Multi-object Filters demo

This is a README file for the Python implementation of a set of multiple multi-object filters based on point processes (or random finite sets), using Gaussian mixtures, along with a demonstration script to compare the filters. In particular, the following filters have been implemented:

• Probability Hypothesis Density (PHD) filter
• Cardinalized PHD (CPHD) filter
• Discrete-gamma with Marks (DGM) filter (DG filter with labels)
• Generalized Labeled Multi-Bernoulli (GLMB) filter
• Joint GLMB (JGLMB) filter (with joint prediction and update steps)
• Linear Complexity Cumulant (LCC) filter
• Linear Complexity Cumulant with Marks (LCCM) filter (LCC filter with labels)

The implementation of the PHD, CPHD, GLMB and JGLMB filters are based on the Matlab implementation of those filters made public in Ba Tuong Vo's web page. The implementation of the CPHD filter has been changed by the owner of this repository to make it more efficient by precomputing many factors necessary in the prediction and update of the cardinality distribution.

Licensing

This code is licensed by Flávio Eler De Melo under the Apache 2.0 license. A copy of the terms and conditions can be found in the LICENSE file.

Contact

Please get in touch by messages for addressing doubts or suggestions. If you have found a bug, please open a ticket.

Requirements

• Python >= 3.5
• Numpy
• Scipy
• Termcolor
• Argparse
• Munkres >= 1.0.12
• Pickle5
• Nuitka
• PyYAML
• Natsort
• Azure-batch
• Azure-mgmt-batch
• Azure-identity

Installation

1. Clone the repository

git clone https://github.com/femelo/multi-object-filters

2. Install Python dependencies

Change directory to the code base directory

cd multi-object-filters

and install required packages

python3 -m pip install -r requirements.txt

Usage

For example, to run the PHD, CPHD and LCC filters (and compare their outputs) with the default parameters in verbose mode, type

python3 demo.py -f phd cphd lcc -v

All options for the demo script can be seen by checking the help as:

python3 demo.py -h

All figures for performance evaluation are saved in the folder 'figures/'.

Precompilation

The application can be precompiled to run faster via Nuitka. First install the patchelf package:

sudo apt-get install patchelf

Then, for the compilation, just run the bash script:

./build.sh

The created standalone application demo.bin can be called as, for instance,

./demo.bin -f phd cphd lcc -v