以視訊為基礎之煙火偵測系統

Video-based Fire and Smoke Detection System

張吉弘(Chi-Hung Chung)

指導教授 ： 石聖文

國立暨南國際大學/科技學院/資訊工程學系/碩士(2014年)

https://doi.org/10.6837/NCNU.2014.00070

摘要

隨著電腦科技的進展，影像處理技術和辨識技術變得更加成熟，使得智能視頻監控系統應用也慢慢擴展當中。在這篇論文中，我們使用影像處理技術來分析火災所產生的火焰及煙霧特徵，透過分析這些特徵來發展出一套基於視頻的煙火偵測系統。我們的煙火偵測系統分為兩個部分，首先是火焰偵測，取得影像經過前景偵測找出感興趣的物件之後，對這些物件使用我們訓練好的火焰顏色查看表、LDA 分析還有G-channel 標準差分析和火焰風險評估來偵測影像當中是否有火焰發生。接著是煙霧偵測，取得影像先經過暗通道 (Dark Channel) 抽取出疑似煙霧的部分還有利用小波分析 (Wavelet analysis) 計算影像高頻能量是否減弱，最後從前景偵測中找出屬於煙霧的目標，並偵測它是否會隨著時間慢慢擴散，用以決定影像裡是否有煙霧的存在。實驗結果顯示，我們可以每秒約 100 張的速度完成煙與火的偵測，且正確率高達 92%。

關鍵字

火焰偵測 ； 煙霧偵測 ； 線性判別分析 ； 綠通道標準差分析 ； 火焰風險評估 ； 暗通道 ； 小波分析

並列摘要

Along with the progress of computer technology, sophisticated image processing/understanding methods have developed and the application of intelligent video surveillance system are becoming more and more popular. In this thesis, we use image processing techniques to analyze image features of flame and smoke. The image features are then used to develop a video-based fire and smoke detection system. The proposed system consists of the fire detection module and the smoke detection module. In the fire detection module, we first detect foreground objects with a proper background model. Then, three pre-trained fire color look up tables, an LDA model, the standard deviation of the G-channel, an evaluated flame risk value are used to detect flame in video. In the smoke detection module, we use dark channel analysis to extract suspicious blurry regions from video. Also, we use wavelet analysis to determine whether the high frequency image energy is reducing. Then, smoke candidate regions are computed and are tracked to examine if the area of any of them keeps growing. When the area of a smoke candidate is increasing, it is determined to be a smoke region. Experimental results show that, when the input video resolution is 640×480, the fire and smoke detection speed is 100 frames/sec., and the recognition accuracy is about 92%.

並列關鍵字

Fire Detection ； Smoke Detection ； LDA ； G-channel Standard Deviation Analysis ； Flame Risk Estimation ； Dark Channel ； Wavelet Analysis

參考文獻

[1] T. Celik, H. Ozkaramanli, and H. Demirel, “Fire and smoke detection without sensors: Image processing-based approach,” in Proceedings of the 15th European Signal Processing Conference, pp. 147 – 158, 2007.

[2] B. U. Toreyin, Y. Dedeoˇ glu, and A. E. Cetin, “Wavelet based real-time smoke detection in video,” in Proceedings of the 13-th European Signal Processing Conference, pp. 4 – 8, 2005.

[3] S. Calderara, P. Piccinini, and R. Cucchiara, “Smoke detection in video surveillance: A mog model in the wavelet domain,” in Proceedings of the Computer Vision Systems, pp. 119 – 128, 2008.

[4] S. Verstockt, “Multi-modal video analysis for early fire detection,” in Ph.D. Thesis,Ghent University, 2011.

[5] K. He, J. Sun, and X. Tang, “Single image haze removal using dark channel prior,” in Proceedings of IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 33, pp. 2341 – 2353, 2011.