Research
Introduction
I worked on the background estimation problem. This problem (also known as background estimation, reconstruction, or initialization problem) consists in generating a unique image estimating the background of an input video sequence acquired from a fixed viewpoint. Generating an estimation of the background is helpful for many applications including video surveillance, segmentation, compression, inpainting, privacy protection, and computational photography [6]. As an example, background subtraction algorithms, able to classify any pixel of a sequence as belonging to the background or not, could benefit from such an estimation to initialize their model. We provide several methods for the generation of a background image: LaBGen and other variants. Note that all of the following methods have been extensively described and assessed in my PhD thesis!
LaBGen
LaBGen is an award-winning background generation method combining a pixel-wise median filter and a patch selection mechanism based on a motion detection performed by a background subtraction algorithm. It was introduced in [1], and is extensively described in [2]. It appears to be the best method among the ones submitted to the Scene Background Modeline and Initialization (SBMI) 2015 workshop, and has been ranked number one during the IEEE Scene Background Modeling Contest (SBMC) 2016. Its C++ source code, provided with a ready-to-use program, can be found on the following GitHub repository:
https://github.com/benlaug/labgen
The corresponding paper can be found at:
http://hdl.handle.net/2268/203572
Here is a video showing some backgrounds estimated by LaBGen:
If you use LaBGen in your work, please cite papers [2] and [1] as follows:
@article{Laugraud2017LaBGen,
title = {{LaBGen}: A method based on motion detection for generating the background of a scene},
author = {B. Laugraud and S. Pi{\'e}rard and M. {Van Droogenbroeck}},
journal = {Pattern Recognition Letters},
publisher = {Elsevier},
volume = {96},
pages = {12-21},
year = {2017},
doi = {10.1016/j.patrec.2016.11.022}
}
@inproceedings{Laugraud2015Simple,
title = {Simple median-based method for stationary background generation using background subtraction algorithms},
author = {B. Laugraud and S. Pi{\'e}rard and M. Braham and M. {Van Droogenbroeck}},
booktitle = {International Conference on Image Analysis and Processing (ICIAP), Workshop on Scene Background Modeling and Initialization (SBMI)},
publisher = {Springer},
series = {Lecture Notes in Computer Science},
volume = {9281},
pages = {477-484},
year = {2015},
month = {September},
address = {Genova, Italy},
doi = {10.1007/978-3-319-23222-5_58}
}
LaBGen-P
LaBGen-P is a simpler variant of LaBGen working at the pixel-level. It is described in [2] and has been ranked number two during the IEEE Scene Background Modeling Contest (SBMC) 2016. Its C++ source code, provided with a ready-to-use program, can be found on the following GitHub repository:
https://github.com/benlaug/labgen-p
The corresponding paper can be found at:
http://hdl.handle.net/2268/201146
Here is a video showing some backgrounds estimated by LaBGen-P:
If you use LaBGen-P in your work, please cite paper [3] as follows:
@inproceedings{Laugraud2016LaBGen-P,
title = {{LaBGen-P}: A Pixel-Level Stationary Background Generation Method Based on {LaBGen}},
author = {B. Laugraud and S. Pi{\'e}rard and M. {Van Droogenbroeck}},
booktitle = {IEEE International Conference on Pattern Recognition (ICPR), IEEE Scene Background Modeling Contest (SBMC)},
pages = {107-113},
year = {2016},
month = {December},
address = {Canc{\'u}n, Mexico},
doi = {10.1109/ICPR.2016.7899617}
}
LaBGen-OF
LaBGen-OF is a variant of LaBGen using optical flow algorithms (instead of background subtraction algorithms) for motion detection. It is described in [4]. Its C++ source code, provided with a ready-to-use program, can be found on the following GitHub repository:
https://github.com/benlaug/labgen-of
The corresponding paper can be found at:
http://hdl.handle.net/2268/213147
Here is a video showing some backgrounds estimated by LaBGen-OF:
If you use LaBGen-OF in your work, please cite paper [4] as follows:
@inproceedings{Laugraud2017IsAMemoryless,
title = {Is a Memoryless Motion Detection Truly Relevant for Background Generation with {LaBGen}?},
author = {B. Laugraud and M. {Van Droogenbroeck}},
booktitle = {Advanced Concepts for Intelligent Vision Systems (ACIVS)},
publisher = {Springer},
series = {Lecture Notes in Computer Science},
year = {2017},
month = {September},
address = {Antwerp, Belgium}
}
LaBGen-P-Semantic
LaBGen-P-Semantic is a variant of LaBGen-P leveraging semantic segmentation for motion detection. It is described in [5]. The corresponding paper can be found at:
http://hdl.handle.net/2268/225850
Here is a video showing some backgrounds estimated by LaBGen-P-Semantic:
If you use LaBGen-P-Semantic in your work, please cite paper [5] as follows:
@article{Laugraud2018LaBGen-P-Semantic,
title = {{LaBGen-P-Semantic}: A First Step for Leveraging Semantic Segmentation in Background Generation},
author = {B. Laugraud and S. Pi{\'e}rard and M. {Van Droogenbroeck}},
journal = {Journal of Imaging},
publisher = {MDPI},
volume = {4},
number = {7},
pages = {86},
month = {June},
year = {2018},
doi = {10.3390/jimaging4070086}
}
References
[6] L. Maddalena and A. Petrosino. Background model initialization for static cameras. In Background Modeling and Foreground Detection for Video Surveillance, chapter 3, pages 3.1–3.16. Chapman and Hall/CRC, 2014.