Fire Detection Algorithms Using Multimodal ... - Bilkent University
Fire Detection Algorithms Using Multimodal ... - Bilkent University Fire Detection Algorithms Using Multimodal ... - Bilkent University
CHAPTER 6. WILDFIRE DETECTION 101Weighted Majority(x,n)for i = 1 to M dow i (0) = 1 , InitializationMend forif ∑ i:d i (x,n)=1 w i(n) ≥ ∑ i:d i (x,n)=−1 w i(n) thenreturn 1elsereturn -1end iffor i = 1 to M doif d i (x, n) ≠ y thenw i (n + 1) ← w i(n)2end ifend forFigure 6.7: The pseudo-code for the Weighted Majority Algorithmbinary, i.e., d i (x, n) ∈ {−1, 1}, which are simply the quantized version of real valuedD i (x, n) defined in Section 6.2. In the WMA, the weights of sub-algorithmsyielding contradictory decisions with that of the oracle are reduced by a factor oftwo in an un-controlled manner, unlike the proposed LMS based algorithm andthe ULP scheme. Initial weights for WMA are taken as 1 , as in the proposedMLMS based scheme.The LMS based scheme, the ULP based scheme, the WMA based scheme, andthe non-adaptive approach with fixed weights are compared with each other inthe following experiments. In Tables 6.1 and 6.2, 6-hour-long forest surveillancerecordings containing actual forest fires and test fires as well as video sequenceswith no fires are used.We have 7 actual forest fire videos and 5 test fire videos ranging from 2 kmto 8 km captured in Antalya and Mugla regions in Turkey, in the summers of2007 and 2008. All of the above mentioned decision fusion methods detect forestfires within 8 seconds, as shown in Table 6.1. The detection rates of the methodsare comparable to each other. On the other hand, the proposed adaptive fusionstrategy significantly reduces the false alarm rate of the system by integrating thefeedback from the guard (oracle) into the decision mechanism within the activelearning framework described in Section 6.3.In Fig. 6.8 a typical false alarm
CHAPTER 6. WILDFIRE DETECTION 102issued for shadows of clouds by an untrained algorithm with decision weightsequal to 1 is presented as a sequence of snapshots taken at seconds 1, 7, 13 and418 from the clip V 15. The proposed algorithm does not produce a false alarm inthis video.When a false alarm is issued by the compound algorithm, the learning processis much faster for the LMS based scheme in comparison to the ULP and WMAbased methods. This is reflected in Fig. 6.9, in which the average squared errorfunctions ĒLMS(n), ĒULP (n) and ĒW MA(n) at image frame n, for the LMS, ULPand the WMA based methods are shown, respectively. The average squared errorvalues are computed over all x in the image frame n, and defined as:Ē(n) = 1 ∑(e(x, n)) 2 (6.40)N Ixwhere N I is the total number of pixels in the image frame, e(x, n) = y(x, n) − ŷ(x, n)and ŷ(x, n) = ∑ iw i D i (x, n). The average squared error values ĒLMS(n),Ē ULP (n) and ĒW MA(n) are computed after the first false alarm is issued. Theaverage squared error value ĒLMS(n) corresponding to the LMS based schemedecays fast reaching around 0 within 5 frames of video which is about 1 sec. Ittakes about 3 sec and 6 sec for the average squared error values ĒULP (n) andĒ W MA (n) corresponding to the ULP and the WMA based schemes to converge.The proposed LMS based method produces the lowest number of false alarmsin our data set. A set of video clips containing moving cloud shadows is used togenerate Table 6.2. These video clips are especially selected. Number of imageframes in which false alarms are issued by different methods are presented inTable 6.2. Total number of false alarms for the clips in Table 6.2 issued by themethods (a) the LMS based scheme, (b) the ULP based scheme, (c) the WMAbased scheme and (d) the non-adaptive approach with fixed weights are 5, 93,240 and 612, respectively.The software is currently being used in 10 forest watch towers in Antalya andMugla regions. We tested the system live and monitored the number of falsealarms for two days in September 2008. The current system produces 0.25 falsealarms per hour. This is an acceptable rate for a look-out tower.
- 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
- Page 115 and 116: CHAPTER 6. WILDFIRE DETECTION 96Sol
- Page 117 and 118: CHAPTER 6. WILDFIRE DETECTION 98as
- Page 119: CHAPTER 6. WILDFIRE DETECTION 100Un
- Page 123 and 124: CHAPTER 6. WILDFIRE DETECTION 104Ta
- Page 125 and 126: Chapter 7Conclusion and Future Work
- Page 127 and 128: CHAPTER 7. CONCLUSION AND FUTURE WO
- Page 129 and 130: Bibliography[1] B. Albers and A. Ag
- Page 131 and 132: BIBLIOGRAPHY 112[18] D. Chetverikov
- Page 133 and 134: BIBLIOGRAPHY 114[38] G. Healey, D.
- Page 135 and 136: BIBLIOGRAPHY 116[56] U. Niesen, D.
- Page 137 and 138: BIBLIOGRAPHY 118[77] C. Stauffer an
CHAPTER 6. WILDFIRE DETECTION 102issued for shadows of clouds by an untrained algorithm with decision weightsequal to 1 is presented as a sequence of snapshots taken at seconds 1, 7, 13 and418 from the clip V 15. The proposed algorithm does not produce a false alarm inthis video.When a false alarm is issued by the compound algorithm, the learning processis much faster for the LMS based scheme in comparison to the ULP and WMAbased methods. This is reflected in Fig. 6.9, in which the average squared errorfunctions ĒLMS(n), ĒULP (n) and ĒW MA(n) at image frame n, for the LMS, ULPand the WMA based methods are shown, respectively. The average squared errorvalues are computed over all x in the image frame n, and defined as:Ē(n) = 1 ∑(e(x, n)) 2 (6.40)N Ixwhere N I is the total number of pixels in the image frame, e(x, n) = y(x, n) − ŷ(x, n)and ŷ(x, n) = ∑ iw i D i (x, n). The average squared error values ĒLMS(n),Ē ULP (n) and ĒW MA(n) are computed after the first false alarm is issued. Theaverage squared error value ĒLMS(n) corresponding to the LMS based schemedecays fast reaching around 0 within 5 frames of video which is about 1 sec. Ittakes about 3 sec and 6 sec for the average squared error values ĒULP (n) andĒ W MA (n) corresponding to the ULP and the WMA based schemes to converge.The proposed LMS based method produces the lowest number of false alarmsin our data set. A set of video clips containing moving cloud shadows is used togenerate Table 6.2. These video clips are especially selected. Number of imageframes in which false alarms are issued by different methods are presented inTable 6.2. Total number of false alarms for the clips in Table 6.2 issued by themethods (a) the LMS based scheme, (b) the ULP based scheme, (c) the WMAbased scheme and (d) the non-adaptive approach with fixed weights are 5, 93,240 and 612, respectively.The software is currently being used in 10 forest watch towers in Antalya andMugla regions. We tested the system live and monitored the number of falsealarms for two days in September 2008. The current system produces 0.25 falsealarms per hour. This is an acceptable rate for a look-out tower.