Fire Detection Algorithms Using Multimodal ... - Bilkent University
Fire Detection Algorithms Using Multimodal ... - Bilkent University Fire Detection Algorithms Using Multimodal ... - Bilkent University
CHAPTER 3. FLAME DETECTION IN INFRA-RED (IR) VIDEO 45Similar to the discussions in Chapter 2, flame flicker process in IR video ischaracterized by a random Markov model. As in the previous chapter, threestateMarkov models are trained off-line for both flame and non-flame pixels torepresent the temporal behavior (cf. Fig. 4.3). These models are trained using afeature signal which is defined as follows: Let I k (n) be the intensity value of thek − th pixel at frame n. The wavelet coefficients of I k are obtained by the samestructure shown in Fig. 4.2, but filtering is implemented temporally.Figure 3.7: Three-state Markov models for a) flame and b) non-flame movingpixels.The states of HMMs are defined as follows: at time n, if |w(n)| < T 1 , thestate is in S1; if T 1 < |w(n)| < T 2 , the state is S2; else if |w(n)| > T 2 , the stateS3 is attained. For the pixels of regular hot objects like walking people, engineof a moving car, etc., no rapid changes take place in the pixel values. When thewavelet coefficients fluctuate between values above the higher threshold T 2 andbelow the lower threshold T 1 in a frequent manner this indicates the existence offlames in the viewing range of the IR camera.The transition probabilities between states for a pixel are estimated during apre-determined period of time around flame boundaries. During the recognitionphase, the HMM based analysis is carried out in pixels near the contour boundariesof bright moving regions whose ρ values exceed ρ T . The state sequence of
CHAPTER 3. FLAME DETECTION IN INFRA-RED (IR) VIDEO 46length 20 image frames is determined for these candidate pixels and fed to theflame and non-flame pixel models. The model yielding higher probability is determinedas the result of the analysis for each of the candidate pixel. A pixel iscalled as a flame or a non-flame pixel according to the result of this analysis. Afire mask composing of flame pixels is formed as the output of the method.3.3 Experimental ResultsThe proposed method was implemented in a personal computer with an AMDAthlonXP 2000+ 1.66GHz processor. The HMMs used in the temporal analysisstep were trained using outdoor IR video clips with fire and ordinary movingbright objects like people and cars. Video clips have 236577 image frames with160 by 120 pixel resolution. All of the clips are captured at 10 fps using a cooledlong-wave IR camera with a spectral range of 8-12 µm. This camera has a longerrange than 30 meters up to which distance the proposed algorithm is developedfor. We cannot detect a starting fire in longer distances using the proposedalgorithm, because flame flicker cannot be observed in long distances.There are moving cars and walking people in most of the test video clips.Image frames from some of the clips are shown in Figs. 3.8 and 3.9. To increasethe number of videos in experimental studies, black and white video clips are alsoused.We used some of our clips for training the Markov models. The fire model wastrained with fire videos and the other model was trained with ordinary movingbright objects. The remaining 48 video clips were used for test purposes. Ourmethod yields no false positives in any of the IR test clips.A modified version of a recent method by Guillemant and Vicente [36] forreal-time identification of smoke in black and white video is implemented forcomparison. This method is developed for forest fire detection from watch towers.In a forest fire, smoke rises first, therefore, the method was tuned for smokedetection. Guillemant and Vicente based their method on the observation that
- Page 14 and 15: LIST OF FIGURESxiv2.6 Three-state M
- Page 16 and 17: LIST OF FIGURESxvi4.4 Two moving ob
- Page 18 and 19: List of Tables1.1 Organization of t
- Page 20 and 21: Chapter 1IntroductionDynamic textur
- Page 22: CHAPTER 1. INTRODUCTION 3No fire de
- Page 27 and 28: Chapter 2Flame Detection in Visible
- Page 29 and 30: CHAPTER 2. FLAME DETECTION IN VISIB
- Page 31 and 32: CHAPTER 2. FLAME DETECTION IN VISIB
- Page 33 and 34: CHAPTER 2. FLAME DETECTION IN VISIB
- Page 35 and 36: CHAPTER 2. FLAME DETECTION IN VISIB
- Page 37 and 38: CHAPTER 2. FLAME DETECTION IN VISIB
- Page 39 and 40: CHAPTER 2. FLAME DETECTION IN VISIB
- Page 41 and 42: CHAPTER 2. FLAME DETECTION IN VISIB
- Page 43 and 44: CHAPTER 2. FLAME DETECTION IN VISIB
- Page 45 and 46: CHAPTER 2. FLAME DETECTION IN VISIB
- Page 47 and 48: CHAPTER 2. FLAME DETECTION IN VISIB
- Page 49 and 50: CHAPTER 2. FLAME DETECTION IN VISIB
- Page 51 and 52: CHAPTER 2. FLAME DETECTION IN VISIB
- Page 53 and 54: CHAPTER 2. FLAME DETECTION IN VISIB
- Page 55 and 56: CHAPTER 3. FLAME DETECTION IN INFRA
- Page 57 and 58: CHAPTER 3. FLAME DETECTION IN INFRA
- Page 59 and 60: CHAPTER 3. FLAME DETECTION IN INFRA
- Page 61 and 62: CHAPTER 3. FLAME DETECTION IN INFRA
- Page 63: CHAPTER 3. FLAME DETECTION IN INFRA
- Page 67 and 68: CHAPTER 3. FLAME DETECTION IN INFRA
- Page 69 and 70: CHAPTER 3. FLAME DETECTION IN INFRA
- Page 71 and 72: CHAPTER 3. FLAME DETECTION IN INFRA
- Page 73 and 74: Chapter 4Short Range Smoke Detectio
- Page 75 and 76: CHAPTER 4. SHORT RANGE SMOKE DETECT
- Page 77 and 78: CHAPTER 4. SHORT RANGE SMOKE DETECT
- Page 79 and 80: CHAPTER 4. SHORT RANGE SMOKE DETECT
- Page 81 and 82: CHAPTER 4. SHORT RANGE SMOKE DETECT
- Page 83 and 84: CHAPTER 4. SHORT RANGE SMOKE DETECT
- Page 85 and 86: CHAPTER 4. SHORT RANGE SMOKE DETECT
- Page 87 and 88: Chapter 5Flame Detection Using PIRS
- Page 89 and 90: CHAPTER 5. FLAME DETECTION USING PI
- Page 91 and 92: CHAPTER 5. FLAME DETECTION USING PI
- Page 93 and 94: CHAPTER 5. FLAME DETECTION USING PI
- Page 95 and 96: CHAPTER 5. FLAME DETECTION USING PI
- Page 97 and 98: CHAPTER 5. FLAME DETECTION USING PI
- Page 99 and 100: CHAPTER 5. FLAME DETECTION USING PI
- Page 101 and 102: CHAPTER 6. WILDFIRE DETECTION 82In
- Page 103 and 104: CHAPTER 6. WILDFIRE DETECTION 84are
- Page 105 and 106: CHAPTER 6. WILDFIRE DETECTION 86D 1
- Page 107 and 108: CHAPTER 6. WILDFIRE DETECTION 88tha
- Page 109 and 110: CHAPTER 6. WILDFIRE DETECTION 90The
- Page 111 and 112: CHAPTER 6. WILDFIRE DETECTION 92D 1
- Page 113 and 114: CHAPTER 6. WILDFIRE DETECTION 94as
CHAPTER 3. FLAME DETECTION IN INFRA-RED (IR) VIDEO 46length 20 image frames is determined for these candidate pixels and fed to theflame and non-flame pixel models. The model yielding higher probability is determinedas the result of the analysis for each of the candidate pixel. A pixel iscalled as a flame or a non-flame pixel according to the result of this analysis. Afire mask composing of flame pixels is formed as the output of the method.3.3 Experimental ResultsThe proposed method was implemented in a personal computer with an AMDAthlonXP 2000+ 1.66GHz processor. The HMMs used in the temporal analysisstep were trained using outdoor IR video clips with fire and ordinary movingbright objects like people and cars. Video clips have 236577 image frames with160 by 120 pixel resolution. All of the clips are captured at 10 fps using a cooledlong-wave IR camera with a spectral range of 8-12 µm. This camera has a longerrange than 30 meters up to which distance the proposed algorithm is developedfor. We cannot detect a starting fire in longer distances using the proposedalgorithm, because flame flicker cannot be observed in long distances.There are moving cars and walking people in most of the test video clips.Image frames from some of the clips are shown in Figs. 3.8 and 3.9. To increasethe number of videos in experimental studies, black and white video clips are alsoused.We used some of our clips for training the Markov models. The fire model wastrained with fire videos and the other model was trained with ordinary movingbright objects. The remaining 48 video clips were used for test purposes. Ourmethod yields no false positives in any of the IR test clips.A modified version of a recent method by Guillemant and Vicente [36] forreal-time identification of smoke in black and white video is implemented forcomparison. This method is developed for forest fire detection from watch towers.In a forest fire, smoke rises first, therefore, the method was tuned for smokedetection. Guillemant and Vicente based their method on the observation that