Skip to Main content Skip to Navigation
New interface

Optimization of video surveillance networks in smart cities using video compressive sensing and deep learning techniques

Abstract : Compressive Sensing, commonly used to approximate solutions for underdetermined linear systems of equations, is gaining a lot of attention as an efficient acquisition and compression paradigm that combines nonlinear reconstruction algorithms and random sampling on sparse basis. It enables to optimize the storage capacity of the wireless systems as well as the speed and cost of acquisition. Recently, Deep Learning architectures have frequently been exploited to optimize the reconstruction phase. The main objective of this thesis is to take advantage of Deep Learning architectures to optimize the compressive sensing technique by applying it on video signals and subsequently optimize the acquisition, transmission, and reconstruction of videos in modern digital systems. Therefore, the strategy adopted during this research work consists in establishing a comparative study on video compressive sensing (VCS) approaches based on Deep Learning by evaluating the quality and the speed of reconstruction as well as the different architectures. Then, two VCS environments have been designed: the first one is based on the prediction of video frames by implementing an approach based on a new recurrent network and the second one exploits the latest performances achieved with the Transformers and the attention mechanism. So, the approach adopted is based on a state-of- the-art analysis followed by an explanation of each architecture and an experimental validation. The different constraints encountered during this work are discussed and appropriate solutions are proposed.
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Monday, November 14, 2022 - 10:47:05 AM
Last modification on : Tuesday, November 15, 2022 - 3:14:25 AM


Version validated by the jury (STAR)


  • HAL Id : tel-03850837, version 1



Wael Saideni. Optimization of video surveillance networks in smart cities using video compressive sensing and deep learning techniques. Signal and Image processing. Université de Limoges, 2022. English. ⟨NNT : 2022LIMO0080⟩. ⟨tel-03850837⟩



Record views


Files downloads