METHOD OF MONITORING FIRE HAZARD IN ROOM Russian patent published in 2025 - IPC A62C3/00 A62C37/00 A62C37/36 A62C37/40 G08B17/00 G08B17/10 G06N3/00 

FIELD: fire-fighting means.

SUBSTANCE: invention relates to fire-fighting equipment, namely to a method for monitoring fire hazard in a room, including detection of fires using a neural network. Method involves the use of a portable fire safety monitoring device made in the form of a suitcase-container, from which thermal fire detectors, flame detectors, oxygen, carbon monoxide and carbon dioxide analytical gas detectors and a video surveillance camera are removed and placed on the ceiling and connect to the suitcase-container connectors and continuously measure the air temperature and its growth rate, the concentration of smoke particles, determine the electromagnetic radiation of the flame, the concentrations of oxygen, carbon dioxide, carbon monoxide in the air. Video surveillance camera, taken out of the suitcase-container, is placed on the ceiling in the corner of the room and is connected to the connectors of the suitcase-container and to a mobile computer taken out of the suitcase-container, the power supply units of which are connected to the mains and a stream of video frames is obtained. During video filming, the floor image of the room is pre-marked out by dividing it into rectangular zones equal in area n. Alternately in each rectangular zone reference object of known area is placed and scale factor is determined for each rectangular zone from expression: , where S_kn is known area of reference object placed in each rectangular zone on image, covered with pixels, cm²; N_p is the number of pixels belonging to the mask of the reference object of known area on the image. Interrogation of said fire detectors and gas-analytical sensors is carried out in predetermined time intervals. Signals from all fire detectors are recorded and is converted from digital form into RS-485 interface signal. Signals of gas-analytical sensors of oxygen, carbon monoxide, carbon dioxide are recorded and converted from analogue into digital form, with subsequent conversion into RS-485 interface signal. Values measured by the fire detectors are compared to preset threshold values, and if they coincide, the fire detectors are actuated. When one of the fire detectors is actuated, they are re-interviewed. During repeated operation of a previously actuated fire detector, or during operation of smoke fire alarm units, or during operation of thermal fire alarm units, or when the fire detectors are triggered, or when one of the gas-analytical sensors of oxygen, carbon monoxide, carbon dioxide is triggered, a signal is generated about the occurrence of a fire. Simultaneously, the stream of video frames obtained with interval of 500 ms is analysed using a neural network pre-trained on fire hazards in a closed room model, obtaining at least 4000 images of fire hazards of different combustion intensity with resolution of 1280×720 pixels. At the same time in the training sample fire seats are marked in accordance with their visible boundaries, circling their contours. Location of fire source is determined using trained neural network, selecting mask of fire source as area covering flame of fire source, and contour of said mask in form of polygon corresponding to flame contour, as well as mask of objects visually similar to fire source. In each rectangular zone, searching for x and y coordinates of geometric centre of fire source, and if in zone coordinate x of geometrical centre of fire source on image is greater than or equal to coordinate x₁ of first along abscissa upper corner of rectangular zone and is less than or equal to coordinate x₂ of opposite lower corner of this rectangular zone, as well as if the y coordinate of the geometric centre of the fire source on the image is greater than or equal to the coordinate y₁ of first along ordinate axis angle of rectangular zone and is less than or equal to coordinate y₂ of opposite corner of said rectangular zone, then geometrical centre of fire source belongs to this zone, and area of fire source is determined by formula: S_fs=μ_p×N_fs, where N_fs is the number of pixels on the image belonging to the mask of the fire source located in the rectangular zone where the geometric centre of the fire source is located.

EFFECT: control of fire hazard in rooms of any category, as well as on objects during field operations, possibility of determining size and location of fire.

1 cl, 6 dwg