Skip to content

jorgesep/BGS

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Software for evaluating BGS algorithms

This Repository contains the implementation of some background subtraction methods, based on statistical modeling of pixels behavior within an image sequence. The background behavior of an image sequence can be modeled either with a mixture of Gaussian distributions[1] or a set of Gaussian Kernels[2]. The code is set of C++ classes along with the OpenCV library. A binary program is also available to evaluate the results of background subtraction algorithms included in this repository; the foreground masks generated by the methods can be compared against a ground truth (reference frame). The program compares pixel-to-pixel the produced silhouettes with its ground truth, resulting in a group of True/False Positive/Negative metrics which are combined to produce a general F1 score measure.

The software is the outcome of a software project[4] aimed to automatically generate a set of segmented silhouettes from a common human actions dataset. Several executions of the algorithms to separate foreground from the background were conducted, they were applied over MuHAVI[3] dataset; a dedicated collection of common human actions (run, walk, jump, etc) played on a scene with a night rough environment. On the first part of the proejct, the specific background subtraction methods run on MuHAVi (Multicamera Human Action Video). While the second part, the generated foreground masks were compared with the set of manually annotate images (ground-truth) available on the dataset (MuHAVi-MAS, manually annotated set). The F1-score was the metric to make performance evaluation of the algorithms. Finally the whole MuHAVI dataset was processed choosing the best score algorithm, resulting in a set of no perfect segmented silhouettes (MuHAVi "uncut") available to experiment with human action recognition algorithms.

Getting Started

These instructions will get you a copy of the project up and running on a Linux machine (e.g. Ubuntu, Mint). All BGS modules can be build outside of the source directory with CMake tool.

Prerequisites

Check that all OpenCV libraries must be installed previous to build BGS modules.

sudo apt-get update
sudo apt-get install cmake git libopencv-dev libboost-all-dev

Installing

Get BGS modules from github

git clone https://github.com/jorgesep/BGS.git

Build BGS modules

cd BGS
mkdir build
cd build
cmake ..
make

References

  1. Chen, Z. & Ellis, T. (2011). Self-adaptive Gaussian mixture model for urban traffic monitoring system.. ICCV Workshops (p./pp. 1769-1776), : IEEE. ISBN: 978-1-4673-0062-9
  2. Elgammal, A. M., Harwood, D. & Davis, L. S. (2000). Non-parametric Model for Background Subtraction.. In D. Vernon (ed.), ECCV (2) (p./pp. 751-767), : Springer. ISBN: 3-540-67686-4
  3. Singh S., Velastin, S.A. & Hossein H. (2010). ’MuHAVi: A Multicamera Human Action Video Dataset for the Evaluation of Action Recognition Methods’. In Proceedings of the 2010 7th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS ’10). IEEE Computer Society, Washington, DC, USA, 48-55.
  4. Sepúlveda, J., Velastin, S.A.: ’F1 Score Assesment of Gaussian Mixture Background Subtraction Algorithms Using the MuHAVi Dataset’., 2015., 6th International Conference on Imaging for Crime Prevention and Detection (ICDP-15). ISBN: 978-1-78561-131-5

About

Background subtraction methods, based on statistical modeling of pixels behavior within an image sequence.

Resources

Stars

Watchers

Forks

Packages

No packages published