Fire scene deep detection method and device based on machine vision

Abstract

The invention discloses a fire scene deep detection method and device based on machine vision, and aims to solve the problems that in the prior art, flames and interference light sources cannot be distinguished, the recognition efficiency of the flames is low, misjudgment of the fire situation is easily caused, the rescue result is affected, and the detection efficiency is low. And the fire hazard grade cannot be judged according to flame and smoke conditions, so that rescue workers are easily injured. The detection method comprises the following steps: step 1, starting the mobile control terminal; and 2, the wheels are controlled to work through the mobile control terminal. According to the detection method and device, flames and interference light sources are distinguished through circularity, so that the flame recognition efficiency is improved, fire disasters and smoke are divided into multiple grades, on the premise that the life safety of rescue workers is guaranteed, personnel and properties in a fire scene are rescued, an ultraviolet detector, a smoke detector and an infrared detector are protected, and the safety of the rescue workers is improved. And the ultraviolet detector, the smoke detector and the infrared detector are prevented from being damaged.

Description

Technical Field

[0001] This invention belongs to the field of fire detection technology, specifically relating to a method and apparatus for in-depth fire scene detection based on machine vision. Background Technology

[0002] Blindly entering a fire scene after it breaks out could result in casualties. Therefore, it is necessary to use detection devices to convert the temperature, smoke, gas, and radiation intensity in the fire into electrical signals to identify the situation at the fire scene and facilitate subsequent rescue and handling.

[0003] Currently, patent CN201410768586.6 discloses a vision-based fire detection device, including a smoke detection light source and a detection host. In the first lens of the detection host, a polarizer, a high-pass filter, and a low-pass filter are arranged in any order in front of the optical lenses. A first CMOS sensor is located behind the optical lenses, connected to an analog video interface and bidirectionally connected to a data processing device, transmitting data to the data processing device. The data processing device is bidirectionally connected to a memory, receiving and retrieving data from the data processing device. The data processing device is also connected to an output module. The smoke detection light source is an infrared LED array, connected to an m-sequence generator. It uses infrared LEDs with emission spectra between the cutoff frequencies of high and low pass filters as brightness target references, and uses an m-sequence to generate illumination. The detector uses detection synchronized with the m-sequence to eliminate other asynchronous interference information and greatly reduce misjudgments. However, this detection method and device cannot distinguish between flames and interfering light sources, has low flame identification efficiency, and is prone to misjudging fire conditions, affecting rescue results. Furthermore, it cannot determine the fire level based on flame and smoke conditions, which can easily cause injury to rescue personnel.

[0004] Therefore, the problem of low efficiency in flame identification needs to be addressed in order to improve the application scenarios of in-depth fire detection methods and devices. Summary of the Invention

[0005] (1) Technical problems to be solved

[0006] In view of the shortcomings of the prior art, the purpose of this invention is to provide a method and device for in-depth fire detection based on machine vision. This detection method and device aim to solve the technical problems of the prior art, which cannot distinguish between flames and interfering light sources, has low flame identification efficiency, is prone to misjudging the fire situation, affects the rescue results, and cannot determine the fire level based on flames and smoke, which can easily cause injury to rescue personnel.

[0007] (2) Technical solution

[0008] To address the aforementioned technical problems, this invention provides a machine vision-based method for in-depth fire scene detection, the steps of which are as follows:

[0009] Step 1: Activate the mobile control terminal, use the Wi-Fi wireless Bluetooth module to connect the detection device to the mobile control terminal, and then send the detection device into the fire scene;

[0010] Step 2: Control the wheels via the mobile control terminal to move the detection device forward, backward, turn, or stop;

[0011] Step 3: The ultrasonic obstacle avoidance module drives the ultrasonic obstacle avoidance sensor to work, emitting a 40KHz sound wave. When the sound wave returns, the ultrasonic obstacle avoidance module records the total time t from the emission to the return of the ultrasonic wave. Based on the speed of sound in the air s, the distance d = t / s is calculated.

[0012] Step 4: The ultrasonic obstacle avoidance module sends the distance d to the obstacle analysis unit. The obstacle analysis unit has a preset safe distance. When d ≤ d, the detection device stops moving.

[0013] Step 5: The ultraviolet sensing module, infrared sensing module, and smoke sensing module drive the ultraviolet detector, infrared detector, and smoke detector to work, and convert the collected flame data and smoke data through the digital-to-analog conversion module;

[0014] Step Six: Perform Flame Analysis:

[0015] 1) The flame analysis unit preprocesses the flame image, including denoising and sharpening.

[0016] 2) Distinguish between flames and interfering light sources by their roundness. The formula for calculating roundness is: ;

[0017] 3) Calculate the number N of flaming pixels in the flame image. The preset threshold for flaming pixels in the flame analysis unit is... , , and The number of lit pixels N is compared with the lit pixel threshold. If N < If so, it is determined that there is no fire; if ≤N≤ If it is a flame from a general fire, then it is determined to be a flame from a common fire; if ≤N≤ If it is a large fire flame, then it is determined to be a large fire flame; if ≤N≤ If N > 0, it is determined to be a major fire; if N > 0, it is determined to be a major fire. If the fire is identified as an exceptionally serious fire, the corresponding result will be sent to the mobile control terminal.

[0018] Step 7: The results detected by the smoke detector are sent from the smoke sensing module to the smoke analysis unit. The smoke analysis unit has a preset smoke concentration threshold. , , and The smoke concentration Y is compared with the smoke concentration threshold. If Y < If it is smokeless, it is considered smokeless; if ≤Y≤ If it is smoke from a general fire, then it is determined to be smoke from a general fire; if ≤Y≤ If it is smoke from a large fire, it is determined to be smoke from a large fire; if ≤Y≤ If Y > 0, it is determined to be a major fire smoke; If the smoke indicates a particularly serious fire, the result will be sent to the mobile control terminal.

[0019] A machine vision-based fire scene deep detection device includes the machine vision-based fire scene deep detection method described above. The device includes a chassis with wheels at its lower end. A heat-insulating protective cover with heat insulation function is fixedly connected to the upper end of the chassis. A microcontroller and a power supply are fixedly connected to the upper end of the chassis and disposed inside the heat-insulating protective cover. Ultrasonic obstacle avoidance sensors are fixedly connected around the heat-insulating protective cover. An ultraviolet detector, a smoke detector, and an infrared detector are fixedly connected sequentially from front to back on the upper end of the heat-insulating protective cover. Supports are fixedly connected to the four corners of the upper end of the heat-insulating protective cover. The support column has a protective plate sleeved on its upper outer side. The upper end of the support column has a threaded hole with a hand-tightening bolt threaded into it. An elastic element that can deform elastically is sleeved on the outer side of the support column. The front end of the heat insulation protective cover is provided with a charging port for charging the power supply. The microcontroller integrates a Wi-Fi wireless Bluetooth module, an ultraviolet sensor module, a digital-to-analog converter module, a power supply module, an ultrasonic obstacle avoidance module, an infrared sensor module, a smoke sensor module, and a data analysis module. The data analysis module includes a flame analysis unit, a smoke analysis unit, and an obstacle analysis unit.

[0020] Preferably, the flame analysis unit has five preset flame levels: extremely serious fire flame, serious fire flame, relatively large fire flame, general fire flame, and no fire. The analysis rule of the flame analysis unit is: compare the number of lit pixels N with the lit pixel threshold; if N < If so, it is determined that there is no fire; if ≤N≤ If it is a flame from a general fire, then it is determined to be a flame from a common fire; if ≤N≤ If it is a large fire flame, then it is determined to be a large fire flame; if ≤N≤ If N > 0, it is determined to be a major fire; if N > 0, it is determined to be a major fire. If it is a fire, it is classified as an exceptionally serious fire.

[0021] Preferably, the preset smoke concentration levels within the smoke analysis unit are: smoke concentration for extremely serious fires, smoke concentration for major fires, smoke concentration for relatively large fires, smoke concentration for general fires, and no smoke. The analysis rule of the smoke analysis unit is: comparing the smoke concentration Y with a smoke concentration threshold; if Y < If it is smokeless, it is considered smokeless; if ≤Y≤ If it is smoke from a general fire, then it is determined to be smoke from a general fire; if ≤Y≤ If it is smoke from a large fire, it is determined to be smoke from a large fire; if ≤Y≤ If Y > 0, it is determined to be a major fire smoke; If so, it is determined to be smoke from an exceptionally serious fire.

[0022] Preferably, the preset safe distance within the obstacle analysis unit is 10cm-200cm.

[0023] Preferably, the Wi-Fi wireless Bluetooth module has a maximum transmission rate of 54Mbps and a maximum coverage range of 100 meters.

[0024] Preferably, the heat insulation protective cover is made of vacuum heat insulation board or fireproof board, and the heat insulation protective cover is a double-layer design.

[0025] Preferably, in the circularity calculation of the flame analysis unit... The circularity of the primitive numbered k is represented by the perimeter of the k-th primitive. Let be the area of ​​the k-th primitive, and m be the number of suspicious flame primitives in the image.

[0026] Preferably, the height of the elastic element is greater than the height of the ultraviolet detector.

[0027] (3) Beneficial effects

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows: The detection method and device of the present invention utilize roundness to distinguish between flames and interfering light sources, thereby improving the efficiency of flame identification and avoiding misjudgment of fire conditions. Furthermore, through the flame analysis unit and smoke analysis unit, the situation in the fire scene is analyzed, and the fire and smoke are divided into multiple levels, which facilitates rescuers to formulate appropriate rescue plans based on the fire scene situation. Under the premise of ensuring the safety of rescuers, people and property in the fire scene can be rescued. Moreover, through the design of the protective plate, the ultraviolet detector, smoke detector and infrared detector are protected when objects fall in the fire scene, avoiding damage to the ultraviolet detector, smoke detector and infrared detector, and improving the service life of the detection device. Attached Figure Description

[0029] Figure 1 This is a three-dimensional structural schematic diagram of a specific embodiment of the detection device of the present invention;

[0030] Figure 2 This is a front structural diagram of a specific embodiment of the detection device of the present invention;

[0031] Figure 3 This is a side view of a specific embodiment of the detection device of the present invention;

[0032] Figure 4 This is a top view schematic diagram of a specific embodiment of the detection device of the present invention;

[0033] Figure 5 This is a schematic diagram of the microcontroller's frame structure in one specific embodiment of the detection device of the present invention;

[0034] Figure 6 This is a schematic diagram of the framework structure of the data analysis module in one specific embodiment of the detection device of the present invention.

[0035] The labels in the attached diagram are as follows: 1. Chassis; 2. Wheels; 3. Heat insulation protective cover; 4. Microcontroller; 5. Power supply; 6. Ultrasonic obstacle avoidance sensor; 7. Ultraviolet detector; 8. Smoke detector; 9. Infrared detector; 10. Support column; 11. Protective plate; 12. Threaded hole; 13. Hand-tightening bolt; 14. Elastic element; 15. Charging port; 16. Wi-Fi wireless Bluetooth module; 17. Ultraviolet sensor module; 18. Digital-to-analog conversion module; 19. Power supply module; 20. Ultrasonic obstacle avoidance module; 21. Infrared sensor module; 22. Smoke sensor module; 23. Data analysis module; 231. Flame analysis unit; 232. Smoke analysis unit; 233. Obstacle analysis unit. Detailed Implementation

[0036] Example 1

[0037] This specific embodiment is a machine vision-based method and device for in-depth fire scene detection, and its three-dimensional structural schematic diagram is shown below. Figure 1 As shown, its front structural diagram is as follows Figure 2 As shown, the steps of this machine vision-based method for in-depth fire scene detection are as follows:

[0038] Step 1: Start the mobile control terminal, use the Wi-Fi wireless Bluetooth module 16 to connect the detection device with the mobile control terminal, and then send the detection device into the fire scene.

[0039] Step 2: Control the wheel 2 via the mobile control terminal to drive the detection device forward, backward, turn, or stop;

[0040] Step 3: The ultrasonic obstacle avoidance module 20 drives the ultrasonic obstacle avoidance sensor 6 to work and emits a 40KHz sound wave. When the sound wave returns, the ultrasonic obstacle avoidance module 20 records the total time t from the emission to the return of the ultrasonic wave. Based on the speed of sound in the air s, the distance d=t / s is calculated.

[0041] Step 4: The ultrasonic obstacle avoidance module 20 sends the distance d to the obstacle analysis unit 233. The obstacle analysis unit 233 has a preset safe distance. When d ≤ d, the detection device stops moving.

[0042] Step 5: The ultraviolet sensing module 17, the infrared sensing module 21, and the smoke sensing module 22 drive the ultraviolet detector 7, the infrared detector 9, and the smoke detector 8 to work, and convert the collected flame data and smoke data through the digital-to-analog conversion module 18;

[0043] Step Six: Perform Flame Analysis:

[0044] 1) The flame analysis unit 231 preprocesses the flame image, including denoising and sharpening.

[0045] 2) Distinguish between flames and interfering light sources by their roundness. The formula for calculating roundness is: ;

[0046] 3) Calculate the number N of flaming pixels in the flame image. The preset threshold for flaming pixels in the flame analysis unit 231 is... , , and The number of lit pixels N is compared with the lit pixel threshold. If N < If so, it is determined that there is no fire; if ≤N≤ If it is a flame from a general fire, then it is determined to be a flame from a common fire; if ≤N≤ If it is a large fire flame, then it is determined to be a large fire flame; if ≤N≤ If N > 0, it is determined to be a major fire; if N > 0, it is determined to be a major fire. If the fire is identified as an exceptionally serious fire, the corresponding result will be sent to the mobile control terminal.

[0047] Step 7: The result detected by the smoke detector 8 is sent by the smoke sensing module 22 to the smoke analysis unit 232. The preset smoke concentration threshold in the smoke analysis unit 232 is... , , and The smoke concentration Y is compared with the smoke concentration threshold. If Y < If it is smokeless, it is considered smokeless; if ≤Y≤ If it is smoke from a general fire, then it is determined to be smoke from a general fire; if ≤Y≤ If it is smoke from a large fire, it is determined to be smoke from a large fire; if ≤Y≤ If Y > 0, it is determined to be a major fire smoke; If the smoke indicates a particularly serious fire, the result will be sent to the mobile control terminal.

[0048] The machine vision-based fire scene deep detection device includes the machine vision-based fire scene deep detection method described above. The device includes a chassis 1, with wheels 2 mounted on the lower end of the chassis 1. A heat-insulating protective cover 3 with heat insulation function is fixedly connected to the upper end of the chassis 1. A microcontroller 4 and a power supply 5 are fixedly connected to the upper end of the chassis 1, located inside the heat-insulating protective cover 3. Ultrasonic obstacle avoidance sensors 6 are fixedly connected around the heat-insulating protective cover 3. An ultraviolet detector 7, a smoke detector 8, and an infrared detector 9 are fixedly connected sequentially from front to back on the upper end of the heat-insulating protective cover 3. Support columns 10 are fixedly connected to the four corners of the upper end of the heat-insulating protective cover 3, with the upper outer side of the support columns 10 sleeved with… The device includes a protective plate 11, a threaded hole 12 at the upper end of a support column 10, a hand-tightening bolt 13 connected to the threaded hole 12, an elastic element 14 that can be elastically deformed fitted on the outside of the support column 10, a charging port 15 for charging the power supply 5 at the front end of the heat insulation protective cover 3, and a microcontroller 4 integrating a Wi-Fi wireless Bluetooth module 16, an ultraviolet sensor module 17, a digital-to-analog conversion module 18, a power supply module 19, an ultrasonic obstacle avoidance module 20, an infrared sensor module 21, a smoke sensor module 22, and a data analysis module 23. The data analysis module 23 includes a flame analysis unit 231, a smoke analysis unit 232, and an obstacle analysis unit 233.

[0049] The flame analysis unit 231 has five preset flame levels: extremely serious fire flame, serious fire flame, relatively large fire flame, general fire flame, and no fire. The analysis rule of the flame analysis unit 231 is: compare the number of lit pixels N with the lit pixel threshold; if N < If so, it is determined that there is no fire; if ≤N≤ If it is a flame from a general fire, then it is determined to be a flame from a common fire; if ≤N≤ If it is a large fire flame, then it is determined to be a large fire flame; if ≤N≤ If N > 0, it is determined to be a major fire; if N > 0, it is determined to be a major fire. If the smoke concentration is too high, it is determined to be a particularly serious fire. The smoke concentration levels preset in the smoke analysis unit 232 are: particularly serious fire smoke concentration, serious fire smoke concentration, relatively large fire smoke concentration, general fire smoke concentration, and no smoke. The analysis rule of the smoke analysis unit 232 is: compare the smoke concentration Y with the smoke concentration threshold. If Y < If it is smokeless, it is considered smokeless; if ≤Y≤ If it is smoke from a general fire, then it is determined to be smoke from a general fire; if ≤Y≤ If it is smoke from a large fire, it is determined to be smoke from a large fire; if ≤Y≤ If Y > 0, it is determined to be a major fire smoke; If so, it is determined to be smoke from an exceptionally serious fire.

[0050] Meanwhile, the obstacle analysis unit 233 has a preset safe distance of 10cm-200cm, the Wi-Fi wireless Bluetooth module 16 has a maximum transmission rate of 54Mbps and a maximum coverage range of 100 meters, and the heat insulation protective cover 3 is made of vacuum heat insulation board or fireproof board, and the heat insulation protective cover 3 has a double-layer design.

[0051] In addition, in the circularity calculation of flame analysis unit 231 The circularity of the primitive numbered k is represented by the perimeter of the k-th primitive. Let be the area of ​​the k-th primitive, and m be the number of suspicious flame primitives in the image.

[0052] In addition, the height of the elastic element 14 is greater than the height of the ultraviolet detector 7.

[0053] A side structural diagram of the detection method and its device is shown below. Figure 3 As shown, its top view structural diagram is as follows: Figure 4 As shown, the schematic diagram of the frame structure of its microcontroller 4 is as follows: Figure 5 As shown, the framework structure diagram of its data analysis module 23 is as follows: Figure 6 As shown.

[0054] When using the device of this technical solution, step one: start the mobile control terminal, use the Wi-Fi wireless Bluetooth module 16 to connect the detection device with the mobile control terminal, and then send the detection device into the fire scene; step two: control the wheels 2 through the mobile control terminal to drive the detection device forward, backward, turn, or stop; step three: the ultrasonic obstacle avoidance module 20 drives the ultrasonic obstacle avoidance sensor 6 to work, emitting 40KHz sound waves outward. When the sound waves return, the ultrasonic obstacle avoidance module 20 records the total time t of the ultrasonic waves from emission to return, and calculates the distance d=t / s based on the speed of sound in the air; step four: the ultrasonic obstacle avoidance module 20 will... The distance d is sent to the obstacle analysis unit 233. The preset safe distance within the obstacle analysis unit 233 is 50cm. When d ≤ 50cm, the detection device stops moving. Step 5: The ultraviolet sensing module 17, the infrared sensing module 21, and the smoke sensing module 22 drive the ultraviolet detector 7, the infrared detector 9, and the smoke detector 8 to work, and convert the collected flame data and smoke data through the digital-to-analog conversion module 18. Step 6: Perform flame analysis: 1) The flame analysis unit 231 preprocesses the flame image, including denoising and sharpening. 2) The flame and interfering light source are distinguished by circularity. The circularity calculation formula is: where ... The circularity of the primitive numbered k is represented by the perimeter of the k-th primitive. Let m be the area of ​​the k-th primitive, and m be the number of suspicious flame primitives in the image; 3) Calculate the number of flame pixels N in the flame image, where the preset threshold for flame pixels in the flame analysis unit 231 is... , , and The number of lit pixels N is compared with the lit pixel threshold. If N < If so, it is determined that there is no fire; if ≤N≤ If it is a flame from a general fire, then it is determined to be a flame from a common fire; if ≤N≤ If it is a large fire flame, then it is determined to be a large fire flame; if ≤N≤ If N > 0, it is determined to be a major fire; if N > 0, it is determined to be a major fire. If the smoke detector 8 detects a fire, it is determined to be a particularly serious fire, and the corresponding result is sent to the mobile control terminal; Step 7: The result detected by the smoke detector 8 is sent by the smoke sensing module 22 to the smoke analysis unit 232, where the preset smoke concentration threshold is... , , and The smoke concentration Y is compared with the smoke concentration threshold. If Y < If it is smokeless, it is considered smokeless; if ≤Y≤ If it is smoke from a general fire, then it is determined to be smoke from a general fire; if ≤Y≤ If it is smoke from a large fire, it is determined to be smoke from a large fire; if ≤Y≤ If Y > 0, it is determined to be a major fire smoke; If the smoke indicates a particularly serious fire, the result will be sent to the mobile control terminal.

Claims

1. A machine vision-based method for in-depth fire scene detection, the steps of which are as follows: Step 1: Start the mobile control terminal, use the wifi wireless Bluetooth module (16) to connect the detection device with the mobile control terminal, and then send the detection device into the fire scene; Step 2: Control the wheels (2) through the mobile control terminal to drive the detection device forward, backward, turn or stop; Step 3: The ultrasonic obstacle avoidance module (20) drives the ultrasonic obstacle avoidance sensor (6) to work and emits a 40KHz sound wave. When the sound wave returns, the ultrasonic obstacle avoidance module (20) records the total time t from the emission to the return of the ultrasonic wave. Based on the speed of sound in the air s, the distance d=t / s is calculated. Step 4: The ultrasonic obstacle avoidance module (20) sends the distance d to the obstacle analysis unit (233). The obstacle analysis unit (233) has a preset safe distance. When d ≤ d, the detection device stops moving. Step 5: The ultraviolet sensing module (17), infrared sensing module (21) and smoke sensing module (22) drive the ultraviolet detector (7), infrared detector (9) and smoke detector (8) to work, and convert the collected flame data and smoke data through the digital-to-analog conversion module (18); Step Six: Perform Flame Analysis: 1) The flame analysis unit (231) preprocesses the flame image, including denoising and sharpening. 2) Distinguish between flames and interfering light sources by their roundness. The formula for calculating roundness is: ； 3) Calculate the number of lit pixels N in the flame image. The preset lit pixel threshold in the flame analysis unit (231) is: , , and The number of lit pixels N is compared with the lit pixel threshold. If N < If so, it is determined that there is no fire; if ≤N≤ If it is a flame from a general fire, then it is determined to be a flame from a common fire; if ≤N≤ If it is a large fire flame, then it is determined to be a large fire flame; if ≤N≤ If N > 0, it is determined to be a major fire; if N > 0, it is determined to be a major fire. If the fire is identified as an exceptionally serious fire, the corresponding result will be sent to the mobile control terminal. Step 7: The result detected by the smoke detector (8) is sent by the smoke sensing module (22) to the smoke analysis unit (232). The preset smoke concentration threshold in the smoke analysis unit (232) is... , , and The smoke concentration Y is compared with the smoke concentration threshold. If Y < If it is smokeless, it is considered smokeless; if ≤Y≤ If it is smoke from a general fire, then it is determined to be smoke from a general fire; if ≤Y≤ If it is smoke from a large fire, it is determined to be smoke from a large fire; if ≤Y≤ If Y > 0, it is determined to be a major fire smoke; If the smoke indicates a particularly serious fire, the result will be sent to the mobile control terminal.

2. A machine vision-based fire scene depth detection device, characterized in that... The device includes a machine vision-based fire scene detection method according to claim 1, comprising a chassis (1), wheels (2) at the lower end of the chassis (1), a heat-insulating protective cover (3) with heat insulation function fixedly connected to the upper end of the chassis (1), a single-chip microcomputer (4) and a power supply (5) fixedly connected to the upper end of the chassis (1) and disposed inside the heat-insulating protective cover (3), ultrasonic obstacle avoidance sensors (6) fixedly connected around the heat-insulating protective cover (3), an ultraviolet detector (7), a smoke detector (8) and an infrared detector (9) fixedly connected sequentially from front to back at the upper end of the heat-insulating protective cover (3), and support columns (10) fixedly connected at the four corners of the upper end of the heat-insulating protective cover (3), with a protective plate (11) sleeved on the outer side of the upper end of the support column (10). The upper end of the support column (10) is provided with a threaded hole (12), and a hand-tightening bolt (13) is connected to the threaded hole (12). An elastic element (14) that can be elastically deformed is sleeved on the outside of the support column (10). The front end of the heat insulation protective cover (3) is provided with a charging port (15) for charging the power supply (5). The microcontroller (4) integrates a Wi-Fi wireless Bluetooth module (16), an ultraviolet sensor module (17), a digital-to-analog conversion module (18), a power supply module (19), an ultrasonic obstacle avoidance module (20), an infrared sensor module (21), a smoke sensor module (22), and a data analysis module (23). The data analysis module (23) includes a flame analysis unit (231), a smoke analysis unit (232), and an obstacle analysis unit (233).

3. The machine vision-based fire scene depth detection device according to claim 2, characterized in that, The flame analysis unit (231) has five preset flame levels: extremely serious fire flame, serious fire flame, relatively large fire flame, general fire flame, and no fire. The analysis rule of the flame analysis unit (231) is: compare the number of fire pixels N with the fire pixel threshold. If N < If so, it is determined that there is no fire; if ≤N≤ If it is a flame from a general fire, then it is determined to be a flame from a common fire; if ≤N≤ If it is a large fire flame, then it is determined to be a large fire flame; if ≤N≤ If N > 0, it is determined to be a major fire; if N > 0, it is determined to be a major fire. If it is a fire, it is classified as an exceptionally serious fire.

4. The machine vision-based fire scene depth detection device according to claim 2, characterized in that, The smoke concentration levels preset in the smoke analysis unit (232) are: smoke concentration for extremely serious fires, smoke concentration for major fires, smoke concentration for relatively large fires, smoke concentration for general fires, and no smoke. The analysis rule of the smoke analysis unit (232) is: compare the smoke concentration Y with the smoke concentration threshold. If Y < If it is smokeless, it is considered smokeless; if ≤Y≤ If it is smoke from a general fire, then it is determined to be smoke from a general fire; if ≤Y≤ If it is smoke from a large fire, it is determined to be smoke from a large fire; if ≤Y≤ If Y > 0, it is determined to be a major fire smoke; If so, it is determined to be smoke from an exceptionally serious fire.

5. The machine vision-based fire scene depth detection device according to claim 2, characterized in that, The preset safe distance within the obstacle analysis unit (233) is 10cm-200cm.

6. The machine vision-based fire scene depth detection device according to claim 2, characterized in that, The Wi-Fi wireless Bluetooth module (16) has a maximum transmission rate of 54Mbps and a maximum coverage range of 100 meters.

7. The machine vision-based fire scene depth detection device according to claim 2, characterized in that, The heat insulation protective cover (3) is made of vacuum heat insulation board or fireproof board, and the heat insulation protective cover (3) is a double-layer design.

8. The machine vision-based fire scene depth detection device according to claim 2, characterized in that, In the flame analysis unit (231), the roundness calculation is performed. The circularity of the primitive numbered k is represented by the perimeter of the k-th primitive. Let be the area of ​​the k-th primitive, and m be the number of suspicious flame primitives in the image.

9. The machine vision-based fire scene depth detection device according to claim 2, characterized in that, The height of the elastic element (14) is greater than the height of the ultraviolet detector (7).

Images