A preset unmanned aerial vehicle fire extinguishing method and system for cultural relic protection buildings

Abstract

This invention relates to the intersection of fire protection technology and drone application technology, and provides a pre-positioned drone firefighting method and system for protected cultural heritage buildings. By completing liquid pressurization and automatic docking and delivery to storage tanks within the drone base station, utilizing a visual inertial navigation system to achieve non-contact static hovering relative to the building structure, employing an infrared thermal imaging array for differentiated spraying of the core high-temperature zone and the peripheral smoldering zone, and enabling aerial relay and automatic liquid replenishment between the first and second drones, this invention solves the technical problems of drone drift and collision due to GPS signal loss within protected cultural heritage buildings, uncontrollable water damage from traditional firefighting methods, and the limited liquid capacity of a single drone preventing continuous operation. It offers advantages such as precise positioning without external infrastructure dependence, controllable water damage risk during firefighting, and uninterrupted firefighting operations.

Description

Technical Field

[0001] This application relates to the intersection of fire protection technology and drone application technology, specifically to a pre-positioned drone fire extinguishing method and system for the protection of cultural heritage buildings. Background Technology

[0002] Fire protection of cultural heritage buildings (such as ancient temples, palaces, libraries, and grottoes) is a global challenge. Traditional water-based fire suppression systems are generally prohibited inside these buildings for several reasons: First, the wooden structures, painted murals, ancient books, and silk embroidery within these structures are extremely sensitive to water damage, and the secondary damage caused by the large flow of water from traditional sprinklers or fire hydrants often exceeds the damage caused by the fire itself. Second, cultural heritage protection regulations explicitly prohibit drilling, grooving, or laying pipes on the cultural relics themselves. This "legally and legally" prevents these sites from installing effective automatic water-based fire suppression systems, leaving them reliant on portable fire extinguishers and manual monitoring, which suffers from fatal flaws such as slow response, blind spots, and high risks to personnel safety.

[0003] Existing drone firefighting technologies mainly include three modes: fire extinguishing bomb dropping, external water hose spraying, and tethered water supply. Fire extinguishing bomb dropping has poor accuracy, limited effectiveness, and is uncontrollable, posing an extremely high risk in complex and valuable protected buildings. External water hose spraying involves dragging heavy hoses, resulting in large water flows and heavy water stains, making it completely unsuitable for indoor environments. While tethered water supply allows for continuous operation, the drone is limited by the tether cable, resulting in poor maneuverability and inability to enter complex indoor spaces. Furthermore, all of the above solutions rely on GPS positioning, which is difficult for drones to hover stably in areas with weak GPS signals, such as indoors or under eaves, making them prone to drifting and collisions, causing secondary damage to cultural relics.

[0004] More importantly, existing technologies generally employ a single-drone operation and overall return-to-base mode. The limited liquid capacity of drones makes them unable to cope with sustained fires; furthermore, they lack sophisticated control over water damage, failing to meet the core requirement of water-free fire suppression for cultural heritage preservation. Therefore, the market completely lacks a dedicated fire protection solution for cultural heritage preservation that simultaneously meets the requirements of zero water damage risk, non-contact precise positioning, continuous uninterrupted operation, and compliant installation.

[0005] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0006] The purpose of this application is to provide a pre-positioned drone fire extinguishing method and system for cultural heritage buildings, which has the advantages of enabling non-contact precise hovering and positioning of drones in environments without GPS signals, performing directional differentiated spraying based on temperature-sensing zones to reduce water damage, and forming uninterrupted fire extinguishing operations through multi-drone aerial relay and automatic liquid replenishment.

[0007] Firstly, this application provides a pre-positioned unmanned aerial vehicle (UAV) firefighting method for protected historical buildings, including: A pre-positioned unmanned aerial vehicle (UAV) firefighting method for protected cultural relics buildings, characterized by including: Deploy drone base stations in non-cultural relic areas of protected buildings, wherein the drone base station includes a drone parking platform, a liquid storage container and a pressurization device; After receiving a fire alarm signal, the drone base station controls a first drone carrying a liquid storage tank to fly to the fire source location. The liquid storage tank is filled with pressurized liquid. The first drone hovers at the fire source location and remains stationary relative to the building structure using a visual inertial navigation system, and applies the pressurized liquid to the fire source in a continuous jetting manner; Before the liquid in the first drone's storage tank runs out, the first drone returns to the drone base station to replenish the liquid. At the same time, the second drone takes off from the drone base station and flies to the fire source to take over spraying, forming an uninterrupted fire extinguishing operation.

[0008] Furthermore, the visual inertial navigation system includes a visible light camera and an inertial measurement unit; The method of hovering at the fire source location and maintaining stillness relative to the building structure using a visual inertial navigation system includes: The first UAV acquires image sequences of the building structure surface at a preset frame rate using the visible light camera, and extracts the pixel displacements of feature points in the image sequences; The first UAV detects triaxial acceleration and triaxial angular velocity through the inertial measurement unit; The first UAV calculates a first relative motion estimate based on the pixel displacement of the feature points, calculates a second relative motion estimate based on the integral of the three-axis acceleration and the three-axis angular velocity, fuses the first relative motion estimate and the second relative motion estimate to obtain the real-time position and attitude relative to the building structure, and adjusts the motor speed to maintain the real-time position and attitude constant, so that the first UAV and the building structure remain in a non-contact relative stationary state.

[0009] Furthermore, the step of extracting the feature point pixel displacement in the image sequence includes: performing corner detection on each frame of the image sequence to identify the texture corners of the building structure surface; performing optical flow tracking on the texture corners in adjacent frames to calculate the pixel displacement of the texture corners on the image plane, so as to obtain the feature point pixel displacement; The first UAV calculates a first relative motion estimate based on the pixel displacement of the feature points, including: calculating the translational and rotational motion of the first UAV relative to the building structure based on the pixel displacement, as the first relative motion estimate.

[0010] Furthermore, the step of applying the pressurized liquid to the ignition source in a continuous jetting manner includes: Based on the infrared thermal imaging array carried by the first UAV, thermal imaging data of the fire source area is collected through the infrared thermal imaging array; according to the temperature distribution of the thermal imaging data, the fire source area is divided into a core high-temperature zone and a peripheral smoldering zone; The nozzle is controlled to spray liquid into the core high-temperature zone and the peripheral smoldering zone respectively, wherein the amount of liquid per unit area sprayed into the peripheral smoldering zone is less than the amount of liquid per unit area sprayed into the core high-temperature zone.

[0011] Furthermore, the first UAV controls the nozzle to directionally spray liquid into the core high-temperature zone, including: The center coordinates of the core high-temperature zone are determined based on the thermal imaging data; the fuselage attitude is adjusted so that the spray direction of the nozzle is aligned with the center coordinates; Liquid is injected into the core high-temperature zone at a first injection pressure, and the first injection pressure is positively correlated with the temperature value of the core high-temperature zone.

[0012] Furthermore, the control nozzle sprays liquid into the core high-temperature zone and the peripheral smoldering zone respectively, including: The boundary range of the peripheral smoldering zone is determined based on the thermal imaging data; Liquid is injected into the boundary area at a second injection pressure, which is less than the first injection pressure, so that the amount of liquid injected per unit area into the peripheral smoldering zone is less than the amount of liquid injected per unit area into the core high-temperature zone.

[0013] Furthermore, before the liquid in the first drone's storage tank is depleted, the first drone returns to the drone base station to replenish the liquid. Simultaneously, the second drone takes off from the drone base station and flies to the fire source location to take over the spraying, including: The remaining liquid level in the storage tank of the first drone is detected in real time; when the remaining liquid level reaches a preset threshold, the first drone sends a return-to-home request to the drone base station and continues spraying until the second drone reaches the fire source location; Upon receiving the return-to-home request, the drone base station immediately releases the second drone, which then flies along the flight path of the first drone to the location of the fire source. After the second drone reaches the fire source and begins spraying, the first drone stops spraying and returns to base.

[0014] Furthermore, the second drone flies along the flight path of the first drone to the location of the fire source, including: The first UAV sends the verified flight path data and corresponding historical environmental image features to the UAV base station; The UAV base station loads the flight path data and corresponding historical environmental image features into the second UAV; The second UAV flies according to the flight path data and corrects path deviations during flight using the visual inertial navigation system.

[0015] Furthermore, the correction of path deviation during flight via the visual-inertial navigation system includes: During flight, the second UAV uses the visible light camera to collect real-time images of the current environment and extract current image features. The current image features are matched with the historical environmental image features recorded in the flight path data to identify the degree of deviation between the current position of the second UAV and the flight path; When the deviation exceeds a preset threshold, the second UAV corrects the integral cumulative error of the inertial measurement unit based on the historical pose data corresponding to the historical environmental image features, so that the second UAV returns to the flight path.

[0016] Furthermore, after receiving a fire alarm signal, the drone base station transports the liquid in the storage container to the storage tank of the drone parked on the parking platform, and pressurizes the liquid in the storage tank through the pressurization device.

[0017] Secondly, this invention also proposes a pre-positioned unmanned aerial vehicle (UAV) fire suppression system for protected cultural heritage buildings, comprising: A drone base station is deployed in the non-cultural relic area of ​​a protected building. The drone base station includes a drone parking platform, a liquid storage container, and a pressurization device. The control center is connected to the drone base station and is used to receive fire alarm signals and, in response to the signals, instruct the first drone to fly to the fire source location. The first drone is placed on the parking platform, and its storage tank is filled with liquid supplied by the storage container and pressurized by the pressurization device. The first drone is used to take off after receiving instructions from the control center, fly to the fire source, hover at the fire source and remain stationary relative to the building structure through the onboard visual inertial navigation system, and apply the pressurized liquid to the fire source in a continuous spray manner. The control center is also used to control the first drone to return to the drone base station for replenishment before the liquid in the first drone's storage tank is exhausted, and at the same time control the second drone to take off from the drone base station and fly to the fire source to take over spraying, forming an uninterrupted fire extinguishing operation.

[0018] As can be seen from the above, the pre-positioned UAV fire extinguishing method and system for cultural relic protection buildings provided in this application solves the core contradictions of cultural relic protection buildings, such as the prohibition of water damage, the prohibition of physical contact, the difficulty of stable operation of UAVs in environments without GPS signals, and the insufficient endurance of a single UAV, by deploying UAV base stations in non-cultural relic areas for liquid pressurization and replenishment, using a visual inertial navigation system to achieve non-contact relative static hovering, using an infrared thermal imaging array for temperature-sensing zoned directional spraying, and using multiple UAVs in aerial relay to form uninterrupted fire extinguishing operations. It has the advantages of being able to achieve precise fire extinguishing of cultural relic protection buildings with zero water damage risk, non-intrusive deployment, centimeter-level stable hovering in environments without GPS, and continuous uninterrupted fire extinguishing operations. Attached Figure Description

[0019] The above and other objects, features, and advantages of the present invention will become more apparent from the detailed description of exemplary embodiments with reference to the accompanying drawings. The drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0020] Figure 1 This is a flowchart illustrating the steps of the pre-positioned drone firefighting method for cultural relic protection buildings disclosed in an embodiment of the present invention. Figure 2 This is a schematic diagram of the process disclosed in an embodiment of the present invention, in which a visual inertial navigation system is used to hover at the location of the fire source and remain stationary relative to the building structure. Figure 3 This is a schematic diagram of the process disclosed in this embodiment of the invention, in which the first drone returns to the drone base station to replenish the liquid before the liquid in the storage tank of the first drone is exhausted, and at the same time, the second drone takes off from the drone base station and flies to the fire source location to take over the spraying. Figure 4 This is a schematic diagram of the pre-installed drone fire extinguishing system for cultural relic protection buildings disclosed in an embodiment of the present invention. Detailed Implementation

[0021] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, they are provided so that the invention will be thorough and complete, and the concept of the exemplary embodiments will be fully conveyed to those skilled in the art. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.

[0022] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a full understanding of embodiments of the invention. However, those skilled in the art will recognize that the technical solutions of the invention can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of the invention.

[0023] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0024] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily need to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0025] It should be noted that the first and second drones in this invention are not limited to two drones, nor are the first and second drones limited to one; they can be understood as two batches. Furthermore, the solution of this invention is not limited to just two batches; there can also be a third drone, a fourth drone, etc.

[0026] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of exemplary embodiments, and the modules or processes in the drawings are not necessarily essential for implementing the present invention, and therefore cannot be used to limit the scope of protection of the present invention.

[0027] Firstly, this embodiment proposes a pre-positioned unmanned aerial vehicle (UAV) firefighting method for cultural relic protection buildings, such as... Figure 1 As shown, the method includes: S101, Deploy drone base stations in non-cultural relic areas of protected buildings, wherein the drone base station includes a drone parking platform, a liquid storage container and a pressurization device.

[0028] Specifically, in this embodiment, in S101, a drone base station is deployed in the non-cultural relic area of ​​the protected building. This means the base station is located outside the main body of the protected building, such as in a courtyard corner, outside an auxiliary hall, or inside a dedicated fire-fighting facility room. The distance from the main body of the protected building is typically controlled within 50 to 300 meters, preferably no more than 200 meters. The drone base station includes a drone landing platform, a liquid storage container, and a pressurization device. The landing platform is designed with an automatic centering mechanism and a wireless charging module to ensure accurate drone landing and automatic recharging. The liquid storage container is typically designed with a capacity of 200 to 500 liters and has a built-in liquid level sensor to monitor the liquid level in real time. The pressurization device uses a high-pressure plunger pump with a maximum output pressure of 8 to 12 MPa to meet the requirements for fine water mist spraying.

[0029] In practical applications, the selection of deployment locations must comprehensively consider both cultural relic protection requirements and fire safety coverage. For example, for a Ming Dynasty wooden structure hall covering an area of ​​approximately 800 square meters, a base station can be deployed on each of its east and west sides, with each base station approximately 60 meters away from the main body of the hall. This ensures that drones taking off from either base station can reach any area of ​​the hall within 30 seconds. This deployment method, outside the main building's structure, avoids destructive construction such as drilling holes and installing supports, thus adhering to the principle of minimal intervention in cultural relic protection.

[0030] S102, after receiving the fire alarm signal, the drone base station controls the first drone carrying a liquid storage tank to fly to the fire source location. The liquid storage tank is filled with pressurized liquid. In the most preferred embodiment of the present invention, the liquid is water. In this method, the water can be pre-loaded and pre-pressurized in the outlet tank, or pressurization can be initiated only after a fire alarm is received. This standby state, where the pressure load is relieved, can extend the service life of the storage tank. In this document, when "fine water mist" is mentioned, it can be understood as an embodiment where the extinguishing agent or the liquid is water. However, the present invention does not exclude other liquid extinguishing agents, including protein foam extinguishing agents, fluoroprotein foam extinguishing agents, aqueous film-forming foam extinguishing agents, synthetic foam extinguishing agents, alcohol-resistant foam extinguishing agents, and Class A foam extinguishing agents, etc. In these cases, in S102, after receiving the fire alarm signal, the drone base station delivers the liquid in the storage container to the storage tank of the drone parked on the parking platform, and pressurizes the liquid in the storage tank through a pressurization device. The fire alarm signal can come from smoke detectors, heat detectors, or manual alarm buttons installed in the protected building. After receiving the signal, the base station starts the infusion pump and injects the extinguishing liquid into the drone's storage tank through a quick-connect connector.

[0031] Another implementation method involves loading the fire extinguishing agent onto the first drone while it is in standby mode, and then initiating the loading process on the second drone upon receiving a fire alarm. Specifically, a pressurization device pressurizes the liquid in the storage tank by injecting the high-pressure liquid output from the pressurization device into the drone's storage tank through a high-pressure pipeline, creating a high-pressure environment inside the tank. This way, the storage tank is already pressurized when the drone takes off, eliminating the need for additional battery power to pressurize it in the air.

[0032] The first drone, parked on the platform, has completed liquid filling and pressurization. Its storage tank is filled with fire extinguishing liquid, and the pressure inside the tank has reached the preset working pressure, making it ready for takeoff at any time. At this moment, the base station control unit sends a takeoff command to the first drone, controlling it to fly to the fire source.

[0033] The system controls the first drone to fly to the fire source location. The base station sends the fire source coordinates and flight path data to the first drone, which then autonomously flies to the target area. The base station control unit calculates the specific spatial coordinates of the fire source based on the detector's coded address in the fire alarm signal and a pre-recorded 3D model of the protected historical building. The system then plans a flight path based on these coordinates and the drone's current position. The path planning must fully consider the structural characteristics of the protected historical building, avoiding protruding components such as brackets, eaves, roof ornaments, and coffered ceilings to ensure that the drone does not collide with the building during flight.

[0034] After the flight path planning is completed, the base station control unit transmits the path data to the first UAV via a wireless link. Upon receiving the instructions and path data, the first UAV immediately starts its motors and takes off. During takeoff, the UAV uses its onboard visual-inertial navigation system for real-time positioning and obstacle avoidance.

[0035] S103, the first drone hovers at the fire source location and remains stationary relative to the building structure using a visual inertial navigation system, and applies the pressurized liquid to the fire source in a continuous jetting manner.

[0036] Specifically, in this embodiment, in S103, after the first drone takes off, it can fly towards the fire source location according to the flight path planned in S102 above. During the flight, the visual inertial navigation system carried by the drone collects image data of the building structure surface in real time and fuses it with the data from the inertial measurement unit to achieve accurate positioning in the absence of GPS signals.

[0037] The process of hovering the drone at the fire source location and maintaining its stationary position relative to the building structure using a visual-inertial navigation system involves: a visible light camera on the drone acquiring image sequences of the building structure's surface at a frequency of 30 frames per second, extracting areas with distinct texture features such as brackets, beams, and painted decorations as feature points; tracking the pixel displacement of feature points in adjacent frames using optical flow, combining the three-axis acceleration and three-axis angular velocity detected by the inertial measurement unit, and fusing visual and inertial data using an extended Kalman filter algorithm to calculate the drone's real-time six-degree-of-freedom pose relative to the building structure; and then adjusting the rotational speeds of the four rotors through the flight control system to keep the drone's pose relative to the building structure constant. For example, when the drone hovers near the coffered ceiling of the main hall to extinguish a fire, even with a level 3 wind outside, the visual-inertial navigation system can still control the drone's drift to within 5 centimeters, ensuring that the drone does not collide with the precious coffered ceiling structure.

[0038] Taking water as an example, in one specific embodiment, pressurized liquid is applied to the fire source in a continuous spray manner. Specifically, after the drone reaches a height of approximately 3 to 5 meters above the fire source, it activates its nozzles and sprays downwards in the form of a fine water mist. The diameter of the water mist droplets is typically 50 to 200 micrometers, allowing them to rapidly vaporize on the fire surface, absorbing heat and isolating oxygen. For instance, in the case of an initial fire involving wooden components covering an area of ​​approximately 2 square meters, a single drone carrying 20 liters of pressurized liquid can continuously spray for approximately 2 to 3 minutes, effectively controlling the fire.

[0039] S104, before the liquid in the storage tank of the first drone is exhausted, the first drone returns to the drone base station to replenish the liquid. At the same time, the second drone takes off from the drone base station and flies to the fire source to take over spraying, forming an uninterrupted fire extinguishing operation.

[0040] Specifically, in this embodiment, in S104, before the liquid in the first drone's storage tank is exhausted, the first drone returns to the drone base station to replenish the liquid. At the same time, the second drone takes off from the drone base station and flies to the fire source to take over spraying, forming an uninterrupted firefighting operation. The realization of this step depends on real-time communication and coordinated control between the base station and the drones.

[0041] Specifically, during the spraying process, the first drone monitors the remaining liquid level in real time using an onboard liquid level sensor. When the remaining liquid level drops to a preset threshold, for example, 20% of the total capacity (i.e., 4 liters), the first drone sends a return-to-home request to the base station while continuing spraying. Upon receiving the return-to-home request, the base station immediately launches the second drone, loading the flight path data of the first drone into it. The second drone then flies along the same path to the fire source. To ensure the continuity of the handover process, the flight paths and spraying areas of the two drones need to be precisely coordinated. For example, the first drone returns from the fire source location on the east side of the hall, approximately 80 meters from the base station, with a flight time of about 15 seconds; the second drone takes off from the base station to the same location, with a flight time of about 20 seconds. When the first drone sends the return-to-home request when the remaining liquid level is 20%, the remaining liquid can still support approximately 30 seconds of spraying, sufficient to cover the 5-second time difference before the second drone arrives, ensuring that the fire source is always covered by a fine water mist.

[0042] After the second drone reaches the fire location and begins spraying, the first drone stops spraying and returns along the original path. Upon returning to the base station, the base station automatically replenishes the first drone's fuel via a robotic arm or quick-connect mechanism, checks its battery level, and wirelessly charges it, allowing it to return to standby status within, for example, 5 to 8 minutes, ready to be deployed again as a third or fourth drone. This relay mode ensures the continuity of firefighting operations, which is especially important for deep fires requiring prolonged cooling or suppression.

[0043] This application's solution pre-positions the drone base station in areas outside the cultural relic itself, avoiding the damage caused by installing fixed fire-fighting facilities on the relic building while enabling rapid drone deployment. The application of a visual-inertial navigation system solves the positioning difficulties caused by weak GPS signals and complex structures inside ancient buildings, allowing the drone to safely hover close to the relic structure. The dual-drone relay liquid replenishment mechanism overcomes the bottleneck of limited liquid capacity of a single drone, forming a continuous fire-fighting capability. Based on the organic combination of these technical features, this solution can achieve effective, precise, and continuous fire suppression of fires in relic buildings while protecting the safety of the cultural relic.

[0044] Furthermore, the visual inertial navigation system includes a visible light camera and an inertial measurement unit; like Figure 2 As shown, the method of hovering at the fire source location and maintaining stillness relative to the building structure using a visual inertial navigation system includes: S201, the first UAV acquires an image sequence of the building structure surface at a preset frame rate using the visible light camera, and extracts the feature point pixel displacement in the image sequence; S202, the first UAV detects triaxial acceleration and triaxial angular velocity through the inertial measurement unit; S203, the first UAV calculates a first relative motion estimate based on the pixel displacement of the feature points, calculates a second relative motion estimate based on the integral of the three-axis acceleration and the three-axis angular velocity, fuses the first relative motion estimate and the second relative motion estimate to obtain the real-time position and attitude relative to the building structure, and adjusts the motor speed to maintain the real-time position and attitude constant, so that the first UAV and the building structure remain in a non-contact relative stationary state.

[0045] Specifically, in this embodiment, the first UAV uses a visible light camera to capture image sequences of the building structure surface at a preset frame rate. This industrial-grade visible light camera typically has a frame rate of 30 frames per second (fps) or higher and a resolution of 1280×720 pixels or higher, continuously capturing images of the building structure surface below during flight and hovering. For cultural heritage buildings, their surfaces often possess rich textural information, such as the brackets of wooden structures, the painted beams, and the arrangement of tiles, all of which provide natural feature points for visual navigation. The pixel displacements of feature points in the image sequence are extracted; that is, image processing algorithms identify identical feature points (such as the edges of painted surfaces or the nodes of wood grain) in adjacent frames and calculate the changes in pixel coordinates of these feature points on the image plane. For example, when hovering under the caisson ceiling of a Qing Dynasty hall, the image captured by the drone's camera shows that the pixel coordinates of a corner point of a bracket in the previous frame are (320, 240) and the pixel coordinates in the next frame are (322, 238). Therefore, the pixel displacement of its feature point is (2, -2) pixels.

[0046] The first type of UAV uses an inertial measurement unit (IMU) to detect triaxial acceleration and triaxial angular velocity. Specifically, the IMU module onboard the UAV measures the UAV's linear acceleration and rotational angular velocity in three-dimensional space in real time at a higher frequency (typically 200Hz to 1000Hz). For example, during hovering, if disturbed by a crosswind, the IMU detects an acceleration of 0.2 m / s² in the X-axis direction and an angular velocity of 0.05 rad / s around the Z-axis. These data reflect the UAV's motion state, but due to the IMU's bias and integral accumulation errors, prolonged use alone can lead to significant positional drift.

[0047] The first UAV calculates a first relative motion estimate based on the pixel displacement of feature points. Specifically, it uses visual odometry to calculate the three-dimensional motion of the UAV relative to the building structure by combining the two-dimensional pixel displacement of feature points in the image with the camera's intrinsic parameter matrix and epipolar geometric constraints. In detail, when the UAV displaces relative to the building structure, there is a specific geometric relationship between the pixel displacement of the feature points on the image plane and the actual motion of the UAV. For example, assuming a camera focal length of 4mm, a pixel size of 3μm, and the UAV is 5 meters away from the building structure surface, if the pixel displacement of the feature points is 10 pixels, then according to the pinhole camera model, the actual displacement of the UAV in that direction can be calculated to be approximately (10 pixels × 3μm / pixel × 5 meters) / 4mm = 0.0375 meters, or 3.75 centimeters. This vision-based motion estimation has high accuracy over short periods and does not exhibit cumulative drift, but it may fail when texture is lacking or the movement is too fast.

[0048] The second relative motion estimate is calculated by integrating the triaxial acceleration and triaxial angular velocity. Specifically, the attitude change is obtained by integrating the angular velocity measured by the IMU, the velocity is obtained after gravity compensation of the acceleration, and the position change is obtained by further integration. For example, within a 0.1-second time window, if the X-axis acceleration is measured to be 0.2 m / s² and the initial velocity is 0, the velocity change obtained by integration is 0.02 m / s², and the position change obtained by further integration is 0.001 meters, or 1 millimeter. The advantage of IMU estimation is its high update frequency and short-term smoothness; however, the integration error accumulates over time. For example, after 10 seconds, the position error may reach tens of centimeters.

[0049] In this embodiment, the first and second relative motion estimates are fused. Specifically, algorithms such as Extended Kalman Filter (EKF) or factor graph optimization are used to optimally fuse the low-frequency, high-precision pose estimated by vision with the high-frequency, low-precision pose estimated by IMU. The specific algorithm flow is as follows: IMU measurements are used as the state prediction equation, and visual estimates are used as the observation update equation. Each time new visual data is obtained, the visual estimate is used to correct the error accumulated by the IMU integral. For example, if the IMU integral shows that the drone has moved 5 cm eastward within one second, but visual feature matching analysis shows that the drone has actually only moved 1 cm eastward relative to the building structure, then the Kalman filter will determine that the IMU has produced a 4 cm drift and will correct this drift in the subsequent fusion estimation, ultimately outputting a fused optimal estimate: the drone's current position is offset eastward by 1 cm relative to the building structure.

[0050] After obtaining the real-time position and attitude relative to the building structure, the UAV flight control system calculates the required motor speed adjustment using a PID controller based on the deviation between the fused attitude and the desired hovering attitude (usually the initial attitude recorded at takeoff or the attitude at the moment of arrival at the fire source). For example, if the fused attitude shows that the UAV has deviated 3 centimeters eastward relative to the building structure, the flight control system will appropriately increase the speed of the two left motors and decrease the speed of the two right motors, generating a westward corrective force to return the UAV to the desired position. This process is continuously performed hundreds of times per second, thereby maintaining the relative stillness between the UAV and the building structure.

[0051] In practical applications, for example, in the event of a fire inside a Song Dynasty wooden hall, a drone hovered approximately 4 meters away from the murals inside the hall to extinguish the fire. Since ancient buildings typically lack GPS signals, and the fire scene may be hampered by dense smoke, relying solely on IMU (Inertial Measurement Unit) data would cause the drone to drift and potentially collide with the precious murals. By employing the aforementioned visual-inertial navigation system, the drone uses a camera to collect real-time texture features of the mural surface, fusing this data with IMU data to control hover drift to within 2 centimeters. Even if firefighters move around or water jets cause airflow disturbances, the drone can maintain a stable relative position to the murals, ensuring effective fire extinguishing without causing secondary damage to the artifacts.

[0052] Furthermore, the step of extracting the feature point pixel displacement in the image sequence includes: performing corner detection on each frame of the image sequence to identify the texture corners of the building structure surface; performing optical flow tracking on the texture corners in adjacent frames to calculate the pixel displacement of the texture corners on the image plane, so as to obtain the feature point pixel displacement; The first UAV calculates a first relative motion estimate based on the pixel displacement of the feature points, including: calculating the translational and rotational motion of the first UAV relative to the building structure based on the pixel displacement, as the first relative motion estimate.

[0053] Specifically, in this embodiment, corner detection is performed on each frame of the image sequence. This involves using a feature detection algorithm to select pixels with significant gradient changes as candidate feature points in each frame. Commonly used corner detection algorithms include Harris corner detection, FAST corner detection, or Shi-Tomasi corner detection. Taking the Shi-Tomasi algorithm as an example, its basic principle is to calculate the gray-level change matrix M of each pixel and solve for two eigenvalues ​​λ1 and λ2 of this matrix. When the smaller eigenvalue λmin is greater than a preset threshold, the pixel is identified as a corner. For cultural relics and protected buildings, their surfaces typically have rich natural textures, such as the annual ring nodes of wooden structures, the mortise and tenon edges of brackets, the intersections of painted lines, and the boundaries of tile arrangements. These provide excellent feature sources for corner detection. For example, in a Ming Dynasty palace, a drone camera captured a 1280×720 resolution image of the caisson ceiling. After Shi-Tomasi corner detection, the algorithm selected approximately 150 texture corner points from the image. Each corner point corresponds to a specific physical location, such as the eye of the dragon and phoenix pattern in the center of the caisson ceiling, the tip of a bracket, or the corner of a painted cloud pattern. These corner points have precise pixel coordinates in the image coordinate system; for example, the coordinates of corner point A are (425, 318), and the coordinates of corner point B are (760, 492).

[0054] Optical flow tracking is performed on texture corner points in adjacent frames. This involves searching for the best-matching position of a corner point in the next frame using optical flow methods, thus obtaining the corner point's motion vector on the image plane. Optical flow tracking is based on the assumption of grayscale invariance, meaning the grayscale value of the same spatial point remains constant across different image frames. Commonly used optical flow tracking algorithms include the Lucas-Kanade sparse optical flow method. This algorithm solves the optical flow constraint equations within a local window to obtain the pixel displacements (u, v) of the corner point in the x and y directions. For example, if corner point A in the current frame has coordinates (425, 318), in the next frame, Lucas-Kanade optical flow tracking finds its matching position at (428, 315), then the pixel displacement of corner point A is (3, -3). Similarly, corner point B moves from (760, 492) to (763, 490), with a pixel displacement of (3, -2). After optical flow tracking is completed for all 150 corner points, a sparse optical flow field containing 150 two-dimensional displacement vectors is obtained. In practical applications, due to image noise or illumination changes, tracking of some corner points may fail or produce incorrect matches. In this case, it is necessary to filter out these outliers using Random Sample Consensus (RANSAC) or error culling algorithms, and retain the inliers for subsequent calculations.

[0055] The translational and rotational motions of the first UAV relative to the building structure are calculated based on pixel displacements. This involves recovering the UAV's six-degree-of-freedom motion parameters from a two-dimensional optical flow field based on multi-view geometry principles. Specifically, when the UAV moves in three-dimensional space, there is a definite mathematical relationship between the pixel displacements of feature points in the image and the UAV's motion. Assuming the UAV's translational motion is T = (Tx, Ty, Tz), its rotational motion is R = (ωx, ωy, ωz), the spatial coordinates of the feature point are P = (X, Y, Z), and its projection point on the image plane is p = (x, y), then the relationship between the pixel displacements (u, v) and the UAV's motion can be described by the following optical flow equation: u = ( -Tx + xTz ) / Z + ωx·xy - ωy·(1+x²) + ωz·y v = ( -Ty + yTz ) / Z + ωx·(1+y²) - ωy·xy - ωz·x This is a system of equations containing multiple unknowns. By using observations from multiple feature points, the motion parameters of the UAV can be jointly solved. (u, v) represents the pixel displacement (i.e., optical flow) of the feature point on the image plane; u represents the displacement of the feature point in the x-direction (horizontal direction) of the image; v represents the displacement of the feature point in the y-direction (vertical direction) of the image. (x, y) represents the coordinates of the feature point on the normalized image plane, indicating the projection component of the unit direction vector from the camera's optical center to the feature point in the camera coordinate system; xy is the product of these two coordinate values. Z is the depth value of the feature point relative to the camera coordinate system (i.e., the UAV); when the UAV hovers close to an ancient building (e.g., 5 meters from a coffered ceiling), Z is approximately the actual distance from the UAV to the feature point, for example, 5 meters. Tx, Ty, and Tz are the translational motion components of the UAV (corresponding to lateral, longitudinal, and axial translation, respectively). ωx, ωy, and ωz are the rotational motion components (angular velocities) of the UAV, representing the rotational angular velocities of the UAV around the three coordinate axes; ωx is the rotation around the X-axis (pitch angle change), ωy is the rotation around the Y-axis (yaw angle change), and ωz is the rotation around the Z-axis (roll angle change).

[0056] In practical calculations, the essential matrix decomposition method or the direct linear transformation method are usually used. Taking the essential matrix method as an example, firstly, based on the pixel coordinates and pixel displacements of the feature points, combined with the camera's intrinsic parameter matrix, the coordinates of the feature points on the normalized plane are calculated; then, the essential matrix E is estimated using the five-point method or the eight-point method; finally, singular value decomposition is performed on the essential matrix to decompose it into the rotation matrix R and the translation vector T. For example, in a set of simulated data, using the optical flow tracking results of 80 effective feature points, the calculated drone motion is as follows: translation vector T = (0.05 m, -0.02 m, 0.03 m), indicating that the drone moved 5 cm to the right, 2 cm upward, and 3 cm forward relative to the building structure; rotation angle R = (1.2°, -0.5°, 0.8°), indicating that the drone rotated 1.2 degrees around the X-axis, -0.5 degrees around the Y-axis, and 0.8 degrees around the Z-axis. This calculation result is used as the first relative motion estimate for subsequent fusion with IMU data.

[0057] Specifically, in some of the above embodiments, the step of calculating the first relative motion estimate based on pixel displacement can be further refined. When the UAV is close to the surface of the building structure, the change in scene depth Z cannot be ignored. At this time, it is necessary to adopt a method based on the essential matrix and triangulation to estimate the depth information of feature points while solving for motion. The specific process is as follows: First, the initial rotation matrix R and translation vector T are obtained through essential matrix decomposition; then, based on R and T, the three-dimensional spatial coordinates of each feature point are calculated using the triangulation method; next, the calculated three-dimensional coordinates are reprojected onto the image plane to obtain the reprojection error; finally, bundle adjustment (BA) is used to jointly optimize R, T and the three-dimensional coordinates of the feature points to minimize the reprojection error. For example, when hovering at a distance of about 3 meters from the surface of the coffered ceiling, the above optimization can control the reprojection error to within 0.5 pixels, and the corresponding spatial position error is about 2 millimeters, which significantly improves the accuracy of motion estimation.

[0058] Furthermore, the step of applying the pressurized liquid to the ignition source in a continuous jetting manner includes: Based on the infrared thermal imaging array carried by the first UAV, thermal imaging data of the fire source area is collected through the infrared thermal imaging array; according to the temperature distribution of the thermal imaging data, the fire source area is divided into a core high-temperature zone and a peripheral smoldering zone; The nozzle is controlled to spray liquid into the core high-temperature zone and the peripheral smoldering zone respectively; wherein the amount of liquid per unit area sprayed into the peripheral smoldering zone is less than the amount of liquid per unit area sprayed into the core high-temperature zone.

[0059] Specifically, in this embodiment, thermal imaging data of the fire source area is collected based on an infrared thermal imaging array mounted on a first UAV. This array is an uncooled infrared focal plane array detector installed beneath the UAV, typically with a resolution of 640×512 pixels or higher and a temperature sensitivity better than 0.05 degrees Celsius. It can penetrate smoke in dense smoke environments to obtain the temperature distribution on the surface of the fire source. The infrared thermal imaging array collects data at a rate of 30 frames per second, with each pixel corresponding to a temperature value, forming a two-dimensional temperature distribution cloud map. For example, in the event of a fire involving wooden components in a Qing Dynasty palace, the UAV hovers approximately 5 meters above the fire source. The data collected by the infrared thermal imaging array shows pixel temperatures in the flame area ranging from 580 to 620 degrees Celsius, surface temperatures of the burning wooden components ranging from 320 to 450 degrees Celsius, and surface temperatures of the surrounding unburned wooden structures ranging from approximately 25 to 40 degrees Celsius.

[0060] Based on the temperature distribution of thermal imaging data, the fire source area is divided into a core high-temperature zone and a peripheral smoldering zone. This is achieved through a temperature threshold segmentation algorithm to partition the thermal imaging image. Specifically, two temperature thresholds can be set: a high-temperature threshold (Thigh) and a low-temperature threshold (Tlow). For example, based on the ignition point of wood (approximately 300 degrees Celsius) and smoldering temperature (approximately 150 degrees Celsius), Thigh is set to 300 degrees Celsius and Tlow to 150 degrees Celsius. Pixel areas in the thermal imaging image with temperatures above 300 degrees Celsius are classified as the core high-temperature zone, pixel areas with temperatures between 150 and 300 degrees Celsius are classified as the peripheral smoldering zone, and areas with temperatures below 150 degrees Celsius are considered normal background areas. Assuming that in the thermal imaging data of a fire, there are 1250 pixels with temperatures above 300 degrees Celsius, the actual physical area corresponding to each pixel is calculated using the field of view and object distance of the optical system. Taking a drone hovering at a height of 5 meters and a camera field of view of 60 degrees as an example, the ground area covered by a single frame of image is approximately 5.8 meters × 4.3 meters, and the actual area corresponding to a single pixel is approximately 0.9 centimeters × 0.9 centimeters. Based on this calculation, the actual area of ​​the core high-temperature zone is approximately 1250 × 0.9 centimeters × 0.9 centimeters ≈ 0.1 square meters, while the outer smoldering zone, with a temperature between 150 and 300 degrees Celsius, has a total of 3800 pixels, with an actual area of ​​approximately 0.31 square meters.

[0061] The drone controls the nozzles to spray liquid into the core high-temperature zone and the outer smoldering zone separately. This means the drone dynamically adjusts its spraying strategy based on the zone division results. For the core high-temperature zone, the nozzles directly cover it with higher spray pressure and flow rate to quickly reduce the temperature of the fire source. For the outer smoldering zone, the nozzles cover it with lower spray pressure and flow rate, primarily to prevent reignition or spread of the smoldering area, while avoiding excessive water spraying that could damage the artifacts. For example, the dual-fluid nozzles on the drone can achieve stepless adjustment of the spray pressure, ranging from 2 MPa to 10 MPa. In the core high-temperature zone, the nozzles spray at 8 MPa, with droplet diameters of approximately 50 to 80 micrometers, effectively penetrating the flame plume to reach the burning surface. In the outer smoldering zone, the nozzles spray at 3 MPa, with droplet diameters of approximately 150 to 200 micrometers, settling at a lower velocity on the surface of the smoldering area. This moistens the wooden structure surface to prevent reignition without causing excessive impact that could peel off the painted surfaces of the artifacts.

[0062] The amount of liquid sprayed per unit area to the outer smoldering zone is less than that sprayed per unit area to the core high-temperature zone. This means that differentiated liquid application rates are achieved for different areas through precise control of the spray flow rate and spray time. Specifically, the liquid flow rate per unit time can be controlled by adjusting the pulse width modulation (PWM) duty cycle of the nozzle or by changing the spray pressure. For example, if the core high-temperature zone area is 0.1 square meters and the target application rate is set at 5 liters / square meter, then 0.5 liters of liquid need to be sprayed; if the outer smoldering zone area is 0.31 square meters and the target application rate is set at 1.5 liters / square meter, then 0.465 liters of liquid need to be sprayed. Based on the area of ​​each region and the target application rate, the drone calculates the required spraying time (Thigh) for the core high-temperature zone and the required spraying time (Tlow) for the outer smoldering zone. Assuming the nozzle's flow rate is 0.3 L / s at 8 MPa pressure and 0.15 L / s at 3 MPa pressure, then Thigh = 0.5 L / 0.3 L / s ≈ 1.67 s, and Tlow = 0.465 L / 0.15 L / s ≈ 3.1 s. When performing the spraying mission, the UAV first aims the nozzle at the core high-temperature zone and sprays continuously at 8 MPa pressure for 1.67 seconds. Then, the nozzle is adjusted to the outer smoldering zone and sprays continuously at 3 MPa pressure for 3.1 seconds.

[0063] Furthermore, the first UAV controls the nozzle to directionally spray liquid into the core high-temperature zone, including: The center coordinates of the core high-temperature zone are determined based on the thermal imaging data; the fuselage attitude is adjusted so that the spray direction of the nozzle is aligned with the center coordinates; Liquid is injected into the core high-temperature zone at a first injection pressure, and the first injection pressure is positively correlated with the temperature value of the core high-temperature zone.

[0064] Specifically, in this embodiment, the center coordinates of the core high-temperature zone are determined based on thermal imaging data. This involves extracting all pixels classified as belonging to the core high-temperature zone from the temperature distribution matrix acquired by the infrared thermal imaging array, and calculating the geometric center or temperature centroid of this region using a weighted average algorithm. Common calculation methods include the arithmetic mean centroid method and the temperature-weighted centroid method. The temperature-weighted centroid method uses the temperature value of each pixel as a weight to perform a weighted average of the pixel coordinates, resulting in center coordinates that better reflect the location where the fire source energy is most concentrated. The calculation formulas are: xcenter = Σ(Ti × xi) / ΣTi, ycenter = Σ(Ti × yi) / ΣTi, where Ti is the temperature value of the i-th high-temperature pixel, and (xi, yi) are the coordinates of that pixel in the image coordinate system. For example, in a fire scene, the core high-temperature zone contains 156 pixels, with the highest temperature point located at (320, 240) at 612 degrees Celsius, and the temperatures of the remaining pixels distributed between 480 and 590 degrees Celsius. The center coordinates, calculated using temperature weighting, are (325.6, 238.4). This location is slightly biased towards a warmer region, rather than a simple geometric center. These center coordinates are then transformed from the image coordinate system to the drone's body coordinate system. This transformation process takes into account the camera's mounting position and angle, as well as the drone's current attitude angle. Assuming the camera is mounted on the drone's gimbal and always points vertically downwards, the image center corresponds to the point directly below the drone. The geographic orientation of the target center relative to the drone can then be calculated using pixel offset.

[0065] Adjusting the drone's attitude aligns the nozzle's spray direction with the center coordinates. Specifically, the drone's flight control system achieves precise alignment by adjusting yaw and pitch angles based on the deviation between the target center and the current nozzle direction. In practice, the nozzle is typically fixedly mounted below the drone's fuselage, with its spray direction parallel to or at a fixed angle to the drone's vertical axis. Therefore, aligning with the target requires adjusting the drone's horizontal position and attitude angles. Assuming the coordinates of the point directly below the drone's current hovering position are (X0, Y0, Z0), and the ground projection coordinates of the target's center point are (Xt, Yt), then the drone needs to move the following horizontal distances: ΔX = Xt - X0, ΔY = Yt - Y0. If the drone has omnidirectional movement capabilities, it can directly align the nozzle with the target through horizontal displacement; however, in actual firefighting scenarios, the drone typically hovers, and the nozzle axis is aligned with the target by adjusting the yaw and pitch angles. For example, if the target center is located 2.3 meters in front of the UAV and 1.5 meters to the right, with a hovering height of 5 meters, then the horizontal distance is √((2.3²+1.5²)) = 2.74 meters. The pitch angle adjustment θpitch = arctan(2.74 / 5) = 28.7 degrees, and the yaw angle adjustment θyaw = arctan(1.5 / 2.3) = 33.1 degrees. The UAV flight control system tilts the fuselage forward by 28.7 degrees by increasing the speed of the two rear motors and decreasing the speed of the two front motors, while simultaneously adjusting the differential speed of the left and right motors to yaw the fuselage to the right by 33.1 degrees, ensuring that the nozzle axis is precisely pointed to the target center. The entire process is completed automatically by the flight control system, with a response time typically within 0.2 to 0.5 seconds.

[0066] Liquid is injected into the core high-temperature zone at a first injection pressure. This first injection pressure is positively correlated with the temperature of the core high-temperature zone; that is, the injection pressure is dynamically adjusted according to the fire source temperature. The higher the temperature, the greater the injection pressure, ensuring that the fine water mist can penetrate the high-temperature flame plume and reach the combustion surface. In specific implementation, a temperature-pressure mapping relationship can be established, such as a linear mapping or a piecewise mapping. A linear mapping can be expressed as P = k × T + b, where P is the injection pressure, T is the highest or average temperature of the core high-temperature zone, and k and b are preset coefficients. Taking a fire in the beam frame of a Song Dynasty palace as an example, the highest temperature of the core high-temperature zone was measured to be 586 degrees Celsius by infrared thermal imaging. According to the preset mapping relationship: the reference temperature Tbase = 300 degrees Celsius corresponds to the reference pressure Pbase = 6 MPa. For every 100 degrees Celsius increase in temperature, the pressure increases by 2 MPa. Therefore, the pressure calculation formula is P = 6 + 2 × (Tmax - 300) / 100. Substituting Tmax = 586, we calculate P = 6 + 2 × (286 / 100) = 6 + 2 × 2.86 = 6 + 5.72 = 11.72 MPa. In actual control, considering the system's safety upper limit, the pressure may be limited to the range of 10 MPa to 12 MPa. Therefore, we take 11.7 MPa as the actual injection pressure.

[0067] The physical basis for the positive correlation between pressure and temperature lies in the fact that the higher the flame temperature, the faster the upward velocity of the hot plume and the greater the turbulence intensity, resulting in a stronger carrying and evaporation effect on the water mist. Water mist at lower pressure may be entrained by the hot plume or completely vaporized before reaching the combustion surface, failing to effectively cool the fire source. By increasing the injection pressure, the initial velocity of the water mist increases, and the droplet diameter decreases, allowing it to penetrate high-temperature areas and directly act on the combustion surface. For example, in laboratory tests, for a wood-based fire source at approximately 300 degrees Celsius, a fine water mist at 6 MPa pressure has a penetration rate of about 75%; for a fire source at approximately 600 degrees Celsius, the penetration rate drops to below 40% at the same pressure; while increasing the pressure to 12 MPa restores the penetration rate to over 70%.

[0068] In practical applications, after hovering and aiming at the core high-temperature area, the drone dynamically adjusts the spray pressure based on the real-time detected highest temperature. Assuming the fire source temperature changes during spraying, for example, the initial temperature is 586 degrees Celsius, and after 3 seconds of spraying, it drops to 420 degrees Celsius, then the drone adjusts the pressure in real time to P = 6 + 2 × (420-300) / 100 = 6 + 2 × 1.2 = 8.4 MPa. This dynamic adjustment ensures fire extinguishing efficiency while avoiding unnecessary water mist diffusion and impact on cultural relics caused by excessive pressure.

[0069] Furthermore, the control nozzle sprays liquid into the core high-temperature zone and the peripheral smoldering zone respectively, including: The boundary range of the peripheral smoldering zone is determined based on the thermal imaging data; Liquid is injected into the boundary area at a second injection pressure, which is less than the first injection pressure, so that the amount of liquid injected per unit area into the peripheral smoldering zone is less than the amount of liquid injected per unit area into the core high-temperature zone.

[0070] Specifically, in this embodiment, the boundary range of the peripheral smoldering zone is determined based on thermal imaging data. That is, all pixels with temperatures between the smoldering threshold and the high-temperature threshold are extracted from the temperature distribution matrix acquired by the infrared thermal imaging array, and the contour boundaries of the connected regions formed by these pixels are determined using image processing algorithms. In the aforementioned embodiment, the high-temperature threshold Thigh is set to 300 degrees Celsius, and the smoldering threshold Tlow is set to 150 degrees Celsius. All pixels in the thermal imaging image with temperatures between 150 degrees Celsius and 300 degrees Celsius constitute one or more connected regions, which are the peripheral smoldering zones. The boundary range is typically determined using edge detection algorithms, such as Canny edge detection or the Sobel operator, to extract the outer contour of the smoldering zone. For example, in an infrared thermal image of a fire in a Qing Dynasty palace, the core high-temperature zone was located at the bottom of the beam frame, appearing as an approximately elliptical area with an area of ​​about 0.1 square meters. The smoldering zone surrounding the core high-temperature zone appeared as an irregular ring-shaped area, with a maximum width of about 0.8 meters and a minimum width of about 0.3 meters. After pixel counting and area conversion, the total area was 0.31 square meters. The precise coordinate data of the boundary range was stored in the UAV's onboard computer for subsequent spray path planning. For multiple unconnected smoldering zones, the system recorded the vertex coordinates of the boundary polygon of each zone separately.

[0071] Liquid is sprayed into the boundary area at a second spray pressure, lower than the first spray pressure. This means the drone uses a lower pressure than the core high-temperature zone spray pressure for a comprehensive coverage spray, based on the boundary of the outer smoldering zone. The setting of the second spray pressure must comprehensively consider the temperature distribution of the smoldering zone, the material properties of the artifact's surface, and the settling characteristics of the water mist. For wooden structures, excessively high spray pressure may result in excessive water mist impact, causing paint peeling or damage to wood fibers; excessively low pressure may prevent the water mist from effectively covering the smoldering area. According to experimental data, when the spray pressure is in the range of 2 MPa to 4 MPa, the droplet diameter is approximately 150 to 250 micrometers, with a moderate settling velocity, effectively wetting the wooden structure surface while preventing runoff. For example, in the aforementioned hall fire case, the first spray pressure used in the core high-temperature zone was 11.7 MPa, while the second spray pressure in the outer smoldering zone was set at 3.2 MPa. During spraying, the drone follows a preset scanning path, aiming the nozzle at the area within the boundary of the smoldering zone and spraying evenly at a pressure of 3.2 MPa. The scanning path typically employs a reciprocating or spiral pattern to ensure that every surface within the smoldering zone is covered. Assuming the nozzle's spray coverage diameter is 1.2 meters at a pressure of 3.2 MPa, and the drone hovers at a height of 5 meters, a single hover can cover an area of ​​approximately 1.1 square meters, sufficient to cover the entire 0.31 square meter smoldering zone. Therefore, the drone only needs to adjust its hovering position once to complete the spraying of the entire smoldering zone.

[0072] To ensure that the amount of liquid sprayed per unit area to the outer smoldering zone is less than that sprayed per unit area to the core high-temperature zone, differential liquid application is achieved through pressure control and spraying time control. The formula for calculating the amount of liquid per unit area is: Liquid application per unit area = Nozzle flow rate × Spraying time / Coverage area. With a relatively fixed coverage area, the main means of controlling the liquid application is to adjust the nozzle flow rate and spraying time. The nozzle flow rate is positively correlated with the spraying pressure. For example, a certain type of fine water mist nozzle has a flow rate of 0.16 liters / second at 3.2 MPa pressure and 0.35 liters / second at 11.7 MPa pressure. For the outer smoldering zone, the target liquid application is usually set to 30% to 50% of that of the core high-temperature zone. Taking a target liquid application of 5 liters / m² for the core high-temperature zone as an example, the target liquid application for the outer smoldering zone can be set to 1.5 liters / m². For an outer smoldering zone with an area of ​​0.31 m², the required total liquid amount is 0.31 × 1.5 = 0.465 liters. At a pressure of 3.2 MPa and a nozzle flow rate of 0.16 L / s, the required spraying time is approximately 0.465 / 0.16 ≈ 2.91 seconds. During the spraying mission, the UAV, following the planned scanning path, completes uniform coverage of the entire smoldering zone within 2.91 seconds. In contrast, the core high-temperature zone, with an area of ​​0.1 square meters and a target application rate of 5 L / m², requires a total liquid volume of 0.5 L. At a pressure of 11.7 MPa and a flow rate of 0.35 L / s, the spraying time is only approximately 0.5 / 0.35 ≈ 1.43 seconds. This comparison shows that the application rate per unit area in the outer smoldering zone (1.5 L / m²) is indeed less than that in the core high-temperature zone (5 L / m²), achieving differentiated application.

[0073] Specifically, in some of the above embodiments, the spraying method for the peripheral smoldering zone can be further refined. For peripheral smoldering zones with large areas or complex shapes, the UAV can use a multi-point hovering scanning method for coverage. The system first grids the boundary of the smoldering zone, dividing it into several sub-regions with diameters equivalent to the nozzle coverage diameter. Then, it plans the optimal hovering path and sequentially completes local spraying at each hovering point. The spraying time at each hovering point is dynamically adjusted according to the actual area of ​​the sub-region to ensure the consistency of the overall liquid application per unit area. For example, for a narrow smoldering strip with uneven width, it can be divided into 3 hovering points, with spraying times of 1.2 seconds, 0.8 seconds, and 1.0 seconds for each point, so that the liquid application rate of the entire smoldering strip is uniformly controlled at about 1.5 liters / square meter.

[0074] Furthermore, such as Figure 3 As shown, before the liquid in the first drone's storage tank is depleted, the first drone returns to the drone base station to replenish the liquid. Simultaneously, the second drone takes off from the drone base station and flies to the fire source location to take over the spraying, including: S301, Real-time detection of the remaining liquid level in the storage tank of the first drone; When the remaining liquid level reaches a preset threshold, the first drone sends a return-to-home request to the drone base station and continues spraying until the second drone reaches the fire source location; S302, after receiving the return-to-home request, the drone base station immediately releases the second drone, and the second drone flies along the flight path of the first drone to the fire source location; S303, after the second drone reaches the fire source location and begins spraying, the first drone stops spraying and returns to base.

[0075] Specifically, in this embodiment, the liquid level sensor monitors the liquid level in the storage tank 20 times per second, with a measurement accuracy of ±1%. The preset threshold setting needs to comprehensively consider the return time, the arrival time of the second drone, and the safety overlap. Taking a large hall as an example, the distance from the first drone returning from the fire scene to the base station is 80 meters, with a flight speed of 8 meters per second, taking 10 seconds, plus take-off and landing adjustments, totaling 20 seconds; the second drone takes 15 seconds (including 5 seconds of take-off) to take off from the base station along the same path to the fire scene. Setting a safety overlap time of 5 seconds means that 20 seconds need to be covered from the time the return request is sent to the time the second drone starts spraying. The nozzle flow rate is 0.3 liters / second, and 6 liters are consumed in 20 seconds. After adding a 30% safety factor, the preset threshold is set to 8 liters. When the remaining liquid volume drops to 8 liters, the first drone sends a return request, at which point it can continue spraying for about 26 seconds. After receiving the request, the base station immediately starts the second drone and loads the flight path data uploaded by the first drone into the second drone. The second drone flies along the same path and arrives at the fire scene 14 seconds later. After the second drone began spraying, the first drone stopped spraying and returned to base, with approximately 2.3 liters of liquid remaining. The fire source remained under a fine water mist cover throughout. This mechanism, through precise threshold setting and path reuse, achieved uninterrupted firefighting operations. Furthermore, the second drone flies to the fire source location along the flight path of the first drone, including: the first drone sending verified flight path data and corresponding historical environmental image features to the drone base station; the drone base station loading the flight path data and corresponding historical environmental image features into the second drone; the second drone flying according to the flight path data, and correcting path deviations during flight using the visual inertial navigation system.

[0076] Furthermore, the step of correcting path deviation through the visual inertial navigation system during flight includes: during flight, the second UAV acquires current environmental images in real time through the visible light camera and extracts current image features; the current image features are matched with historical environmental image features recorded in the flight path data to identify the degree of deviation between the current position of the second UAV and the flight path; when the degree of deviation exceeds a preset threshold, the second UAV corrects the integral cumulative error of the inertial measurement unit based on the historical pose data corresponding to the historical environmental image features, so that the second UAV returns to the flight path.

[0077] Specifically, in this embodiment, during its initial flight to the fire source location, the first UAV records its flight path data in real time using a visual inertial navigation system. This includes the three-dimensional coordinates of each path point, the UAV's attitude angle, and the surface image features of the building structure acquired at the corresponding time. For example, the first UAV takes off from the base station and flies approximately 95 meters to the fire source in the central hall's caisson ceiling, taking about 12 seconds. During the flight, the visual inertial navigation system acquires images at 30 frames per second and extracts feature points from each frame, resulting in 360 frames. Each frame contains approximately 150 to 200 feature points. These feature points correspond to the unique textures of cultural relics and buildings, such as the edges of the brackets, painted patterns, and the tenons of the beams. The first UAV then packages and sends the verified flight path data—including the path point coordinate sequence (a total of 120 path points, spaced approximately 0.8 meters apart), the UAV's attitude angles (yaw, pitch, and roll) corresponding to each path point, and the keyframe image and its feature point descriptor at each path point—to the UAV base station. The data volume is approximately 35 megabytes.

[0078] After receiving the aforementioned data, the UAV base station transmits it to the second UAV via a high-speed data transmission link. Upon receiving the return-to-home command, the second UAV immediately loads this data and prepares for takeoff. The loading process includes importing the flight path point sequence into the flight control system's navigation module and storing keyframe image features in the visual navigation database. The entire data loading process is completed within 2 seconds, without affecting the rapid deployment of the second UAV.

[0079] The second UAV flies based on flight path data and corrects path deviations during flight using a visual-inertial navigation system. After takeoff, the second UAV flies according to the path point sequence recorded by the first UAV. However, due to slight differences in takeoff position and wind disturbances, the actual flight trajectory of the second UAV may deviate from the preset path. The visual-inertial navigation system corrects this through real-time image matching. For example, when the second UAV flies to about 30 meters from the base station, the visible light camera captures an image of the current environment and extracts 152 feature points. The onboard computer matches these current feature points with historical environmental image features recorded by the first UAV in the same area. The matching algorithm uses Fast Nearest Neighbor (FLANN) and successfully matches 138 feature points. By calculating the pixel coordinate differences of the matched feature points on the image plane and combining this with essential matrix factorization, the pose deviation of the second UAV's current position relative to the first UAV at that path point is calculated. The calculation shows that the actual position of the second UAV deviates 0.32 meters eastward, 0.18 meters southward, and 0.12 meters lower in altitude than the preset path point, with a yaw angle deviation of 2.3 degrees.

[0080] When the deviation exceeds a preset threshold, the second UAV corrects the cumulative error of the inertial measurement unit (IMU) based on historical pose data corresponding to historical environmental image features, allowing the second UAV to return to the flight path. The preset thresholds are typically set based on the safe distance to historical buildings, such as a horizontal offset threshold of 0.2 meters, an altitude offset threshold of 0.1 meters, and an attitude angle deviation threshold of 2.0 degrees. In the example above, the horizontal offset of 0.32 meters exceeds the 0.2-meter threshold, thus triggering path correction. The correction process includes: first, using the three-dimensional spatial coordinates corresponding to the matched historical image feature points as a reference, calculating the desired pose of the second UAV using the perspective n-point (PnP) algorithm; then, fusing this desired pose with the position obtained from the current IMU integration, updating the state estimate through extended Kalman filtering, and correcting the IMU's cumulative error; finally, the flight control system adjusts the motor speed based on the deviation between the corrected pose and the desired pose, allowing the UAV to return to the preset path. For example, after one correction, the horizontal offset of the second UAV at the next path point is reduced to 0.08 meters, and the altitude offset to 0.03 meters, both within the threshold. The entire correction process took about 0.3 seconds. The drone flew at a speed of 8 meters per second and only flew forward about 2.4 meters during the process, and the path deviation was corrected in time.

[0081] In a practical application, a fire broke out in the east annex of a main hall. The first drone took off from the base station and flew to the fire scene, traversing complex areas such as the north side of the bell tower, the south eaves of the main hall, and the gap in the gable wall of the west annex. The flight path needed to precisely avoid protruding objects such as brackets and roof ornaments. The first drone uploaded 120 path points and corresponding 360 frames of image features recorded throughout the flight to the base station. When the second drone took over, it loaded this data and flew along the same path. When it reached the south eaves of the main hall, a crosswind caused the second drone to deviate outward by 0.27 meters, approaching the brackets at the edge of the eaves. The visual inertial navigation system detected the deviation through real-time image matching and immediately triggered a correction, pulling the drone back to a safe path, successfully avoiding the brackets, and finally safely reaching the fire scene. During the entire flight, the path correction was triggered 4 times, and the deviation was controlled within 0.1 meters after each correction, ensuring the safety of the cultural relics and the continuity of the mission.

[0082] The proposed solution achieves the reuse of safe flight paths by transmitting verified flight path data and historical environmental image features from a first UAV to a second UAV. During flight, the second UAV accurately identifies deviations between its position and the preset path through real-time image matching and historical feature comparison, and promptly corrects the integrated accumulation error of the IMU. It is precisely this visual feature-based path tracking and correction mechanism that enables the second UAV to accurately reproduce the safe flight trajectory of the first UAV even under wind interference or differences in takeoff position, avoiding the time consumption of replanning the path and ensuring flight safety in complex ancient architectural environments.

[0083] Furthermore, after receiving a fire alarm signal, the drone base station transports the liquid in the storage container to the storage tank of the drone parked on the parking platform, and pressurizes the liquid in the storage tank through the pressurization device.

[0084] Specifically, in this embodiment, the base station control unit receives alarm information from smoke detectors, heat detectors, manual alarm buttons, or linked monitoring systems installed within the protected historical building via wired or wireless communication. The fire alarm signal must at least contain the coded address of the area where the fire source is located or the specific detection point, for example, "Smoke detector No. 03 in the beam frame of the east secondary bay of a certain main hall." Upon receiving the signal, the base station control unit immediately confirms its validity and triggers the emergency response procedure.

[0085] The liquid in the storage container is transferred to the storage tank of the drone, which is parked on the landing platform. The base station activates the infusion pump, filling the drone's storage tank with fire extinguishing liquid through pipes and quick-connect fittings. When the drone is parked on the platform, the one-way valve quick-connect interface at the bottom of its storage tank automatically connects to the ground-based infusion pipeline. After the base station control unit detects that the connection is normal, it activates the infusion pump and the electric valve. A flow meter and pressure sensor are installed on the infusion pump outlet pipeline to monitor the infusion status in real time. During the filling process, the drone's onboard liquid level sensor collects liquid level data in real time and transmits it back to the base station control unit via a wireless link. When the liquid level reaches the preset capacity, the control unit issues a command to shut down the infusion pump and the electric valve, stopping the filling. A small margin is left to prevent abnormal pressure increases inside the tank due to thermal expansion of the liquid.

[0086] The liquid in the storage tank is pressurized by a pressurization device. This means that after filling, pressure is applied to the liquid in the drone's storage tank to reach the working pressure required for fine water mist spraying. The pressurization device automatically connects to the pressurization interface on the top of the drone's storage tank via a high-pressure pipeline. After filling stops, the base station control unit activates the pressurization device, injecting high-pressure liquid back into the storage tank, applying pressure to the liquid through compressed air or by directly applying pressure to the liquid surface. A pressure sensor inside the storage tank monitors the internal pressure in real time. When the pressure reaches the preset working pressure, the control unit issues a command to stop pressurization and closes the valve at the pressurization interface, maintaining a constant pressure inside the storage tank.

[0087] In this embodiment, the entire injection and pressurization process ensures that the drone is fully pressurized before takeoff. During this process, injection and pressurization are performed in parallel with tasks such as fire information processing and flight path planning, without causing additional time delays. The working pressure setting needs to comprehensively consider the building height and the characteristics of the fire target to ensure that the fine water mist can effectively cover the fire source area.

[0088] Secondly, this application also proposes a pre-installed unmanned aerial vehicle (UAV) fine water mist fire suppression system for protected cultural heritage buildings, such as... Figure 4 As shown, it includes: The drone base station 401 is deployed in the non-cultural relic area of ​​the protected building. The drone base station includes a drone parking platform, a liquid storage container and a pressurization device. The control center 402 is communicatively connected to the drone base station and is used to receive fire alarm signals and respond to the signals to instruct the first drone to fly to the fire source location. The first drone 403 is parked on the parking platform. Its storage tank contains liquid supplied by the storage container and pressurized by the pressurization device. The first drone is used to take off after receiving instructions from the control center, fly to the fire source, hover at the fire source and remain stationary relative to the building structure through the onboard visual inertial navigation system, and apply the pressurized liquid to the fire source in a continuous spray manner. The control center 402 is also used to control the first drone to return to the drone base station for replenishment before the liquid in the first drone's storage tank is exhausted, and at the same time control the second drone to take off from the drone base station and fly to the fire source to take over spraying, forming an uninterrupted fire extinguishing operation.

[0089] This system can be used to implement the pre-positioned drone firefighting method for cultural heritage buildings described in the first aspect, which will not be elaborated further here.

[0090] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A pre-positioned unmanned aerial vehicle (UAV) firefighting method for protected cultural relics buildings, characterized in that, include: Deploy drone base stations in non-cultural relic areas of protected buildings, wherein the drone base station includes a drone parking platform, a liquid storage container and a pressurization device; After receiving a fire alarm signal, the drone base station controls the first drone carrying a liquid storage tank to fly to the fire source location. The liquid storage tank is filled with pressurized liquid. The first drone hovers at the fire source location and remains stationary relative to the building structure using a visual inertial navigation system, and applies the pressurized liquid to the fire source in a continuous jetting manner; Before the liquid in the first drone's storage tank runs out, the first drone returns to the drone base station to replenish the liquid. At the same time, the second drone takes off from the drone base station and flies to the fire source to take over spraying, forming an uninterrupted fire extinguishing operation.

2. The pre-positioned unmanned aerial vehicle (UAV) firefighting method for cultural relic protection buildings according to claim 1, characterized in that, The visual inertial navigation system includes a visible light camera and an inertial measurement unit; The method of hovering at the fire source location and maintaining stillness relative to the building structure using a visual inertial navigation system includes: The first UAV acquires image sequences of the building structure surface at a preset frame rate using the visible light camera, and extracts the pixel displacements of feature points in the image sequences; The first UAV detects triaxial acceleration and triaxial angular velocity through the inertial measurement unit; The first UAV calculates a first relative motion estimate based on the pixel displacement of the feature points, calculates a second relative motion estimate based on the integral of the three-axis acceleration and the three-axis angular velocity, fuses the first relative motion estimate and the second relative motion estimate to obtain the real-time position and attitude relative to the building structure, and adjusts the motor speed to maintain the real-time position and attitude constant, so that the first UAV and the building structure remain in a non-contact relative stationary state.

3. The pre-positioned unmanned aerial vehicle (UAV) firefighting method for cultural relic protection buildings according to claim 2, characterized in that, The step of extracting the feature point pixel displacement in the image sequence includes: performing corner detection on each frame of the image sequence to identify the texture corners of the building structure surface; performing optical flow tracking on the texture corners in adjacent frames to calculate the pixel displacement of the texture corners on the image plane, so as to obtain the feature point pixel displacement. The first UAV calculates a first relative motion estimate based on the pixel displacement of the feature points, including: calculating the translational and rotational motion of the first UAV relative to the building structure based on the pixel displacement, as the first relative motion estimate.

4. The pre-positioned unmanned aerial vehicle (UAV) fire extinguishing method for cultural relic protection buildings according to claim 1, wherein the pressurized liquid is applied to the fire source in a continuous spray manner, comprising: Based on the infrared thermal imaging array carried by the first UAV, thermal imaging data of the fire source area is collected through the infrared thermal imaging array. Based on the temperature distribution of the thermal imaging data, the fire source area is divided into a core high-temperature zone and a peripheral smoldering zone; The nozzle is controlled to spray liquid into the core high-temperature zone and the peripheral smoldering zone respectively, wherein the amount of liquid per unit area sprayed into the peripheral smoldering zone is less than the amount of liquid per unit area sprayed into the core high-temperature zone.

5. The pre-positioned drone fire extinguishing method for cultural relic protection buildings according to claim 4, wherein the first drone controls the nozzle to spray liquid directionally into the core high-temperature zone, comprising: The center coordinates of the core high-temperature zone are determined based on the thermal imaging data. Adjust the machine body attitude so that the spray direction of the nozzle is aligned with the center coordinate; Liquid is injected into the core high-temperature zone at a first injection pressure, and the first injection pressure is positively correlated with the temperature value of the core high-temperature zone.

6. The pre-positioned unmanned aerial vehicle (UAV) fire extinguishing method for cultural relic protection buildings according to claim 4 or 5, wherein the control nozzle sprays liquid into the core high-temperature zone and the peripheral smoldering zone respectively, comprising: The boundary range of the peripheral smoldering zone is determined based on the thermal imaging data; Liquid is injected into the boundary area at a second injection pressure, which is less than the first injection pressure, so that the amount of liquid injected per unit area into the peripheral smoldering zone is less than the amount of liquid injected per unit area into the core high-temperature zone.

7. The pre-positioned drone firefighting method for cultural relic protection buildings according to claim 1, wherein before the liquid in the storage tank of the first drone is exhausted, the first drone returns to the drone base station to replenish the liquid, and simultaneously, the second drone takes off from the drone base station and flies to the fire source location to take over the spraying, comprising: Real-time monitoring of the remaining liquid level in the storage tank of the first drone; When the remaining liquid volume reaches a preset threshold, the first drone sends a return-to-home request to the drone base station and continues spraying until the second drone reaches the fire source location; Upon receiving the return-to-home request, the drone base station immediately releases the second drone, which then flies along the flight path of the first drone to the location of the fire source. After the second drone reaches the fire source and begins spraying, the first drone stops spraying and returns to base.

8. The pre-positioned unmanned aerial vehicle (UAV) fire extinguishing method for cultural relic protection buildings according to claim 7, wherein the second UAV flies along the flight path of the first UAV to the fire source location, comprising: The first UAV sends the verified flight path data and corresponding historical environmental image features to the UAV base station; The UAV base station loads the flight path data and corresponding historical environmental image features into the second UAV; The second UAV flies according to the flight path data and corrects path deviations during flight using the visual inertial navigation system.

9. The pre-positioned unmanned aerial vehicle (UAV) firefighting method for cultural relic protection buildings according to claim 8, wherein correcting path deviation through the visual-inertial navigation system during flight includes: During flight, the second UAV uses the visible light camera to collect real-time images of the current environment and extract current image features. The current image features are matched with the historical environmental image features recorded in the flight path data to identify the degree of deviation between the current position of the second UAV and the flight path; When the deviation exceeds a preset threshold, the second UAV corrects the integral cumulative error of the inertial measurement unit based on the historical pose data corresponding to the historical environmental image features, so that the second UAV returns to the flight path.

10. The pre-positioned unmanned aerial vehicle (UAV) firefighting method for cultural relic protection buildings according to claim 1, After receiving a fire alarm signal, the drone base station transports the liquid in the storage container to the storage tank of the drone parked on the parking platform, and pressurizes the liquid in the storage tank through the pressurization device.

11. A pre-positioned unmanned aerial vehicle (UAV) fire suppression system for protected cultural relics buildings, characterized in that, include: A drone base station is deployed in the non-cultural relic area of ​​a protected building. The drone base station includes a drone parking platform, a liquid storage container, and a pressurization device. The control center is connected to the drone base station and is used to receive fire alarm signals and, in response to the signals, instruct the first drone to fly to the fire source location. The first drone is placed on the parking platform, and its storage tank is filled with liquid supplied by the storage container and pressurized by the pressurization device. The first drone is used to take off after receiving instructions from the control center, fly to the fire source, hover at the fire source and remain stationary relative to the building structure through the onboard visual inertial navigation system, and apply the pressurized liquid to the fire source in a continuous spray manner. The control center is also used to control the first drone to return to the drone base station for replenishment before the liquid in the first drone's storage tank is exhausted, and at the same time control the second drone to take off from the drone base station and fly to the fire source to take over spraying, forming an uninterrupted fire extinguishing operation.

Images