本論文主要是研究一些即時火焰辨識演算法並提出改善方法,根據國科會計畫「建構網際網路協定為基礎的數位監視系統的Java自由軟體平台」[1],在其架構下增加火災偵測功能,利用本論文提出的演算法建置出火災偵測及查詢系統 (Fire Detection and Query System,FDQS) [2]。本論文之演算法結合了火焰燃燒時亮度變化的不規則特性及適用於一般網路攝影機所拍攝畫面的火焰顏色模型,能簡單、快速地擷取出火焰區域,並且藉由火焰顏色分佈隨時間變化的特性來大幅減低像火顏色的動態物體所造成的錯誤警報。最後設計一個判別火災的警報流程使此演算法更有效地應用在數位監視系統上,進而達到有效率的即時火災偵測、警報以及火災事件記錄與查詢等目的。
In this paper, we study real-time flame recognition algorithms and propose an improved method. The proposed fire detection algorithm is applied to extend the National Science Council project - The Design and Implementation of Java Open Source Platform for IP-based Video Surveillance System (IVSS) [1], to develop a Fire Detection and Query System (FDQS) [2]. The algorithm not only combines the irregular property of intensity variation in burning with the fire color model which can adapt to generic web camera, but also well reduces the false alarm caused by the moving object with fire-like color according to the characteristics of time-varying fire color distribution. Thus we can extract the fire region efficiently. Moreover, we design a fire alarm mechanism for the flame recognition algorithm such that it can be applied on digital surveillance system. In summary, we have attained the goal of detecting fire, fire alarm, event recording and querying in real-time.