Weighting Observations

61 views

Skip to first unread message

Alan Buchanan

unread,

Dec 26, 2016, 7:00:54 AM12/26/16

to Ceres Solver

Hello,

I want to add weighting to my bundle adjustment.

My intent is to add extra observation types to my adjustment.

- Weighted Intrinsics (rather than holding fixed or constrained)

- Observations for the camera extreior orientation

- Ground control points

- Geometric constraints in the model

Hello,

I want to add weighting to my bundle adjustment.

My intent is to add extra observation types to my adjustment.

- Weighted Intrinsics (rather than holding fixed or constrained)

- Observations for the camera extreior orientation

- Ground control points

- Geometric constraints in the model

I have added weighting to the Reprojection Cost function and a cost function for adding Auxillary parameters (Intrinsics) with a weight (1/standard deviation) (see below)

However the results have the same effect as constraining the intrinsics. That is it is not pushing any of the errors onto the intrinsics.

I have experimented with very loose weighting for the Intrinsics with the same results.

Any advice is appreciated.

Alan

struct AuxillaryParameterError

{

AuxillaryParameterError(double f,double xp,double yp,double k1,double k2,double k3,double p1, double p2, double sd) :

f(f),xp(xp),yp(yp),k1(k1),k2(k2),k3(k3),p1(p1),p2(p2), sd(sd)

{}

template<typename T>

bool operator()( const T* const intrinsics,

T* residuals) const

{

const T& focal_length = intrinsics[OFFSET_FOCAL_LENGTH];

const T& principal_point_x = intrinsics[OFFSET_PRINCIPAL_POINT_X];

const T& principal_point_y = intrinsics[OFFSET_PRINCIPAL_POINT_Y];

const T& kk1 = intrinsics[OFFSET_K1];

const T& kk2 = intrinsics[OFFSET_K2];

const T& kk3 = intrinsics[OFFSET_K3];

const T& pp1 = intrinsics[OFFSET_P1];

const T& pp2 = intrinsics[OFFSET_P2];

// The error is the difference between the predicted and observed position.

residuals[0] = (focal_length - T(f))/sd;

residuals[1] = (principal_point_x - T(xp))/sd;

residuals[2] = (principal_point_y - T(yp))/sd;

residuals[3] = (kk1 - T(k1))/sd;

residuals[4] = (kk2 - T(k2))/sd;

residuals[5] = (kk3 - T(k3))/sd;

residuals[6] = (pp1 - T(p1))/sd;

residuals[7] = (pp2 - T(p2))/sd;

return true;

}

// Factory to hide the construction of the CostFunction object from

// the client code.

static ceres::CostFunction* Create(const double f,const double xp,const double yp,const double k1,const double k2,const double k3,const double p1,const double p2, const double sd)

{

return (new ceres::AutoDiffCostFunction<AuxillaryParameterError, 9, 8>(

new AuxillaryParameterError(f,xp,yp,k1,k2,k3,p1,p2,sd)));

}

double f,xp,yp,k1,k2,k3,p1,p2;

double sd;

};

Alan Buchanan

unread,

Dec 27, 2016, 4:20:08 AM12/27/16

to Ceres Solver

I need to put my original post on hold.

My problem is associated with passing standard deviation to the cost function.

The standard deviation was being considered as a parameter to be solved and therefore was falling over.