影像火災偵測一般具有比傳統火災偵測更多的優點,諸如反應快速、非接觸式偵測等,惟大多數的影像火災偵測卻有很高的誤判率。為了改善影像火災偵測的效能,本論文提出一個嶄新的基於火源與煙霧之機器視覺演算法。在即時火災偵測進行影像連續處理時,筆者利用運動歷史影像(MHI)移動物件偵測演算法,暫存影像中可能出現火源與煙霧的位置;然後再針對影像序列中火源與煙霧的光譜特徵以及時間特性進行分析。其中,權重式單一高斯模型與YCrCb模糊邏輯分類器用來偵測火源光譜特徵。飽和度累積偵測與貝氏分類器用來找出影像序列中的煙霧光譜特徵。運動梯度法則是用來量測煙霧運動特徵。最後由實驗結果顯示,以本論文所實現之演算法進行火災偵測則,比以往偵測方式具有更快的反應時間與更高的準確率。
Video fire detection has many advantages over traditional methods, such as fast response, non-contact, and so on. But most of video fire detection systems usually have high false alarms. In order to improve the performance of video fire detection, this paper proposes a new Machine Vision algorithm based on fire and smoke characteristics. The successive processing steps of our real-time fire detection are using the MHI (Motion History Image) motion object detection algorithm to register the possible fire and smoke position in a video; then analyze the spectral and temporal characteristics of fire and smoke in the image sequences. The weight single Gaussian model and YCrCb fuzzy logic classifier are used to detect spectral characteristics of fire. The saturation accumulative method and Bayesian classifier are to find out the spectral characteristics of smoke in the image sequences. The motion gradient method is used to measure the motion characteristics of smoke. Experimental results show that our algorithm has better accuracy and shorter response time for fire detection.