Bayesian Recursive Estimation Workspace in MATLAB

---

A MATLAB wrapper for target tracking applications using BREW.

This package holds general dynamics models, different target models, and filters that work together.

NOTE: This code is still in development and is in no way promised to perform to any specific standard.

Getting Started:

To get started, run the function

build_brew()

After the compile is done, the wrapper should work as intended. A folder +BREW will be created automatically using generator_files/generate_mex.py, and then +utils holds plotting and sampling methods specific for MATLAB for easy-to-run testing.

The tests/ folder should ensure the wrapper is working while showing how to initialize the algorithms.

Adding New C++ Classes

When you add a new class to the brew C++ library, annotate it with @mex comment blocks and run build_brew. The generator will automatically create the MEX gateway code and MATLAB wrappers.

Annotation Reference

Place annotations directly above the class declaration in the C++ header.

Dynamics (stored as handle, skeleton MATLAB wrapper):

// @mex dynamics
// @mex_name MyDynamics
// @mex_args dims:int, param:double
class MyDynamics : public DynamicsBase { ... };

Models (value class in MATLAB, used for filter I/O):

// @mex model
// @mex_name MyModel
// @mex_fields alpha:scalar, beta:scalar, mean:vec, covariance:mat, v:scalar, V:mat
class MyModel : public BaseSingleModel { ... };

Trajectory models (wraps a base model with windowed state):

// @mex model
// @mex_name TrajectoryMyModel
// @mex_trajectory MyModel

Filters (handle class, stub predict/correct for subclassing):

// @mex filter
// @mex_name MyFilter
// @mex_dist MyModel
// @mex_setters window_size:int, decay:scalar, noise:mat
// @mex_handle_setters icp:IcpBase
class MyFilter : public Filter<models::MyModel> { ... };

RFS filters (handle class, full predict/correct/cleanup via C++):

// @mex rfs
// @mex_name MyRFS
// @mex_params threshold:double:1e-4, max_components:int:100
// @mex_init set_intensity
// @mex_has cardinality, track_histories, birth_weights
// @mex_optional_params poisson_rate:double
template <typename T> class MyRFS : public RFSBase { ... };

Clustering (handle class):

// @mex clustering
// @mex_name MyClustering
// @mex_args epsilon:double, min_pts:int
class MyClustering { ... };

ICP algorithms (handle class with align method):

// @mex icp
// @mex_name MyIcp
// @mex_namespace template_matching
// @mex_args inner:clone:IcpBase, template:mat_pc
// @mex_params max_iterations:int:50, tolerance:double:1e-6
class MyIcp : public IcpBase { ... };

Models with handle/matrix constructor fields (e.g., TemplatePose):

// @mex model
// @mex_name MyPose
// @mex_fields mean:vec, covariance:mat, rotation:mat
// @mex_create_mat_fields template_points:PointCloud
// @mex_create_int_vec_fields pos_indices

Field Types

Type	C++	MATLAB (mixture creation)	MATLAB (model class)
vec	Eigen::VectorXd	cell of column vectors	property
mat	Eigen::MatrixXd	cell of matrices	property
scalar	double	double array	property
int	int	scalar	property
clone:Type	unique_ptr<Type> (cloned from handle)	object with .handle_	-
mat_pc	PointCloud (from matrix)	raw d x N matrix	-

What Gets Generated

Source	Output	Overwrite behavior
@mex dynamics	+BREW/+dynamics/<Name>.m	Skip if exists
@mex model	+BREW/+models/<Name>.m + <Name>Mixture.m	Always
@mex filter	+BREW/+filters/<Name>.m	Always
@mex rfs	+BREW/+multi_target/<Name>.m	Always
@mex clustering	(MEX command only)	Always
@mex icp	+BREW/+template_matching/<Name>.m	Always
All annotations	generator_files/brew_mex.cpp	Always
All models	+BREW/+models/Mixture.m	Always

Example 1:

Gamma Gaussian Inverse Wishart applied to PHD filter for simulated measurements using the GGIW Mixture model. Output of test_PHD_GGIW.m.

Example 2:

Trajectory Set Theory GM-PHD filter implemented as an alternative to labeled RFS. Output of test_trajectory_gm_phd.m.

You can see the L-scan window in red, and previous target trajectories in the other colors. The L-scan is performed as an efficient implementation of the trajectory set theory filters.

Example 3:

Same test as Example 1, but with the TST-GGIW. Output of test_PHD_TrajectoryGGIW.m.