Interface

Intarfaces and implementation helper classes. More...

Collaboration diagram for Interface:

Classes
class	GeneralizedModelBuilder< ErrorAccumClassT >
class	Model< ModelT, DatumT >
class	ModelBuilder< ModelClassT >
	Interface for model builders used in RanSaC. More...
class	MultiSolutionModel< ErrorAccumulatorT, CAllocTT >
class	RanSaCOptions< ErrorAccumClassT >
class	SingleSolutionModel< ErrorAccumulatorT >
	Implements model inteface for models which have at most one solution. More...
Modules
	Error Functions
	Data Selectors
Files
file	model.h
	Model and ModelBuilder interfaces used by RanSaC.
Typedefs
typedef Model< Matrix3d, ImageMatch >	EMatrixModel
typedef ModelBuilder < EMatrixModel >	EMatrixModelBuilder
typedef std::vector< bool >	MaskVec
	mask vector for indication of inliers in data vector (RanSaC).
Functions
template<class ErrorAccumClassT >
shared_ptr< typename ErrorAccumClassT::ModelClassT >	ranSaC (shared_ptr< ModelBuilder< typename ErrorAccumClassT::ModelClassT > > model_builder, shared_ptr< DataSelector< typename ErrorAccumClassT::ModelClassT::DataVec > > data_selector, const typename ErrorAccumClassT::ModelClassT::DataVec &data)
	model estimation using ransac.

---

Detailed Description

Intarfaces and implementation helper classes.

In the following we describe how to use this module to implement specific robust estimation and show it on simple example. More useful examples can be found in EMatEst and RanSaC__triangulation.

Description of basic types:

• "ModelT": Data structure that holds the parameters of the model to be estimated.
• "DatumT": Data structure that holds one piece of observation of the situation given by model (which is completely determined by ModelT).
• "ErrorT": Represents a point in the space of errors given by particualr model and its observations.
• "ErrorAccumT" : XXX.

The description of interface concepts:

• error function class concept: DatumErrorFunc is a template base class that defines the interface which all error function classes have to implement. This class is a function object class that computes error of an observation given the observation and a model.

• error accumulator class concept: ErrorAccumulator is a base class template that defines the interface which all error accumulator function classes have to implement. This function includes error function fuctionality (error function object is specified in the constructor), datum inlier test based on point in error space and eccumulator fuction that accumulates errors into ErrorAccumT (this does not affect model estimation in any way and is intendet for debuggigng purposes --- e.g. RealErrorMaxErrorAccumulator computes maximum error over all data, where ErrorT and ErrorAccumT are real numbers)

• abstract class template Model : We provide two implementations of Model + SingleSolutionModel, which assumes that, based on selected subset of available observation, we construct at most one model (i.e. one instace of type ModelT). + MultiSolutionModel, which does not have such restriction.

• abstract class template ModelBuilder : This is an interface for function object style classes. Some subset of available observation of size 'k' (the model size) are selected and pointer to Model is returned. The implementations of this usually utilize SingleSolutionModel or MultiSolutionModel and thus only the implementation of how to determine ModelT from data is needed.

• abstract class template DataSelector : The implementations are initialized with available observations and desired model size, and the method DataSelector::select implements 'k' selections of subsets of observations and on call returns null. 'k' is defined by the particular implementation.

Example:

class CircleAbsError : public DatumErrorFunc<double, Vector2d, double>{
      ErrorT operator()(const ModelT& model, const DatumT& obs){
          return abs(obs.norm() - model);
      }
}

template<ErrorAccumObjT> CircleModelBuilder : public GeneralizedModelBuilder<ErrorAccumObjT>{
    public:      
    CircleModelBuilder(const ErrorAccumClassT& _error_accum) : 
            GeneralizedModelBuilder<ErrorAccumClassT>(_error_accum) { } 
 
    virtual shared_ptr<typename ErrorAccumClassT::ModelClassT> operator()(
            const typename ErrorAccumClassT::ModelClassT::DataVec & selected){
        return shared_ptr<typename ErrorAccumClassT::ModelClassT>(
                new(SingleSolutionModel<ErrorAccumObjT>(1, selected, selected[0], this->error_accum)));
    }

}

double computeModel(const ALIGNED<Vector2d>::vector& data){
    typedef RealErrorMaxErrorAccumulator<CircleAbsError> EAccT;
    EAccT eacc(CircleAbsError());
    shared_ptr<ModeB> model_builder = CircleModelBuilder<EAccT>(eacc);

    //run 100 iterations of ransac.
    shared_ptr<typename EAccT::ModelClassT> model = ranSaC(
            model_builder, RandomDataSelector(data, 1, 100), data);

    if(model && (model->getInlierCount() > 0.2 * data.size())){
        std::cout << "model:" << model->getModel() << std::endl;
        std::cout << "# of inliers:" << model->getInlierCount() << std::endl;
        std::cout << "max error:" << model->getError() << std::endl;
        std::cout << "max inlier error:" << model->getInlierError() << std::endl;
    } else{
        std::cout << "did not find any model" << std::endl;
    }
}

---

Typedef Documentation

typedef Model<Matrix3d, ImageMatch> EMatrixModel

Definition at line 107 of file model.h.

typedef ModelBuilder<EMatrixModel> EMatrixModelBuilder

Definition at line 213 of file model.h.

typedef std::vector<bool> MaskVec

mask vector for indication of inliers in data vector (RanSaC).

Definition at line 18 of file model.h.

---

Function Documentation

template<class ErrorAccumClassT >

shared_ptr<typename ErrorAccumClassT::ModelClassT> ranSaC	(	shared_ptr< ModelBuilder< typename ErrorAccumClassT::ModelClassT > >	model_builder,
		shared_ptr< DataSelector< typename ErrorAccumClassT::ModelClassT::DataVec > >	data_selector,
		const typename ErrorAccumClassT::ModelClassT::DataVec &	data
	)

model estimation using ransac.

number of RanSaC iterations.

Parameters:

min_inliers

minimum number of inliers for a given model to be accepted.

Definition at line 321 of file model_estimation.h.