot2(w1, w2);
      weightedTanLambdas.emplace_back((z2 - z1) / (s2 - s1),  weight);
      totalWeight += weight;
    }
  }
 
  for (WithWeight<double>& weightedTanLambda : weightedTanLambdas) {
    weightedTanLambda.weight() /= totalWeight;
  }
 
  const double tanLambda = weightedMedian(std::move(weightedTanLambdas));
  const double z0 = getMedianZ0(szObservations, tanLambda);
 
  CDCTrajectorySZ trajectorySZ(tanLambda, z0);
  trajectorySZ.setNDF(szObservations.size());
  return trajectorySZ;
}

getMedianZ0()

double getMedianZ0 ( const CDCSZObservations &  szObservations, double  tanLambda  ) const

private

Compute the median z0 intercept from the given observations and an estimated slope.

Definition at line 115 of file CDCRobustSZFitter.cc.

{
  std::vector<double> z0s;
  z0s.reserve(szObservations.size());
  for (unsigned int i = 0; i < szObservations.size(); i++) {
    double s = szObservations.getS(i);
    double z = szObservations.getZ(i);
    z0s.push_back(z - s * tanLambda);
  }
  const double z0 = median(std::move(z0s));
  return z0;
}

---

The documentation for this class was generated from the following files:

• tracking/trackFindingCDC/fitting/include/CDCRobustSZFitter.h
• tracking/trackFindingCDC/fitting/src/CDCRobustSZFitter.cc