Aims
Segmentation of moving foreground objects from a video stream captured by a static cameras is the fundamental step in many applications such as visual surveillance and analysis of human activities, visual observation of animals and insect behaviors, and human-machine interaction. In these applications, the first step is the background subtraction process.
Since the beginning, different well-known background modeling approaches made their own breakthrough such as Gaussians models, sample-based methods and robust subspace learning models. Many scientific efforts have been reported in the literature to improve leader or classical methods in a more progressive way in applications where challenges are becoming more complex. But, no algorithm is able to simultaneously address all the key challenges that are present in videos during long sequences as in the real cases. Since 2016, with the seminal work of Braham and Van Droogenbroeck, a large number of studies on deep neural networks (CNNs, GANs) applied to background subtraction have been published, and a continual gain of performance has been achieved. However, the top background subtraction methods currently compared in CDnet 2014 are based on deep neural networks, and have demonstrated a large performance improvement in comparison to conventional unsupervised approaches based on multi-features or multi-cue strategies. But, their main drawbacks are their computational and memory requirements, and also their supervised aspects requiring labeling of a large amount of data.
The goals of this special session are thus three-fold: 1) designing background subtraction algorithms for real applications in urban and natural environments; 2) proposing new adaptive and incremental algorithms using supervised or unsupervised machine learning models; and 3) proposing robust algorithms to handle the key challenges in background subtraction in the case of either static or moving cameras.
Papers are solicited to address background modeling and subtraction algorithms based on mathematics, machine learning and signal processing theories.
Program
Session 1 – Unsupervised Approaches
« Summarizing the performances of a background subtraction algorithm measured on several videos »
Sébastien Piérard (University of Liège, Belgium), Marc Van Droogenbroeck (University of Liège, Belgium)
« CS-RPCA: Clustered Sparse RPCA for Moving Object Detection »
Sajid Javed (Khalifa University of Science and Technology, UAE), Arif Mahmood, Information Technology University, Pakistan, Jorge Dias, Naoufel Werghi (Khalifa University of Science and Technology, UAE)
« On The Structures of Representation for The Robustness of Semantic Segmentation to Input Corruption »
Charles Lehman (Georgia Institute of Technology, USA), Dogancan Temel (Georgia Institute of Technology, USA), Ghassan AlRegib (Georgia Institute of Technology, USA)
« Real-Time Semantic Background Subtraction »
Anthony Cioppa (University of Liège, Belgium), Marc Van Droogenbroeck (University of Liège, Belgium), Marc Braham (University of Liège, Belgium)
« Dual Information-Based Background Model for Moving Object Detection »
Sujoy Madhab Roy (Indian Statistical Institute, India), Thierry Bouwmans (University of La Rochelle, France)
Session 2 – Semi-supervised and Supervised Approaches
« Semi-Supervised Background Subtraction of Unseen Videos: Minimization of The Total Variation of Graph Signals »
Jhony Heriberto Giraldo Zuluaga (La Rochelle Université, France), Thierry Bouwmans (La Rochelle Université, France)
« Deep Autoencoder Architectures For Foreground Object Detection In Video Sequences Based On Probabilistic Mixture Models »
Jorge García-González, Miguel Ángel Molina-Cabello, Rafael Marcos Luque-Baena, Juan Miguel Ortiz-de-Lazcano-Lobato, Ezequiel López-Rubio (University of Málaga, Spain)
« Dynamic Background Subtraction Using Least Square Adversarial Learning »
Maryam Sultana (KNU, Republic of Korea), Arif Mahmood (ITU, Pakistan), Thierry Bouwmans (Universite de La Rochelle, France), Soon Ki Jung (KNU, Republic of Korea)
« Rethinking Background and Foreground in Deep Neural Network-Based Background Subtraction »
Tsubasa Minematsu (Kyushu university, Japan), Atsushi Shimada (Kyushu university, Japan), Rin-ichiro Taniguchi (Kyushu university, Japan)
« Robustness and Overfitting Behavior of Implicit Background Models »
Shirley Liu (Georgia Institute of Technology, USA), Charles Lehman (Georgia Institute of Technology, USA), Ghassan AlRegib (Georgia Institute of Technology, USA)