Алгоритмічно-програмні аспекти граничної відеоаналітики: від теорії до практичного впровадження

• Дата: 2025
• Автор: Golovin, Oleksandr
• Формат: Стаття
• Мова: Ukrainian
• Опубліковано: V.M. Glushkov Institute of Cybernetics of NAS of Ukraine 2025
• Теми: граничні обчислення; відеоаналітика; комп’ютерний зір; розподілена інтелектуальність; федеративне навчання; оптимізація моделей; автономні системи
• Онлайн доступ: https://jais.net.ua/index.php/files/article/view/528
• Назва журналу: Problems of Control and Informatics

This article presents a comprehensive analysis of algorithmic and software transformations in video analytics under the influence of edge computing and establishes a methodological foundation for developing autonomous visual intelligence systems. The research investigates the evolutionary development of video analytical technologies from basic motion detection of the first generation to fourth-generation edge-optimized artificial intelligence, which achieved 100-fold computational acceleration while maintaining 95 % accuracy even after radical optimization. A systematic analysis of architectural changes is conducted — from centralized processing to distributed intelligence, including federated and hybrid approaches, enabling 95% reduction in network traffic and 70 % energy consumption savings. The study thoroughly examines algorithmic challenges of edge computing: quality degradation during model compression from 32- to 8-bit representation, temporal processing constraints with analytical window reduction from 10–30 to 1–3 seconds, complexity of distributed coordination of probabilistic computer vision results, and ensuring autonomy under edge device isolation conditions. Special attention is given to new algorithmic requirements for edge systems, including adaptive complexity with dynamic scaling from full-featured deep learning models to simplified detectors, continual learning without catastrophic forgetting, and robustness to visual anomalies through adaptive preprocessing techniques. The research results form a conceptual foundation for developing autonomous, energy-efficient video analytical systems of the new generation, capable of functioning in distributed environments with limited resources while maintaining high analysis quality and ensuring data privacy.