Novel Video Surveillance-Based Fire and Smoke Classification Using Attentional Feature Map in Capsule Networks

Area of Science:

• Computer Vision
• Artificial Intelligence
• Surveillance Technology

Background:

• Early fire and smoke detection remain significant challenges, impacting public safety and property.
• Existing surveillance models struggle with varying light, camera angles, and environmental changes.
• CCTV systems are widespread but require enhanced analytical capabilities for reliable fire detection.

Purpose of the Study:

• To develop an effective and affordable visual detection system for early fire and smoke identification.
• To improve the accuracy of fire and smoke classification from outdoor CCTV images.
• To address limitations in capsule network inputs and large image analysis for fire detection.

Main Methods:

• An attention feature map integrated into a capsule network was developed for classifying fire and smoke.
• The model processes single images of fire and smoke for outdoor distance classification.
• The approach compensates for the absence of deep networks using attention mechanisms.

Main Results:

• The proposed model achieved high classification accuracy compared to modern architectures.
• The attention-based capsule network demonstrated robustness and stability across diverse viewpoints.
• The system effectively classifies fire and smoke from outdoor CCTV camera images.

Conclusions:

• The developed attention-based capsule network offers a practical and accurate solution for fire and smoke detection.
• This method enhances the utility of existing CCTV infrastructure for public safety.
• The approach shows significant promise for real-world applications in early fire event recognition.