本文提出了一個基於資訊熵的影像處理方法應用在火災偵測上,影像在經過資訊熵的處理後,觀察影片中每幀影像的變化,來判斷是否有火災以及火災發生的位置。 本文提出了兩種偵測火災的方法,方法一:全域熵,分別計算在每幀影像的不同顏色通道的熵值,將每一個通道的熵作圖比較,再觀察每一個通道在影片中的數據變化,即可判斷影片中是否有火災;方法二:局部熵,先將每一幀影像做灰階處理,再設一個特定範圍計算熵值,在影像中重複運算,直到做出一張完整的熵值分布圖,再對每一張熵值分布圖變化分析,即可找出火災發生的位置,最後我們會用我們的方法對火災影片跟非火災影片分析,比較兩者數據的差異。 本文的貢獻如下: 1. 有效率:可以即時偵測火災 2. 精準性:可以找出火災位置 3. 低成本:不需額外安裝設備
In the thesis, we propose a method to do fire detection based on information entropy. After the image processing, we can observe the variations frames by frames in the video to determine whether there is a fire and where the fire appears. In the thesis, we propose two methods to detect the fire. One is global entropy method which we need to calculate a value in different color channels and grayscale separately. Then, make a comparison with the data of these channels to determine whether there is a fire or not. The other is local entropy which we calculate the value of entropy in specific range repeatedly to make a complete entropy distribution map. And we analyze the variations in all of the distribution maps to find out where the fire appears. Finally, we compare the differences of analysis results between fire videos and fire-like videos. The contributions of our research are as follows: 1. Efficient: detecting the fire in time is possible 2. Accurate: detecting the position of the fire is feasible 3. Low-cost: installing another equipment is unnecessary