rus
Issue #1/2020
V. V. Startsev, V. K. Popov
Multispectral Fire Detection System by “Design Bureau “Astron” – Technology and Economics
Multispectral Fire Detection System by “Design Bureau “Astron” – Technology and Economics
DOI: 10.22184/1993-7296.FRos.2020.14.1.106.114
The detectors of fire protection systems for the budget consumer have a high proportion of false alarms. A project of a multispectral fire detection system is presented, where a useful signal is distinguished against false interference by a combination of signals from detectors analyzing images in several spectral (visible, average IR, thermal IR) ranges. The test results are presented.
The detectors of fire protection systems for the budget consumer have a high proportion of false alarms. A project of a multispectral fire detection system is presented, where a useful signal is distinguished against false interference by a combination of signals from detectors analyzing images in several spectral (visible, average IR, thermal IR) ranges. The test results are presented.
Теги: fire fire protection multispectral detection systems open fire smoldering thermal imaging detector возгорание многоспектральные системы обнаружения открытый огонь пожарная охрана тепловизионный детектор тление
Multispectral Fire Detection System by “Design Bureau “Astron” – Technology and Economics
V. V. Startsev, V. K. Popov,
JSC “Design Bureau “Astrohn», Lytkarino, Moscow region, Russia
The detectors of fire protection systems for the budget consumer have a high proportion of false alarms. A project of a multispectral fire detection system is presented, where a useful signal is distinguished against false interference by a combination of signals from detectors analyzing images in several spectral (visible, average IR, thermal IR) ranges. The test results are presented.
Keywords: fire protection, fire, smoldering, open fire, thermal imaging detector, multispectral detection systems
Received: 14.01.2020
Accepted: 28.01.2020
Introduction
Fire detection at an early stage of its occurrence is a key task of modern fire protection systems. Depending on the composition of combustible materials and environmental conditions, a fire at the initial stage can be characterized by various factors: local temperature increase (smoldering), the presence of smoke, a flame with smoke, or only an open flame.
The development of video fire detection systems is also due to economic factors: reducing the cost of detectors, the development of wireless communication systems, the emergence of energy-efficient communication systems for autonomous remote systems, the ability to automatically detect smoke and flame as the main factor, unified fire and security systems for monitoring extended objects and protected space.
The limiting factor in the development of automatic fire detection systems for extended objects is a high proportion of false alarms. This is primarily due to the fact that high-performance thermal-looking detectors of the “looking type” for a range of 3–5 microns are still quite expensive, and the use of other detectors leads to an increase in false alarms.
Another disadvantage of automatic systems is the relatively small range of fire detection.
Setting of the problem
An increase in the range of action, while reducing the likelihood of a false alarm, can be achieved by a combination of detectors that analyze the resulting image in several spectral ranges (visible, average IR, thermal IR). Multispectrality also provides the separation of the useful signal from the false due to lamps, the sun, glare, radiation reflections, etc.
A promising area for the development of early fire detection tools is round-the-clock remote video monitoring in several spectral ranges. The advantages of such monitoring are the ability to detect fire in open extended objects and spaces, the absence of the need for detectors and equipment to be in close proximity to a potential source of ignition or in contact with it, the detection of a fire in the initial stage with accurate determination of the coordinates and location of the source of ignition [1,2].
Solution to the problem – automatic detection system by “Design Bureau “Astron”
JSC “Design Bureau “Astron” has developed a multispectral system (complex) for early fire detection. The main objective of this hardware-software complex is remote automatic fire detection and monitoring by built-in analytics based on the analysis of multispectral images from detectors in the visible range (0.5–0.8 μm) and thermal imaging range (3–5 μm).
The complex has the ability to monitor the situation around the clock and automatically detect a fire in a radius of up to 5 km with a 360-degree circular view.
Built-in analytics uses for color analysis the color, temperature and time characteristics in the sequence of frames of two spectral ranges and integrates the obtained temperature data for real-time fire detection. After receiving a confirmation of the fire, communication and data transfer is carried out via fiber-optic communication channels or via mobile communication channels, as well as trunking communication from complex to complex within direct visibility.
The developed automatic hardware and software complex is an autonomous optoelectronic system for multispectral observation with simultaneous registration of an optical image, a thermal imaging image, and such a key fire factor as the excess of temperature over the background. The complex is designed for the early detection of fire at extended facilities, industrial enterprises, transport infrastructure facilities, in forestry and agriculture, etc.
The multispectral complex consists of a color digital video camera, matrix microbolometric uncooled thermal imaging detectors for a radiation range of 3–5 μm, with fields of view corresponding to the field of view of the camera, power supply, control and processing units placed in a common building (Fig. 1). The image in the visible (0.5–0.8 μm) and thermal imaging (3–5 μm) are combined for analysis purposes in both ranges using the “fusion” technology. Its essence is the complete combination of thermal imaging and television images in one frame. Depending on weather conditions, the operator is able to observe the scene in different spectral ranges: visible 0.35–0.78 μm or thermal imaging 3–5 μm, or a “mixture” of both pictures.
The data processing software shares video, thermal imaging channel data and spatio-temporal changes of objects to classify areas of the fire by secondary features (smoke, heated air) and by the presence of primary features (open fire) in a sequence of frames in real time.
As calculations show, from a distance of already 1000 meters a fire focus of 2 × 2 meters will not be detected by a video camera with a resolution of 2 MPs in the visible range (0.5–0.8 μm) with an angle of view of the lens of about 30 degrees. A fire center of this size with a temperature of 1500K will be detected by a video camera at a distance of not more than 150 m. It is optimal to use video cameras to detect secondary signs of a fire (for example, smoke). To increase the detection range of open fire, it is necessary to use photodetector arrays with a maximum sensitivity in the spectral range of 700–1500K, which corresponds to a spectral range of 3–5 μm.
Structure of the multispectral thermal imaging complex
The complex consists of 12 two-spectral cameras (0.5–0.8 μm, 3–5 μm) with fields of view coinciding in both ranges, installed in one common building with angles between the optical axes of 30 degrees. Each two-spectral camera has a combination of thermal imaging and video images with intelligent analysis of the thermal field. Analysis in the spectral range of 3–5 microns is carried out according to the model of “appearance of a hot object” exceeding the background temperature, without movement, the minimum area of the object is 4 square meters. Analysis in the spectral range of 0.5–0.8 μm is performed by the appearance of the movement of objects at angles of 35–135 degrees to the horizon, the absence of horizontal movements, an increase in area from 100 sq. M is possible. meters or more, an increase in the height of the object above 5 meters. In the process of tuning and testing, intelligent analysis should be able to “learn” using neural network technologies.
The use of intelligent video analytics with three-dimensional perspective construction and calibration at a distance to the horizon, the ability to classify objects according to parameters of area, speed of movement, direction of movement excludes the possibility of using scanning circular survey systems. For reliable operation of algorithms for automatic recognition of threat models and analysis of the contrast-background situation, a fixed position is required without moving the underlying surface. This requirement is ensured only by the fixed position of the optoelectronic systems and their preliminary calibration at the installation site for analysis of the observed scene.
To carry out 360-degree circular viewing without the possibility of circular scanning, it is necessary to use a minimum number of stationary thermal imaging cameras in the range of 3–5 microns. When the viewing angle of each camera is 30 degrees, it is necessary to use 12 cameras [3,4].
Peculiarities of PRD selection
In our opinion, the photo-receiving device (PRD) of the mid-IR range (3–5 microns) is most functional for detecting fire sources, since this range has minimal atmospheric attenuation and maximum sensitivity in the spectral range corresponding to an open flame. The radiation of the fire source in this range is maximum, while light smoke is almost transparent to radiation in this range. The energy parameters of radiation in the spectral range of 3–5 μm are several orders of magnitude higher than the values in the second thermal-imaging spectral range of 7–14 μm. It is for this reason that the 3–5 µm range is used to detect fires and open flames. The radiation range with a wavelength of 3–5 μm corresponds to cooled thermal imaging photodetectors based on indium antimonide (InSb), as well as to cadmium-mercury-tellurium (HgCdTe) structures. In the second radiation range of 7–14 μm, uncooled bolometric receivers are used.
The cost of PRDs based on bolometers in industrial production is two orders of magnitude lower than the cost of matrices based on HgCdTe, InSb, with typical NETD values (temperature sensitivity equal to the minimum equivalent noise of a temperature difference – Noise Equivalent Temperature Difference) for bolometric matrices 50–100 mK (for PRDs based on HgCdTe, values of the order of 10–20 mK are typical). The most important advantage of bolometric infrared detectors is the ability to work without cooling (at temperatures around 300K), while most photon detectors operate at cryogenic temperatures (usually at least 77K).
The spectral range of 7–14 μm corresponds to a maximum in the temperature range of 30–150 °C. In this temperature range, it is possible to detect only heated combustion products, warm smoke. However, in most real atmospheric situations and combustion conditions, the detection of fire by the smoke accompanying it is more effective in the visible range with high-resolution video cameras. Very rarely, the occurrence of fire occurs without combustion products visible in the range of 0.3–0.8 μm, corresponding to the visible range. The use of bolometric PRDs for the range 7–14 μm is not justified for the previously mentioned reasons, taking into account the maximum spectral sensitivity: low detection efficiency of open flames with temperatures above 500 °C, detection of objects with temperatures up to 150 °C, including heated cars, people, leading to false alarms. The possibility of detecting warm smoke and combustion products by microbolometric matrices in the range of 7–14 μm in practice is identical to the possibilities of their automatic detection in the visible range by heated particles of more than 1 μm resulting from combustion and visible in the range of 0.3–0.8 μm. At the same time, the price of microbolometric matrices with optics in the range of 7–14 μm significantly exceeds the cost of cameras in the visible range [5, 6].
The use of cooled systems operating in the range of 3–5 microns leads to a significant increase in the cost of the entire complex. The only uncooled matrix detector for the 3–5 micron range is the Astron‑64017D detector developed at “Design Bureau “Astron”.
Software features
We found that for the complex to work confidently, the overall scene analysis frequency up to 5 times per second is sufficient. Reducing the speed of aggregate analysis is necessary for a full analysis. The architecture of the complex, taking into account this parameter, allows the use of inexpensive and reliable standard network devices that do not require special software. This is optimal for the low bandwidth of a wired or wireless network when the complex is in remote places and work offline.
High reliability of fire detection and low probability of false alarms, as well as the accuracy of determining the coordinates of ignition is achieved both by the hardware (to a greater extent by the characteristics of detectors and infrared matrices), and by intelligent algorithmic support.
The analyzed image in both spectra is characterized by a large number of noise and disturbances. The visible range is affected by objects with similar parameters for analysis: fog, light flare, moving shadows from clouds, sunlight, etc. The thermal imaging range is characterized by a large level of intrinsic noise, intrinsic fluctuations of temperature fields.
To detect smoke in the visible range, it is based on an analysis of the dynamic and static properties of video images. The detector is based on an algorithm that includes the following basic steps: pre-processing frames; building a background frame and searching for moving areas; detection of candidate areas; classification of moving candidate areas, selection and analysis of the probabilities of classifying an event as a fire. The indicated algorithms for the work of intelligent analytics can work only after 3D‑calibration of the analyzed scene in range, size, area and movement. Three-dimensional calibration of the analyzed scene allows you to determine the size of objects and ranges depending on the location of objects relative to the horizon and perspective corresponding to the focal length of the lenses. Three-dimensional calibration of the scene, as well as the use of many video analysis algorithms, cannot be carried out with scanning systems, the presence of a moving underlying surface. It is for this reason that the detection of a fire by optoelectronic systems on board unmanned aerial vehicles is quite difficult, and is carried out only with the help of a person. The use of scanning systems reduces the cost of monitoring systems for all-round viewing, but cannot solve the problem of automatically detecting fires. When observing during the day in good weather, the television image of the visible spectrum is used to a greater extent, at night and in bad weather conditions, the thermal imaging spectrum is more mixed into the image. The advantage of the technology is a more natural picture of the field of view for the human eye, less operator fatigue, better indicators for detecting and recognizing a security threat.
Motion is considered as the primary sign when smoke is detected, and at the first stage of the algorithm, the background subtraction method is used to extract slowly moving areas from the current frame. The combination of moving parts in connected candidate areas is performed using operations of mathematical morphology and contour analysis by mathematical algorithms. A distinctive feature of the algorithm is that the classification of regions is based on the analysis of the direction of their motion vectors, determined by the block method of calculating the optical flux, directed at an angle from 45 ° to 135 ° to the lower horizontal axis of the frame (according to the main direction of smoke propagation), while ground objects (cars, people) move in a horizontal plane, taking into account the perspective. Further algorithms relate to Weber contrast calculations. This approach makes it possible to quite effectively distinguish smoke from real objects with similar behavior. At the final stage, the classification unit gives an alarm in case of smoke detection. The proposed algorithm is implemented programmatically using Visual C++ and the library of algorithms for computer vision and image processing OpenCV [6].
Test results
To confirm the possibility of detecting open flame, tests were carried out on a thermal imaging detector of the short-wave range (3–5 μm). The tests were carried out in order to confirm the possibility of detecting an open flame measuring 2 × 2 meters at distances of 5000 meters or more.
Open fire had a temperature of more than 1500K and was confidently detected at a distance of up to 5000 meters. Testing showed the need for mandatory use of intelligent video analysis.
Conclusions
To detect foci of ignition and the initial stages of fires, it is necessary to use the detection of open flame, as well as smoke and combustion products. For the detection of open flames with a burning or smoldering temperature of more than 500 °C, the spectral range of electromagnetic radiation of 3–5 microns is used. An open flame with an area of 2 × 2 m in projection in this range is detected automatically at a distance of up to 5 km with Astron‑64017D photodetector arrays with a maximum spectral sensitivity of 3–5 microns. The use of bolometric matrices for the spectral range of 7–14 μm with spectral sensitivity at temperatures up to 150 °C is impractical due to poor sensitivity to high temperatures, and their use for the detection of hot smoke and combustion products is less effective in comparison with cameras in the visible range, including cost-effectiveness parameters.
The total cost of the complex for a circular review does not exceed 7 million rubles. This budget is possible only with the use of bolometric matrices for a range of 3–5 microns, the use of refrigerated systems to work in this range in the complex would lead to a significant increase in cost, about 10 times.
REFERENCES
Ponomarenko V. P., Filachev A. M. Infrakrasnaya tekhnika i elektronnaya optika. Stanovlenie nauchnyh napravleniya. – M.: Fizmatkniga, 2016.
A. Rogalski. Infrared Detectors. − CRC-Press Taylor Francis Group, 2-nd ed., London New York, 2011, 876p.
Filachev A. M. , Taubkin I. I. , Trishenkov M. A. Tverdotel’naya fotoelektronika. Fotorezistory i fotopriemnye ustrojstva. − M.: Fizmatkniga, 2012, 368 pp.
Startsev V.V., Popov V.K., Anoshin K.E. IR- and video system based multispectral module of detection and analysis of threats for the protection of extended objects. – Photonics Russia. 2017; 63(3): 82–96. DOI: 10.22184 / 1993–7296.2017.63.3.82.96.
Kul’chickij N. A. , Naumov A. V. , Starcev V. V. Neohlazhdaemye mikrobolometry infrakrasnogo diapazona-sovremennoe sostoyanie i tendencii razvitiya. – Nano- i mikrosistemnaya tekhnika. 2018; 20(10): 613–624.
http://www.astrohn.com.
About authors
Startsev Vadim Valerievich, Candidate of Technical Sciences, ASTROHN Technology Ltd,
https://astrohn.ru, Lytkarino, Moscow region, Russia.
ORCID ID: 0000-0002-2800-544X
Popov Vladimir Konstantinovich,popov@astrohn.ru,
ASTROHN Technology Ltd, https://astrohn.ru, Lytkarino, Moscow region, Russia.
ORCID ID: 0000-0002-2267-1994
Contribution by the members
of the team of authors
The article was prepared on the basis of work by all members of the team of authors.
Development and research are carried out at the expense of ASTROHN Technology Ltd.
Conflict of interest
The authors claim that they have no conflict of interest.
V. V. Startsev, V. K. Popov,
JSC “Design Bureau “Astrohn», Lytkarino, Moscow region, Russia
The detectors of fire protection systems for the budget consumer have a high proportion of false alarms. A project of a multispectral fire detection system is presented, where a useful signal is distinguished against false interference by a combination of signals from detectors analyzing images in several spectral (visible, average IR, thermal IR) ranges. The test results are presented.
Keywords: fire protection, fire, smoldering, open fire, thermal imaging detector, multispectral detection systems
Received: 14.01.2020
Accepted: 28.01.2020
Introduction
Fire detection at an early stage of its occurrence is a key task of modern fire protection systems. Depending on the composition of combustible materials and environmental conditions, a fire at the initial stage can be characterized by various factors: local temperature increase (smoldering), the presence of smoke, a flame with smoke, or only an open flame.
The development of video fire detection systems is also due to economic factors: reducing the cost of detectors, the development of wireless communication systems, the emergence of energy-efficient communication systems for autonomous remote systems, the ability to automatically detect smoke and flame as the main factor, unified fire and security systems for monitoring extended objects and protected space.
The limiting factor in the development of automatic fire detection systems for extended objects is a high proportion of false alarms. This is primarily due to the fact that high-performance thermal-looking detectors of the “looking type” for a range of 3–5 microns are still quite expensive, and the use of other detectors leads to an increase in false alarms.
Another disadvantage of automatic systems is the relatively small range of fire detection.
Setting of the problem
An increase in the range of action, while reducing the likelihood of a false alarm, can be achieved by a combination of detectors that analyze the resulting image in several spectral ranges (visible, average IR, thermal IR). Multispectrality also provides the separation of the useful signal from the false due to lamps, the sun, glare, radiation reflections, etc.
A promising area for the development of early fire detection tools is round-the-clock remote video monitoring in several spectral ranges. The advantages of such monitoring are the ability to detect fire in open extended objects and spaces, the absence of the need for detectors and equipment to be in close proximity to a potential source of ignition or in contact with it, the detection of a fire in the initial stage with accurate determination of the coordinates and location of the source of ignition [1,2].
Solution to the problem – automatic detection system by “Design Bureau “Astron”
JSC “Design Bureau “Astron” has developed a multispectral system (complex) for early fire detection. The main objective of this hardware-software complex is remote automatic fire detection and monitoring by built-in analytics based on the analysis of multispectral images from detectors in the visible range (0.5–0.8 μm) and thermal imaging range (3–5 μm).
The complex has the ability to monitor the situation around the clock and automatically detect a fire in a radius of up to 5 km with a 360-degree circular view.
Built-in analytics uses for color analysis the color, temperature and time characteristics in the sequence of frames of two spectral ranges and integrates the obtained temperature data for real-time fire detection. After receiving a confirmation of the fire, communication and data transfer is carried out via fiber-optic communication channels or via mobile communication channels, as well as trunking communication from complex to complex within direct visibility.
The developed automatic hardware and software complex is an autonomous optoelectronic system for multispectral observation with simultaneous registration of an optical image, a thermal imaging image, and such a key fire factor as the excess of temperature over the background. The complex is designed for the early detection of fire at extended facilities, industrial enterprises, transport infrastructure facilities, in forestry and agriculture, etc.
The multispectral complex consists of a color digital video camera, matrix microbolometric uncooled thermal imaging detectors for a radiation range of 3–5 μm, with fields of view corresponding to the field of view of the camera, power supply, control and processing units placed in a common building (Fig. 1). The image in the visible (0.5–0.8 μm) and thermal imaging (3–5 μm) are combined for analysis purposes in both ranges using the “fusion” technology. Its essence is the complete combination of thermal imaging and television images in one frame. Depending on weather conditions, the operator is able to observe the scene in different spectral ranges: visible 0.35–0.78 μm or thermal imaging 3–5 μm, or a “mixture” of both pictures.
The data processing software shares video, thermal imaging channel data and spatio-temporal changes of objects to classify areas of the fire by secondary features (smoke, heated air) and by the presence of primary features (open fire) in a sequence of frames in real time.
As calculations show, from a distance of already 1000 meters a fire focus of 2 × 2 meters will not be detected by a video camera with a resolution of 2 MPs in the visible range (0.5–0.8 μm) with an angle of view of the lens of about 30 degrees. A fire center of this size with a temperature of 1500K will be detected by a video camera at a distance of not more than 150 m. It is optimal to use video cameras to detect secondary signs of a fire (for example, smoke). To increase the detection range of open fire, it is necessary to use photodetector arrays with a maximum sensitivity in the spectral range of 700–1500K, which corresponds to a spectral range of 3–5 μm.
Structure of the multispectral thermal imaging complex
The complex consists of 12 two-spectral cameras (0.5–0.8 μm, 3–5 μm) with fields of view coinciding in both ranges, installed in one common building with angles between the optical axes of 30 degrees. Each two-spectral camera has a combination of thermal imaging and video images with intelligent analysis of the thermal field. Analysis in the spectral range of 3–5 microns is carried out according to the model of “appearance of a hot object” exceeding the background temperature, without movement, the minimum area of the object is 4 square meters. Analysis in the spectral range of 0.5–0.8 μm is performed by the appearance of the movement of objects at angles of 35–135 degrees to the horizon, the absence of horizontal movements, an increase in area from 100 sq. M is possible. meters or more, an increase in the height of the object above 5 meters. In the process of tuning and testing, intelligent analysis should be able to “learn” using neural network technologies.
The use of intelligent video analytics with three-dimensional perspective construction and calibration at a distance to the horizon, the ability to classify objects according to parameters of area, speed of movement, direction of movement excludes the possibility of using scanning circular survey systems. For reliable operation of algorithms for automatic recognition of threat models and analysis of the contrast-background situation, a fixed position is required without moving the underlying surface. This requirement is ensured only by the fixed position of the optoelectronic systems and their preliminary calibration at the installation site for analysis of the observed scene.
To carry out 360-degree circular viewing without the possibility of circular scanning, it is necessary to use a minimum number of stationary thermal imaging cameras in the range of 3–5 microns. When the viewing angle of each camera is 30 degrees, it is necessary to use 12 cameras [3,4].
Peculiarities of PRD selection
In our opinion, the photo-receiving device (PRD) of the mid-IR range (3–5 microns) is most functional for detecting fire sources, since this range has minimal atmospheric attenuation and maximum sensitivity in the spectral range corresponding to an open flame. The radiation of the fire source in this range is maximum, while light smoke is almost transparent to radiation in this range. The energy parameters of radiation in the spectral range of 3–5 μm are several orders of magnitude higher than the values in the second thermal-imaging spectral range of 7–14 μm. It is for this reason that the 3–5 µm range is used to detect fires and open flames. The radiation range with a wavelength of 3–5 μm corresponds to cooled thermal imaging photodetectors based on indium antimonide (InSb), as well as to cadmium-mercury-tellurium (HgCdTe) structures. In the second radiation range of 7–14 μm, uncooled bolometric receivers are used.
The cost of PRDs based on bolometers in industrial production is two orders of magnitude lower than the cost of matrices based on HgCdTe, InSb, with typical NETD values (temperature sensitivity equal to the minimum equivalent noise of a temperature difference – Noise Equivalent Temperature Difference) for bolometric matrices 50–100 mK (for PRDs based on HgCdTe, values of the order of 10–20 mK are typical). The most important advantage of bolometric infrared detectors is the ability to work without cooling (at temperatures around 300K), while most photon detectors operate at cryogenic temperatures (usually at least 77K).
The spectral range of 7–14 μm corresponds to a maximum in the temperature range of 30–150 °C. In this temperature range, it is possible to detect only heated combustion products, warm smoke. However, in most real atmospheric situations and combustion conditions, the detection of fire by the smoke accompanying it is more effective in the visible range with high-resolution video cameras. Very rarely, the occurrence of fire occurs without combustion products visible in the range of 0.3–0.8 μm, corresponding to the visible range. The use of bolometric PRDs for the range 7–14 μm is not justified for the previously mentioned reasons, taking into account the maximum spectral sensitivity: low detection efficiency of open flames with temperatures above 500 °C, detection of objects with temperatures up to 150 °C, including heated cars, people, leading to false alarms. The possibility of detecting warm smoke and combustion products by microbolometric matrices in the range of 7–14 μm in practice is identical to the possibilities of their automatic detection in the visible range by heated particles of more than 1 μm resulting from combustion and visible in the range of 0.3–0.8 μm. At the same time, the price of microbolometric matrices with optics in the range of 7–14 μm significantly exceeds the cost of cameras in the visible range [5, 6].
The use of cooled systems operating in the range of 3–5 microns leads to a significant increase in the cost of the entire complex. The only uncooled matrix detector for the 3–5 micron range is the Astron‑64017D detector developed at “Design Bureau “Astron”.
Software features
We found that for the complex to work confidently, the overall scene analysis frequency up to 5 times per second is sufficient. Reducing the speed of aggregate analysis is necessary for a full analysis. The architecture of the complex, taking into account this parameter, allows the use of inexpensive and reliable standard network devices that do not require special software. This is optimal for the low bandwidth of a wired or wireless network when the complex is in remote places and work offline.
High reliability of fire detection and low probability of false alarms, as well as the accuracy of determining the coordinates of ignition is achieved both by the hardware (to a greater extent by the characteristics of detectors and infrared matrices), and by intelligent algorithmic support.
The analyzed image in both spectra is characterized by a large number of noise and disturbances. The visible range is affected by objects with similar parameters for analysis: fog, light flare, moving shadows from clouds, sunlight, etc. The thermal imaging range is characterized by a large level of intrinsic noise, intrinsic fluctuations of temperature fields.
To detect smoke in the visible range, it is based on an analysis of the dynamic and static properties of video images. The detector is based on an algorithm that includes the following basic steps: pre-processing frames; building a background frame and searching for moving areas; detection of candidate areas; classification of moving candidate areas, selection and analysis of the probabilities of classifying an event as a fire. The indicated algorithms for the work of intelligent analytics can work only after 3D‑calibration of the analyzed scene in range, size, area and movement. Three-dimensional calibration of the analyzed scene allows you to determine the size of objects and ranges depending on the location of objects relative to the horizon and perspective corresponding to the focal length of the lenses. Three-dimensional calibration of the scene, as well as the use of many video analysis algorithms, cannot be carried out with scanning systems, the presence of a moving underlying surface. It is for this reason that the detection of a fire by optoelectronic systems on board unmanned aerial vehicles is quite difficult, and is carried out only with the help of a person. The use of scanning systems reduces the cost of monitoring systems for all-round viewing, but cannot solve the problem of automatically detecting fires. When observing during the day in good weather, the television image of the visible spectrum is used to a greater extent, at night and in bad weather conditions, the thermal imaging spectrum is more mixed into the image. The advantage of the technology is a more natural picture of the field of view for the human eye, less operator fatigue, better indicators for detecting and recognizing a security threat.
Motion is considered as the primary sign when smoke is detected, and at the first stage of the algorithm, the background subtraction method is used to extract slowly moving areas from the current frame. The combination of moving parts in connected candidate areas is performed using operations of mathematical morphology and contour analysis by mathematical algorithms. A distinctive feature of the algorithm is that the classification of regions is based on the analysis of the direction of their motion vectors, determined by the block method of calculating the optical flux, directed at an angle from 45 ° to 135 ° to the lower horizontal axis of the frame (according to the main direction of smoke propagation), while ground objects (cars, people) move in a horizontal plane, taking into account the perspective. Further algorithms relate to Weber contrast calculations. This approach makes it possible to quite effectively distinguish smoke from real objects with similar behavior. At the final stage, the classification unit gives an alarm in case of smoke detection. The proposed algorithm is implemented programmatically using Visual C++ and the library of algorithms for computer vision and image processing OpenCV [6].
Test results
To confirm the possibility of detecting open flame, tests were carried out on a thermal imaging detector of the short-wave range (3–5 μm). The tests were carried out in order to confirm the possibility of detecting an open flame measuring 2 × 2 meters at distances of 5000 meters or more.
Open fire had a temperature of more than 1500K and was confidently detected at a distance of up to 5000 meters. Testing showed the need for mandatory use of intelligent video analysis.
Conclusions
To detect foci of ignition and the initial stages of fires, it is necessary to use the detection of open flame, as well as smoke and combustion products. For the detection of open flames with a burning or smoldering temperature of more than 500 °C, the spectral range of electromagnetic radiation of 3–5 microns is used. An open flame with an area of 2 × 2 m in projection in this range is detected automatically at a distance of up to 5 km with Astron‑64017D photodetector arrays with a maximum spectral sensitivity of 3–5 microns. The use of bolometric matrices for the spectral range of 7–14 μm with spectral sensitivity at temperatures up to 150 °C is impractical due to poor sensitivity to high temperatures, and their use for the detection of hot smoke and combustion products is less effective in comparison with cameras in the visible range, including cost-effectiveness parameters.
The total cost of the complex for a circular review does not exceed 7 million rubles. This budget is possible only with the use of bolometric matrices for a range of 3–5 microns, the use of refrigerated systems to work in this range in the complex would lead to a significant increase in cost, about 10 times.
REFERENCES
Ponomarenko V. P., Filachev A. M. Infrakrasnaya tekhnika i elektronnaya optika. Stanovlenie nauchnyh napravleniya. – M.: Fizmatkniga, 2016.
A. Rogalski. Infrared Detectors. − CRC-Press Taylor Francis Group, 2-nd ed., London New York, 2011, 876p.
Filachev A. M. , Taubkin I. I. , Trishenkov M. A. Tverdotel’naya fotoelektronika. Fotorezistory i fotopriemnye ustrojstva. − M.: Fizmatkniga, 2012, 368 pp.
Startsev V.V., Popov V.K., Anoshin K.E. IR- and video system based multispectral module of detection and analysis of threats for the protection of extended objects. – Photonics Russia. 2017; 63(3): 82–96. DOI: 10.22184 / 1993–7296.2017.63.3.82.96.
Kul’chickij N. A. , Naumov A. V. , Starcev V. V. Neohlazhdaemye mikrobolometry infrakrasnogo diapazona-sovremennoe sostoyanie i tendencii razvitiya. – Nano- i mikrosistemnaya tekhnika. 2018; 20(10): 613–624.
http://www.astrohn.com.
About authors
Startsev Vadim Valerievich, Candidate of Technical Sciences, ASTROHN Technology Ltd,
https://astrohn.ru, Lytkarino, Moscow region, Russia.
ORCID ID: 0000-0002-2800-544X
Popov Vladimir Konstantinovich,popov@astrohn.ru,
ASTROHN Technology Ltd, https://astrohn.ru, Lytkarino, Moscow region, Russia.
ORCID ID: 0000-0002-2267-1994
Contribution by the members
of the team of authors
The article was prepared on the basis of work by all members of the team of authors.
Development and research are carried out at the expense of ASTROHN Technology Ltd.
Conflict of interest
The authors claim that they have no conflict of interest.
Readers feedback
