Thierry Bouwmans is an Associate Professor at the University of La Rochelle, France. His research interests consist mainly in the detection of moving objects in challenging environments as it is a pre-step for behaviour detection in video surveillance. He has authored more than 100 papers in refereed international journals and conferences in the field of background modeling and foreground detection, and has co-edited two books in CRC Press (background/foreground separation for video surveillance, robust PCA via decomposition in low rank and sparse matrices). His research investigated particularly the use of mathematical concepts (crisp, fuzzy, Dempter-Schäfer), subspace learning concepts (reconstructive, discriminative, robust PCA), neural networks concepts, (CNNs, GANs) and graph signal processing concepts in video surveillance. It also concern full exhaustive surveys on mathematical tools used in foreground/background separation. He has been the lead guest editor of the special issue on “Background Modeling for Foreground Detection in Real-World Dynamic Scenes” in the journal Machine Vision and Applications. He has been the invited talk in the International Workshop on Background Models Challenge at ACCV 2012. He has supervised five Ph.D. students in this field. He is the coordinator of the BGSlibrary and LRS library. He is the creator and the administrator of the Background Subtraction Web Site (33 115 visits and 17 636 visitors). He is a reviewer for prestigious international journals including IEEE (Trans. on Image Processing, Trans. on Multimedia, Trans. on CSVT, etc.), SPRINGER (IJCV, MVA, etc.) and ELSEVIER (CVIU, PR, PRL, etc.), and top-level conferences such as CVPR, ICPR, ICIP, AVSS, etc.
Research Interests
Professional Skills
Research Experience
Judge of The Works of Others
Membership
Prizes and Awards
43.091 (2024), 23.723 (2023), 16.835 (2022), 19.416 (2021),
14.548 (2020), 14.055 (2019), N/A (2018), 42.341 (2017)
Project
Global biodiversity and ecosystems face increasing threats from land use changes and global warming, leading to a pressing need for effective conservation efforts. A profound understanding of landscape heterogeneity, which refers to the variability of natural features, is essential for designing targeted conservation strategies.
Soundscape analysis, which involves studying the diverse sound sources within ecosystems, has emerged as a promising tool for monitoring and understanding these complex environments. However, traditional methods in soundscape analysis often rely on supervised learning models that are limited by their dependency on labeled data, potentially overlooking unknown ecological patterns.
An innovative approach introduced by recent research uses "sonotypes" to characterize the biophony of landscapes, providing a detailed acoustic profile. In response to the limitations of current methodologies, this research proposes an unsupervised graph-based network architecture. The project aims to incorporate the acoustic profile of the sites and temporal data and thereby yield highly interpretable results that capture the dynamic relationships and complexities of ecosystems. Thus, it will provide a comprehensive understanding of acoustic heterogeneity across geographical locations, ultimately helping in the design of more informed and effective conservation strategies.
Ecosystems in Colombia have unique flora and fauna components with high endemism levels. Due to deforestation the degradation is increasing. Then, it is necessary to apply conservation plans based on the biodiversity changes.
Passive Acoustic Monitoring (PAM) is considered as an alternative to overcome the disadvantages of traditional methods as the site’s perturbations, the cost, and study time. Most of PAM methods determine acoustic heterogeneity apply supervised algorithms in sites where the labels are metrics or qualities of study sites as transformations, habitat prototypes landcover types, and landscape types. However, the ecosystem health depends on transitional changes reflected through the species behavior or local communities changes. Sound provides information about ecosystems without predefined information about the distributions or label s of each spot. Whereby, it is necessary to work with unsupervised algorithms that find patterns between study sites.
Teaching Experience
Message
