An Early Flame Detection System Based on Image Block Threshold Selection Using Knowledge of Local and Global Feature Analysis

Ting Wei Hsu, S. Pare, Mahendra Singh Meena, Deepak Kumar Jain, Dong Lin Li, Amit Saxena, Mukesh Prasad, Chin‐Teng LinSustainability 2020 · DOI

TL;DR

This paper introduces a novel, computationally efficient fire detection system that analyzes video frames by dividing them into blocks. By combining local and global feature analysis, including color, motion, texture, and flickering, the system aims to achieve high detection rates and low false alarm rates in diverse real-world conditions, outperforming existing methods.

What Problem Does This Paper Address?

Early fire detection is crucial for minimizing damage to property and forests. While image-based detection systems offer a promising avenue, many existing methods struggle with performance in varied environmental conditions, leading to high false alarm rates or missed detections. Traditional methods often rely on specific techniques that perform well in limited scenarios but lack generalizability. This paper addresses the need for a robust, general-purpose fire detection system that can operate effectively in realistic, dynamic environments, overcoming limitations of previous approaches by integrating a comprehensive set of local and global visual features.

Methods

The proposed system employs a block-based analysis approach. Each video frame is divided into non-overlapping blocks, which are then processed through several stages:

  • Pre-processing: This involves background subtraction using a Gaussian Mixture Model (GMM) to identify foreground objects and then detecting movement within these foreground regions using temporal differencing. Blocks exhibiting significant changes are flagged as "candidate blocks."
  • Local Feature Analysis: Candidate blocks are analyzed for several local features:
  • Flame Color Analysis: Utilizes the YCbCr color space to identify pixels with characteristics similar to fire (yellow, red). It accounts for different intensity levels of fire.
  • Flame Source Analysis: Identifies the fixed position and high-intensity nature of a fire's origin, distinguishing it from transient light sources.
  • Flame Flickering Analysis: Detects the characteristic rapid changes in intensity and color associated with flames.
  • Flame Texture Analysis: Differentiates fire from objects with similar colors but smooth textures by analyzing color variance.
  • Global Feature Analysis: Further refines detection by considering features across multiple blocks, including:
  • Flame Area Analysis: Analyzes the rate of change in the fire's area, distinguishing it from flashing lights which exhibit rapid, instantaneous changes. This involves extending candidate regions and using wavelet analysis to capture high-frequency signals.
  • Alarm Decision Unit (ADU): A buffering mechanism is used to collect detection intervals. An alarm is triggered only when detected intervals exceed a certain percentage of the buffer size, helping to filter out transient false alarms caused by camera noise or sudden brightness changes.

Datasets

The paper mentions using "six non-fire videos and ten fire videos" for experimentation. For testing the algorithm and accuracy discussions, a dataset of "13,257 positive samples as fire events and 62,614 ordinary moving objects as negative samples" was used. Specific named datasets like "Foggia's video dataset" and "Chino's dataset" are mentioned in the literature review but not explicitly stated as used in this paper's experiments.

Key Insights and Ideas

  • Block-based analysis is key: Dividing frames into blocks allows for localized feature extraction and efficient processing, making the system computationally lighter.
  • Multi-feature fusion is superior: Relying on a single feature (like color) is insufficient. Combining local features (color, source, flickering, texture) with global analysis (area change) significantly improves robustness and reduces false positives.
  • Distinguishing fire from similar phenomena: The system explicitly addresses common false alarm sources like flashing lights, street lights, and objects with similar colors, by analyzing their distinct temporal and spatial characteristics.
  • Adaptive color analysis for intensity: The YCbCr color model is adapted to handle different fire intensities, including high-intensity fires that might appear closer to white.
  • Temporal buffering for stability: The Alarm Decision Unit acts as a crucial filter, preventing immediate alarms from transient events and ensuring more reliable detection.

Key Results

  • The proposed system achieved a detection rate of 89.8% and a false alarm rate of 0.8% when using the ADU.
  • Compared to existing methods by Chen et al. and Celik et al., the proposed system detected 10 fire videos correctly with only 1 false alarm, while the others detected 9 videos with 3 and 4 false alarms, respectively.
  • The system demonstrated good performance across various environments, including indoor, outdoor, and low-light conditions.
  • The reaction time for the proposed system with ADU was 3.45 seconds, which is longer than some other methods but attributed to the buffering mechanism for improved accuracy.

Limitations and Future Work

  • Reaction Time: The buffering mechanism for the ADU, while improving accuracy, leads to a longer reaction time compared to some other methods.
  • Minor Shaking and Flame Colors: The authors suggest future work could focus on further improving the system by eliminating false alarms caused by minor camera shaking due to wind and by better handling different flame colors from various fuels.
  • Smoke Detection: The current system focuses on flame detection. The paper notes that smoke detection feature extraction could be a valuable addition for early detection, especially in scenarios like forest fires where flames might not be immediately visible.
  • Smoldering Fire and Volcanic Eruptions: The system could potentially be adapted for detecting smoldering fires or even volcanic eruptions.
  • Embedded Systems: The low computational complexity makes the system suitable for implementation on embedded systems for real-time surveillance.

Summary updated

Abstract

Fire is one of the mutable hazards that damage properties and destroy forests. Many researchers are involved in early warning systems, which considerably minimize the consequences of fire damage. However, many existing image-based fire detection systems can perform well in a particular field. A general framework is proposed in this paper which works on realistic conditions. This approach filters out image blocks based on thresholds of different temporal and spatial features, starting with dividing the image into blocks and extraction of flames blocks from image foreground and background, and candidates blocks are analyzed to identify local features of color, source immobility, and flame flickering. Each local feature filter resolves different false-positive fire cases. Filtered blocks are further analyzed by global analysis to extract flame texture and flame reflection in surrounding blocks. Sequences of successful detections are buffered by a decision alarm system to reduce errors due to external camera influences. Research algorithms have low computation time. Through a sequence of experiments, the result is consistent with the empirical evidence and shows that the detection rate of the proposed system exceeds previous studies and reduces false alarm rates under various environments.
BibTeX
@inproceedings{hsu2020early,
  author = {Ting Wei Hsu and S. Pare and Mahendra Singh Meena and Deepak Kumar Jain and Dong Lin Li and Amit Saxena and Mukesh Prasad and Chin‐Teng Lin},
  title = {{An Early Flame Detection System Based on Image Block Threshold Selection Using Knowledge of Local and Global Feature Analysis}},
  booktitle = {Sustainability},
  year = {2020},
  doi = {10.3390/su12218899},
}
Figure