A flexible perception method of thin smoke based on patch total bounded variation for buildings.

Area of Science:

• Computer Vision
• Image Processing
• Fire Safety Engineering

Background:

• Early fire detection is crucial for power system stability.
• Current methods struggle with subtle features, delaying fire hazard alerts.

Purpose of the Study:

• To develop a novel algorithm for thin smoke detection to improve early fire detection capabilities.
• To address limitations of existing methods in capturing subtle image variations.

Main Methods:

• Proposed the Patch-TBV feature, computing total bounded variation (TBV) at the patch level.
• Incorporated subtle variation magnification using computed TBV values.
• Utilized a dataset of 3,120 images (TIP dataset) for real-world scenario evaluation.

Main Results:

• The algorithm effectively captures subtle image variations indicative of early fire signs.
• Subtle variation magnification enhances detection precision in low smoke concentrations.
• Experimental results confirm the algorithm's robustness and effectiveness in diverse conditions.

Conclusions:

• The proposed algorithm, integrating Patch-TBV and micro-variation amplification, overcomes limitations of existing methods.
• Demonstrates potential as a valuable tool for enhancing fire safety in power systems.
• Provides accurate and timely fire warnings through improved subtle feature detection.