Architectural Optimization Of Computer Vision Surveillance

Résumé: This thesis presents a Robust Framework featuring two algorithms MPSTA and MARCPR, an innovative approach to optimizing automated human surveillance systems, which are often hindered by their inherent complexity and the rapid evolution of computer vision techniques. We introduce a human-behavior profiling system that addresses constraints architecturally, eliminating the need for synchronism. Our system outperforms continuous detection systems in terms of speed, it tackles each detection challenge using various computer vision techniques, thereby enhancing its efficiency and effectiveness. We propose a novel human behavior profiling approach, a system featuring a hybrid computer-vision kernel with dual execution times distributed architecture along with pattern extraction/employment, pre-processing techniques, and feature engineering. These methods have the potential to enhance profiling performance and accuracy in certain contexts, enabling the detection and analysis of both collective behaviors, such as interactions and conflicts among individuals, and distinct individual behaviors.

Publié dans la revue: