contrib/brl/bbas/bsta/algo/bsta_adaptive_updater.h

Go to the documentation of this file.

// This is brl/bbas/bsta/algo/bsta_adaptive_updater.h
#ifndef bsta_adaptive_updater_h_
#define bsta_adaptive_updater_h_
//:
// \file
// \brief Adaptive updaters for mixtures of Gaussians (i.e. Grimson and similar)
// \author Matt Leotta (mleotta@lems.brown.edu)
// \date July 13, 2005
//
// This file contains updaters based on
// "Adaptive background mixture models for real-time tracking"
// Grimson et. al.
//
// \verbatim
//  Modifications
//   (none yet)
// \endverbatim

#include <bsta/bsta_distribution.h>
#include <bsta/bsta_mixture.h>
#include <bsta/bsta_mixture_fixed.h>
#include <bsta/bsta_attributes.h>
#include <bsta/bsta_gauss_ff3.h>
#include "bsta_gaussian_updater.h"
#include "bsta_gaussian_stats.h"


//: A mixture of Gaussians adaptive updater
//  Base class for common functionality in adaptive updating schemes
template <class mix_dist_>
class bsta_mg_adaptive_updater
{
 private:
  typedef typename mix_dist_::dist_type obs_gaussian_; //mixture comp type
  typedef typename obs_gaussian_::contained_type gaussian_; //num_obs parent
  typedef typename gaussian_::math_type T;//the field type, e.g. float
  typedef typename gaussian_::vector_type vector_;// the vector type

 public:
  //: for compatibility with vpdl/vpdt
  typedef typename gaussian_::field_type field_type;
  typedef mix_dist_ distribution_type;

 protected:
  //: Constructor
  bsta_mg_adaptive_updater(const gaussian_& model,
                           unsigned int max_cmp = 5)
   : init_gaussian_(model,T(1)),
     max_components_(max_cmp) {}

  //: insert a sample in the mixture
  void insert(mix_dist_& mixture, const vector_& sample, T init_weight) const
  {
    bool removed = false;
    if (mixture.num_components() >= max_components_){
      removed = true;
      do {
        mixture.remove_last();
      } while (mixture.num_components() >= max_components_);
    }

    // if a mixture is removed renormalize the rest
    if (removed){
      T adjust = T(0);
      for (unsigned int i=0; i<mixture.num_components(); ++i)
        adjust += mixture.weight(i);
      adjust = (T(1)-init_weight) / adjust;
      for (unsigned int i=0; i<mixture.num_components(); ++i)
        mixture.set_weight(i, mixture.weight(i)*adjust);
    }
    init_gaussian_.set_mean(sample);
    //The following statement was originally conditionally applied, which can 
    //insert a single component with weight == init_weight, not one.
    if(mixture.num_components()>0)
      mixture.insert(init_gaussian_,init_weight);
    else // added by JLM June 19, 2010, make weight==1 if only one component
      mixture.insert(init_gaussian_,T(1));
  }

  //: A model for new Gaussians inserted
  mutable obs_gaussian_ init_gaussian_;
  //: The maximum number of components in the mixture
  unsigned int max_components_;
};


//: A mixture of Gaussians statistical updater
//  This updater treats all data equally
template <class mix_dist_>
class bsta_mg_statistical_updater : public bsta_mg_adaptive_updater<mix_dist_>
{
 public:
  typedef typename mix_dist_::dist_type obs_gaussian_;
  typedef typename obs_gaussian_::contained_type gaussian_;
  typedef typename gaussian_::math_type T;
  typedef typename gaussian_::vector_type vector_;
  typedef bsta_num_obs<mix_dist_> obs_mix_dist_;

  //: for compatibility with vpdl/vpdt
  typedef obs_mix_dist_ distribution_type;

  enum { data_dimension = gaussian_::dimension };

  //: Constructor
  bsta_mg_statistical_updater(const gaussian_& model,
                              unsigned int max_cmp = 5,
                              T g_thresh = T(3),
                              T min_stdev = T(0))
    : bsta_mg_adaptive_updater<mix_dist_>(model, max_cmp),
      gt2_(g_thresh*g_thresh), min_var_(min_stdev*min_stdev) {}

  //: The main function
  void operator() ( obs_mix_dist_& mix, const vector_& sample ) const
  {
    mix.num_observations += T(1);
    this->update(mix, sample, T(1)/mix.num_observations);
  }

  void update( mix_dist_& mix, const vector_& sample, T alpha ) const;
#if 0
  void update( mix_dist_& mix, const T & sample, T alpha ) const;
#endif
  //: Squared Gaussian Mahalanobis distance threshold
  T gt2_;
  //: Minimum variance allowed in each Gaussian component
  T min_var_;
};


//: A mixture of Gaussians statistical updater
//  This updater treats all data equally
template <class mix_dist_>
class bsta_mg_window_updater : public bsta_mg_statistical_updater<mix_dist_>
{
 public:
  typedef typename mix_dist_::dist_type obs_gaussian_;
  typedef typename obs_gaussian_::contained_type gaussian_;
  typedef typename gaussian_::math_type T;
  typedef typename gaussian_::vector_type vector_;
  typedef bsta_num_obs<mix_dist_> obs_mix_dist_;

  //: for compatibility with vpdl/vpdt
  typedef obs_mix_dist_ distribution_type;

  enum { data_dimension = gaussian_::dimension };

  //: Constructor
  bsta_mg_window_updater(const gaussian_& model,
                         unsigned int max_cmp = 5,
                         T g_thresh = T(3),
                         T min_stdev = T(0),
                         unsigned int window_size = 40)
  : bsta_mg_statistical_updater<mix_dist_>(model, max_cmp, g_thresh, min_stdev),
    window_size_(window_size) {}

  //: The main function
  void operator() ( obs_mix_dist_& mix, const vector_& sample ) const
  {
    if (mix.num_observations < window_size_)
      mix.num_observations += T(1);
    this->update(mix, sample, T(1)/mix.num_observations);
  }

 protected:
  unsigned int window_size_;
};

//: A mixture of Gaussians weighted statistical updater
//  This updater treats data according to a specified weight
template <class mix_dist_>
class bsta_mg_weighted_updater : bsta_mg_statistical_updater<mix_dist_>
{
 public:
  typedef typename mix_dist_::dist_type obs_gaussian_;
  typedef typename obs_gaussian_::contained_type gaussian_;
  typedef typename gaussian_::math_type T;
  typedef typename gaussian_::vector_type vector_;
  typedef bsta_num_obs<mix_dist_> obs_mix_dist_;

  //: for compatibility with vpdl/vpdt
  typedef obs_mix_dist_ distribution_type;

  enum { data_dimension = gaussian_::dimension };

  //: Constructor
  bsta_mg_weighted_updater(const gaussian_& model,
                           unsigned int max_cmp = 5,
                           T g_thresh = T(3),
                           T min_stdev = T(0))
    : bsta_mg_statistical_updater<mix_dist_>(model, max_cmp, g_thresh, min_stdev){}

  //: The main function
  void operator() ( obs_mix_dist_& mix, const vector_& sample, const T weight ) const
  {
    mix.num_observations += weight;
    this->update(mix, sample, weight/mix.num_observations);
  }
};


//: A mixture of Gaussians statistical updater
//  Using the grimson approximation to prior probabilities
template <class mix_dist_>
class bsta_mg_grimson_statistical_updater : public bsta_mg_adaptive_updater<mix_dist_>
{
 public:
  typedef typename mix_dist_::dist_type obs_gaussian_;
  typedef typename obs_gaussian_::contained_type gaussian_;
  typedef typename gaussian_::math_type T;
  typedef typename gaussian_::vector_type vector_;
  typedef bsta_num_obs<mix_dist_> obs_mix_dist_;

  //: for compatibility with vpdl/vpdt
  typedef obs_mix_dist_ distribution_type;

  enum { data_dimension = gaussian_::dimension };

  //: Constructor
  bsta_mg_grimson_statistical_updater(const gaussian_& model,
                                      unsigned int max_cmp = 5,
                                      T g_thresh = T(3),
                                      T min_stdev = T(0) )
    : bsta_mg_adaptive_updater<mix_dist_>(model, max_cmp),
      gt2_(g_thresh*g_thresh), min_var_(min_stdev*min_stdev) {}

  //: The main function
  void operator() ( obs_mix_dist_& mix, const vector_& sample ) const
  {
    mix.num_observations += T(1);
    this->update(mix, sample, T(1)/mix.num_observations);
  }

  void update( mix_dist_& mix, const vector_& sample, T alpha ) const;

  //: Squared Gaussian Mahalanobis distance threshold
  T gt2_;
  //: Minimum variance allowed in each Gaussian component
  T min_var_;
};

//: A mixture of Gaussians window updater
//  Using the grimson approximation to prior probabilities
template <class mix_dist_>
class bsta_mg_grimson_window_updater : public bsta_mg_grimson_statistical_updater<mix_dist_>
{
 public:
  typedef typename mix_dist_::dist_type obs_gaussian_;
  typedef typename obs_gaussian_::contained_type gaussian_;
  typedef typename gaussian_::math_type T;
  typedef typename gaussian_::vector_type vector_;
  typedef bsta_num_obs<mix_dist_> obs_mix_dist_;

  //: for compatibility with vpdl/vpdt
  typedef obs_mix_dist_ distribution_type;

  enum { data_dimension = gaussian_::dimension };

  //: Constructor
  bsta_mg_grimson_window_updater(const gaussian_& model,
                                 unsigned int max_cmp = 5,
                                 T g_thresh = T(3),
                                 T min_stdev = T(0),
                                 unsigned int window_size = 40)
    : bsta_mg_grimson_statistical_updater<mix_dist_>(model, max_cmp, g_thresh, min_stdev),
      window_size_(window_size) {}

  //: The main function
  void operator() ( obs_mix_dist_& mix, const vector_& sample ) const
  {
    if (mix.num_observations < window_size_)
      mix.num_observations += T(1);
    this->update(mix, sample, T(1)/mix.num_observations);
  }

 protected:
  unsigned int window_size_;
};

//: A mixture of Gaussians statistical weighted updater
//  Using the grimson approximation to prior probabilities
template <class mix_dist_>
class bsta_mg_grimson_weighted_updater : bsta_mg_grimson_statistical_updater<mix_dist_>
{
 public:
  typedef typename mix_dist_::dist_type obs_gaussian_;
  typedef typename obs_gaussian_::contained_type gaussian_;
  typedef typename gaussian_::math_type T;
  typedef typename gaussian_::vector_type vector_;
  typedef bsta_num_obs<mix_dist_> obs_mix_dist_;

  //: for compatibility with vpdl/vpdt
  typedef obs_mix_dist_ distribution_type;

  enum { data_dimension = gaussian_::dimension };

  //: Constructor
  bsta_mg_grimson_weighted_updater(const gaussian_& model,
                                   unsigned int max_cmp = 5,
                                   T g_thresh = T(3),
                                   T min_stdev = T(0) )
    : bsta_mg_grimson_statistical_updater<mix_dist_>(model, max_cmp, g_thresh, min_stdev){}

  //: The main function
  void operator() ( obs_mix_dist_& mix, const vector_& sample, const T weight ) const
  {
    mix.num_observations += weight;
    this->update(mix, sample, weight/mix.num_observations);
  }
};


#endif // bsta_adaptive_updater_h_