CCTV systems use sophisticated cameras (network cameras, smart cameras) and computer servers for video recording in a fully digital system. They often integrate hundreds of cameras generating a huge amount of video data, far beyond human agent monitoring capabilities One of the most important and modern challenges, in this field, is to be able to scale an existing cloud-based video surveillance system with multiple heterogeneous smart cameras and adapt it to a Fog / Cloud architecture to improve performance without a significant cost overhead. Recently, FPGAs are becoming more and more present in FCIoT (FoG-Cloud-IoT) platform architectures. These components are characterized by dynamic and partial configuration modes, allowing platforms to quickly adapt themselves to changes resulting from an event, while increasing the available computing power. Today, such platforms present a certain number of serious scientific challenges, particularly in terms of deployment and positioning of FoGs. This thesis proposes a video surveillance model composed of plug & play smart cameras, equipped with dynamically reconfigurable FPGAs on a hierarchical FOG / CLOUD basis. In this highly dynamic and scalable system, both in terms of intelligent cameras (resources) and in terms of targets to track, we propose an automatic and optimized approach for camera authentication and their dynamic association with the FOG components of the system. The proposed approach also includes a methodology for an optimal allocation of hardware trackers to the electronic resources available in the system to maximize performance and minimize power consumption. All contributions have been validated with a real size prototype.
- Directeurs de thèse : MEFTALI Samy, AOUALI Djamel (co-encadrant) - Rapporteurs : ABOULHAMID El Mostapha , TARI Abdelkamel - Examinateurs : BEESTON Nathalie, VANIER Sonia, BOURENNANE El-Bay
Thesis of the team EAST defended on 20/04/2018