Bayesian Recursive Estimation Workspace (BREW)

---

A header-heavy C++ library for single-target and multi-target tracking, supporting both point and extended targets. Built on Eigen and C++20.

Modules

Module	Description
models	State distribution representations (point, extended, trajectory-aware)
dynamics	Motion models for prediction
filters	Single-target recursive filters (prediction + update)
multi_target	Random Finite Set (RFS) multi-target tracking filters
fusion	Mixture management operations (prune, merge, cap)
assignment	Optimal and K-best assignment solvers
clustering	Measurement clustering for extended targets
metrics	Performance metrics (OSPA, GOSPA)
serialization	Filter state serialization (nlohmann/json)
plot_utils	Optional visualization utilities (requires matplot++)

Building

mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
cmake --build .

Eigen and GTest are fetched automatically if not found on the system. Plotting support is enabled by default and can be disabled with -DBREW_ENABLE_PLOTTING=OFF.

Quick Start

All multi-target filters share a common template interface parameterized on distribution type. A typical tracking loop looks like:

#include <brew/multi_target/pmbm.hpp>
#include <brew/filters/ekf.hpp>
#include <brew/models/gaussian.hpp>
#include <brew/dynamics/integrator_2d.hpp>

// 1. Set up dynamics and single-target filter
auto dynamics = std::make_shared<brew::dynamics::Integrator2D>(dt);
auto ekf = std::make_shared<brew::filters::EKF>(dynamics, R);

// 2. Create birth model
auto birth = std::make_shared<brew::models::Mixture<brew::models::Gaussian>>();
birth->add_component(birth_gaussian, birth_weight);

// 3. Create multi-target filter
brew::multi_target::PMBM<brew::models::Gaussian> tracker(ekf, birth);
tracker.set_prob_detection(0.9);
tracker.set_prob_survive(0.99);
tracker.set_clutter_rate(10.0);

// 4. Predict-update loop
for (const auto& measurements : measurement_sequence) {
    tracker.predict();
    tracker.update(measurements);
    auto estimates = tracker.estimate();
}

Any RFS filter can be swapped in by changing the type (e.g. PHD, MBM, LMB, GLMB) — the predict/update/estimate interface is the same. Swap the distribution type template parameter to switch between point and extended target tracking.

Testing

cd build
ctest --output-on-failure

License

See LICENSE.