Early fire detection method in video for vessels The Journal of Systems and Software, vol. 82, no. 4, pp. 656-667, 2009. 指導教授:李育強 教授 報告者:吳偉誠 2009/11/24 南台科技大學 資訊工程系 Outline 1. Introduction 2. The early fire detection approach Dangerous flame detection Dangerous smoke detection 3. Experimental results and analyses 4. Conclusions 2 1. Introduction A video-based early fire and smoke detection method for vessels. For flame detection, the dominant flame color lookup table (DFCLT) is constructed by using the fuzzy cmeans clustering algorithm. Elementary, medium, or emergency flame alarms are determined by comparing the pixels of the changed regions with the DFCLT. 3 2. The early fire detection approach 2.1 Dangerous flame detection 4 2. The early fire detection approach 1. The dominant flame color lookup table (DFCLT). 2. The changed video frames are automatically selected, and the changed regions are identified from these frames. 3. An elementary, medium, or emergency dangerous flame alarm is triggered by comparing the pixels of the changed regions with the DFCLT. 5 2. The early fire detection approach The DFCLT creation approach 6 2. The early fire detection approach 7 2. The early fire detection approach 8 2. The early fire detection approach Color transformation phase 1. RGB color space 2. HSV color space used only hue (H) and saturation (S) as the clustering attributes. 3. CIE LAB color space used only the A and B components as the clustering attributes. 9 2. The early fire detection approach Fuzzy color clustering phase 其中 X 為資料點,V 為中心點矩陣,n 為資料數, c 為中心點資料個數,w 為權重指數,設定為1.5。 10 2. The early fire detection approach Fuzzy color clustering phase 1. 設定分類個數 c,訂定初始權重矩陣,隨機給定 0~1 之 值,並滿足權重 總和為1。 2. 計算聚類中心點。 11 2. The early fire detection approach Fuzzy color clustering phase 3. 計算目標函數值,當目標函數值小於設定之容忍誤差可 結束迭代計算,否則執行步驟 4。 4. 重新計算權重矩陣 W ,並回到步驟 2 進行運算。 12 2. The early fire detection approach The advantage of using the DFCLTs creation approach is that another DFCLTs can be easily created when the environments changes, such as using different cameras or changing illumination. 13 2. The early fire detection approach Changed images are selected while the pixel number of 1 is greater than a threshold (25 pixels). 14 2. The early fire detection approach 2.2 Dangerous smoke detection 15 2. The early fire detection approach The changed images are selected when the pixel number 1 is greater than the threshold (30 pixels). 16 3. Experimental results and analyses 17 3. Experimental results and analyses 18 3. Experimental results and analyses 19 3. Experimental results and analyses 20 4. Conclusions The proposed method can be applied to other indoor environments. Combine the color, shape, texture and spatial relationship characteristics to improve performance. 21 南台科技大學 資訊工程系