Integrating Deep Learning-Based IoT and Fog Computing with Software-Defined Networking for Detecting Weapons in Video...

Area of Science:

• Computer Science
• Artificial Intelligence
• Network Engineering

Background:

• The proliferation of multimedia traffic from Internet of Things (IoT) and remote applications necessitates improved Quality of Service (QoS).
• Existing network architectures struggle to adapt to the dynamic demands of delay-sensitive applications, particularly in real-time surveillance.
• The COVID-19 pandemic accelerated the need for efficient remote multimedia solutions, highlighting network limitations.

Purpose of the Study:

• To investigate the integration of deep learning (DL) techniques with Software-Defined Network (SDN) architecture.
• To enhance QoS for delay-sensitive applications within Internet of Things (IoT) environments.
• To develop an intelligent edge-based system for real-time weapon detection and efficient data transmission.

Main Methods:

• Trained and evaluated multiple DL models for weapon detection in real-time video surveillance.
• Deployed the highest-performing DL model within an AI model at the edge for frame extraction.
• Utilized SDN for dynamic network programming to provide differentiated QoS based on traffic type, load, and destination.
• Conducted performance evaluation using the mininet emulator.

Main Results:

• Achieved significant performance improvements in delay, throughput, and bandwidth requirements.
• Demonstrated up to 75.0% improvement in average throughput.
• Showcased up to 14.7% improvement in mean jitter and up to 32.5% reduction in packet loss.
• Successfully reduced bandwidth requirements by offloading non-critical traffic.

Conclusions:

• The integration of DL and SDN effectively supports delay-sensitive applications in IoT environments.
• The proposed edge AI model enables rapid threat detection and reduces network load.
• Dynamic network programming based on traffic characteristics optimizes QoS and network resource utilization.