在極端火災環境中，強烈的火光往往會導致局部過曝、光暈以及高亮區的飽和現象。同時，濃煙和不均勻的光照也會使暗部細節嚴重流失，這大大降低了救災人員以及智慧監控系統的辨識能力。傳統影像增強方法（如 Contrast-Limited Adaptive Histogram Equalization, CLAHE）雖可提升局部對比度，但容易放大噪聲並產生區塊化偽影及邊緣失真；而深度學習方法雖具有一定的影像增強能力，但受其模型規模與運算資源的限制，較不適合用於即時嵌入式應用。
針對上述問題，本研究提出一套高效能之即時火場影像清晰化演算法，整合火焰遮罩、自適應局部對比增強、極端亮暗區保護與運算最佳化，旨在兼顧影像品質與即時處理效能。首先，利用結合 HSV 特徵與時域平滑濾波的火焰辨識機制，穩定擷取動態火焰區域並降低背景干擾。接著，設計結合平滑步階函數與動態裁切限制之雙向 CLAHE 模組，以改善影像中光照不均與背光問題，並抑制過度增強所造成的偽影。最後，透過三重極端值防護機制動態調整火源、高光與暗部權重，並以 Alpha 融合重建 RGB 影像，以提升暗部可視性並保留火焰邊界平滑性。
為滿足即時處理與硬體化需求，本研究進一步使用定點運算、位移運算（Bit-shift）及查表法（LUT）取代部分高成本浮點和非線性運算，並結合OpenMP多執行緒平行化與非同步佇列設計，以降低CPU負擔並提升整體效率，為未來硬體化實作奠定基礎。實驗結果顯示，本方法可有效提升火場影像的資訊熵與平均梯度，並在1280 × 720解析度下維持即時處理能力，在影像品質與運算效率之間取得良好平衡，適用於火場救援、智慧監控及邊緣運算等應用場景。

This thesis introduces a high-performance real-time image enhancement algorithm specifically designed for extreme fire scenes, which are often marked by a high dynamic range, intense light sources, and dense smoke occlusion. To overcome the computational constraints of edge computing devices in disaster relief, the proposed framework offers a hardware-friendly digital image processing pipeline that sidesteps expensive floating-point operations.
Initially, the algorithm incorporates a spatiotemporal flame masking strategy based on HSV and luminance features, along with infinite impulse response filtering, to accurately isolate burning regions and avoid overexposure in later stages. For the obscured background, the system employs adaptive shadow compensation via a pre-computed Smoothstep function. This step is succeeded by a dynamic bidirectional Contrast-Limited Adaptive Histogram Equalization (CLAHE) module, which uses mask-aware histograms and dynamic clip limits to boost local contrast while effectively reducing noise amplification in extremely dark areas. To complete the process, an RGB-domain extreme-value protection method merges the enhanced background with the original fire source using alpha-weight blending. This ensures the natural color gradients and structural fidelity of the flames are maintained without color distortion.
On the software side, the architecture is optimized for low latency and high throughput. By replacing intricate mathematical divisions with fixed-point arithmetic, bit-shift operations, and Look-Up Tables (LUTs), the algorithm significantly lightens the CPU's computational load. Additionally, the system enhances multi-core processing capabilities by incorporating OpenMP for data-level parallelism and an asynchronous I/O double-buffering pipeline. These software enhancements not only enable real-time processing on standard CPUs but also set a strong base for future adaptation to Register-Transfer Level (RTL) circuits.

[1] GoPro, "Firefighter's Raw POV," YouTube, Nov. 08, 2024. [Online]. Available: https://www.youtube.com/watch?v=mphHFk5IXsQ
[2] Celik, T. (2010). Fast and efficient method for fire detection using image processing. ETRI Journal, 32(6), pp. 881–890. https://doi.org/10.4218/etrij.10.0109.0695.
[3] Pizer, S. M., Amburn, E. P., Austin, J. D., Cromartie, R., Geselowitz, A., Greer, T., ter Haar Romeny, B., Zimmerman, J. B., & Zuiderveld, K. (1987). Adaptive histogram equalization and its variations. Computer Vision, Graphics, and Image Processing, 39(3), pp. 355–368. https://doi.org/10.1016/S0734-189X(87)80186-X.
[4] Reza, A. M. (2004). Realization of the contrast limited adaptive histogram equalization (CLAHE) for real-time image enhancement. Journal of VLSI Signal Processing Systems for Signal, Image, and Video Technology, 38, pp. 35–44. https://doi.org/10.1023/B:VLSI.0000028532.53893.82
[5] Muhammad, K., Ahmad, J., Mehmood, I., Rho, S., & Baik, S. W. (2018). Convolutional neural networks based fire detection in surveillance videos. IEEE Access, 6, pp. 18174–18183. https://doi.org/10.1109/ACCESS.2018.2812835
[6] Jiao, Z., Zhang, Y., Xin, J., Mu, L., Yi, Y., Liu, H., & Liu, D. (2019). A deep learning based forest fire detection approach using UAV and YOLOv3. In 2019 1st International Conference on Industrial Artificial Intelligence (IAI)，pp. 1–5. IEEE. https://doi.org/10.1109/ICIAI.2019.8850815.
[7] Chen, J., & Ran, X. (2019). Deep learning with edge computing: A review. Proceedings of the IEEE, 107(8), pp. 1655–1674. https://doi.org/10.1109/JPROC.2019.2921977.
[8] Gonzalez, R. C., & Woods, R. E. (2018). Digital image processing (4th ed.). Pearson.
[9] International Telecommunication Union. (2011). Studio encoding parameters of digital television for standard 4:3 and wide-screen 16:9 aspect ratios (ITU-R Recommendation BT.601-7). https://www.itu.int/rec/R-REC-BT.601-7-201103-I/en.
[10] Horng, W.-B., Peng, J.-W., & Chen, C.-Y. (2005). A new image-based real-time flame detection method using color analysis. In Proceedings of the 2005 IEEE International Conference on Networking, Sensing and Control，pp. 100–105. IEEE. https://doi.org/10.1109/ICNSC.2005.1461169
[11] Munshi, A. (2021). Fire detection methods based on various color spaces and Gaussian mixture models. Advances in Science and Technology Research Journal, 15(3), pp. 197–214. https://doi.org/10.12913/22998624/138924
[12] Fire Cam Police and Fire Cameras, "2 Deck Gun House Fire with Interior Attack," YouTube, Oct. 08, 2015. [Online]. Available: https://www.youtube.com/watch?v=fGeMWg85KFU
[13] Fire Cam Police and Fire Cameras, "Fire Helmet Camera Structure Fire 2016," YouTube, Jun. 15, 2016. [Online]. Available: https://www.youtube.com/watch?v=B1bToqWjPgw
[14] Fire Cam Police and Fire Cameras, "House Fire Quick Interior Attack," YouTube, Nov. 09, 2012. [Online]. Available: https://www.youtube.com/watch?v=Fwj2UJhqDW8
[15] WAVY TV 10, "Helmet camera video of Smithfield house fire," YouTube, Jul. 07, 2015. [Online]. Available: https://www.youtube.com/watch?v=Cd43nNGvh00
[16] 1News, "Seven Sharp burns down house live on TV with Fire and Emergency NZ," YouTube, Jun. 23, 2021. [Online]. Available: https://www.youtube.com/watch?v=aCC1-4FKAo8
[17] Hennessy, J. L., & Patterson, D. A. (2017). Computer architecture: A quantitative approach (6th ed.). Morgan Kaufmann.
[18] Wolf, M. (2012). Computers as components: Principles of embedded computing system design (3rd ed.). Elsevier.
[19] Bradski, G., & Kaehler, A. (2008). Learning OpenCV: Computer vision with the OpenCV library. O’Reilly Media
[20] Silberschatz, A., Galvin, P. B., & Gagne, G. (2018). Operating system concepts (10th ed.). Wiley.
[21] Fire Cam Police and Fire Cameras, "House Fire Fire Helmet Camera," YouTube, Jun. 25, 2018. [Online]. Available: https://www.youtube.com/watch?v=y2Z2ybyA_Ks
[22] Shannon, C. E. (1948). A mathematical theory of communication. Bell System Technical Journal, 27(3), pp. 379–423. https://doi.org/10.1002/j.1538-7305.1948.tb01338.x
[23] Liu, C.-L., Wang, Y.-R., & Liu, W.-Y. (2025). Multi-index remote sensing for post-fire damage assessment: Accuracy, carbon loss, and conservation implications. Frontiers in Forests and Global Change, 8, pp. 1–11. https://doi.org/10.3389/ffgc.2025.1577612