Fire false alarm prevention method and system based on sunshine geometry constraint and multi-view cooperation

Abstract

The present application relates to the technical field of forest fire prevention monitoring and the technical field of digital twin simulation, in the extremely complex natural geographical and meteorological environment in the field, due to the complex strong light and optical artifacts, the traditional forest fire visual alarm system has a very high false alarm rate, the present application provides a kind of forest fire false alarm prevention method and system based on sunshine geometric constraint and multi-view cooperation, utilize the multi-view cooperative optimization strategy of sunshine geometric physical constraint, determine the cooperative equipment of the current fire target by evaluating the light receiving state of the monitoring camera around the suspected fire target, combined with multi-factor weighted decision to review, when there is no cooperative equipment, dispatch mobile unmanned aerial vehicle vertical angle verification, filter the false alarm of forest fire caused by strong light interference;The present application realizes dynamic reliability decision based on geometric physical constraint, eliminates the frequent false alarm of fire caused by complex strong light and optical artifacts in the field, and significantly improves the reliability of multi-view cooperative verification.

Description

Technical Field

[0001] This invention relates to the fields of forest fire monitoring technology and digital twin simulation technology, and more specifically, to a method and system for preventing false alarms in forest fires based on solar radiation geometric constraints and multi-view collaboration. Background Technology

[0002] Forest fires are characterized by their suddenness, destructive power, and high risk of failure. Therefore, large-scale automated early fire monitoring is crucial for protecting forest resources. In recent years, with the rapid development of artificial intelligence and computer vision technologies, forest fire video monitoring systems based on high-altitude pan-tilt cameras combined with deep learning object detection algorithms (such as the YOLO series algorithms) have been widely used. These systems can automatically identify smoke or firelight in the footage and trigger alarms by performing real-time image analysis on the video streams transmitted from field cameras.

[0003] However, in the extremely complex natural geographical and meteorological environments of the wild, traditional forest fire visual alarm systems suffer from a very high false alarm rate, becoming a major technical bottleneck for grassroots forestry fire prevention departments. The reasons for this lie primarily in the following shortcomings of existing monitoring methods and systems: First, existing forest fire visual monitoring systems lack an underlying physical verification mechanism to detect false alarms caused by optical artifacts from strong light. During specific times of day, intense direct sunlight in the wild can directly enter the camera lens, causing optical reflections within the lens and creating whitish halos, flares, or ghosting in the video image. Simultaneously, when sunlight shines on exposed rocks, water surfaces, or metal roofs of buildings, it produces strong specular reflections. These dynamic optical artifacts are extremely similar in shape, color, and texture to early forest fire smoke and flames. Traditional target detection algorithms are prone to misidentifying optical artifacts as fires, frequently leading to false alarms. Existing false alarm prevention methods are limited to algorithmic optimization within the two-dimensional image plane and cannot combine the three-dimensional spatial relationship between the sun's position, terrain features, and camera perspective to predict interference risks, making it difficult to reduce the probability of false alarms from the optical causes.

[0004] Second, there is a lack of a precise multi-view collaborative verification mechanism based on three-dimensional spatial geometric constraints. Most existing visual false alarm prevention methods are limited to "image-by-image" analysis within a two-dimensional image plane, completely severing the three-dimensional spatial geographical relationship between the camera, the sun, and the ground fire point. This makes it impossible to determine whether the current alarm area is located in a backlight or strong light reflection area prone to optical interference. Due to the lack of a three-dimensional spatial optimization strategy, when a suspected alarm occurs, existing monitoring methods and systems often can only poll all surrounding surveillance cameras for verification, failing to guarantee the effectiveness of collaborative perspectives. This leads to one false alarm result verifying another false alarm result, rendering collaborative verification meaningless and potentially making erroneous conclusions appear more "credible."

[0005] Third, the lack of automated hierarchical verification and three-dimensional validation methods results in high emergency response costs and low efficiency. Existing alarm systems lack automated, multi-dimensional verification methods when faced with low-reliability, highly suspicious visual alarms. If alarms are issued directly, on-duty personnel often need to review video footage for verification. Some optical artifacts are highly similar to real fires, and manual verification based solely on single-view images is insufficient for accurate identification. This necessitates dispatching mobile drones to the scene for verification, leading to high resource allocation costs for emergency response.

[0006] Therefore, how to deeply integrate the geometric and physical constraints of solar radiation, the solution of digital twin three-dimensional spatial geographic grids, and multi-dimensional stereoscopic verification to achieve forest fire interception-verification-verification and prevent false alarms is a core technical problem that urgently needs to be solved in the field of intelligent forestry monitoring. Summary of the Invention

[0007] This invention aims to address the aforementioned deficiencies in existing technologies by providing a forest fire false alarm prevention method and system based on solar radiation geometric constraints and multi-view collaboration. This invention integrates digital twin three-dimensional ray projection positioning, spatial geometric light-receiving physical constraints, and multi-dimensional stereoscopic verification technology to eliminate frequent false alarms caused by complex strong light and optical artifacts in the wild to the forest fire visual monitoring system, and to achieve a low-cost interception-verification-verification disaster prevention closed loop for forest fires.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A forest fire false alarm prevention system based on solar radiation geometric constraints and multi-view collaboration includes a central management server, several surveillance cameras, and mobile drones. The central management server connects all the surveillance cameras and mobile drones via a network. The central management server includes a target detection module, a digital twin engine, a streaming media distribution module, a global parameter calculation module, a virtual-real mapping positioning module, a directional light reception verification module, an optimization scheduling module, and a multi-factor fusion adjudication module. The target detection module identifies fire targets in the video streams acquired by surveillance cameras and mobile drones, outputs the detection results and original detection confidence of suspected fire targets, and determines the main surveillance camera and candidate surveillance cameras for the current fire target. The digital twin engine imports a three-dimensional digital elevation model of the monitored area to achieve spatial coordinate transformation, spatial ray projection, and terrain parameter analysis. The streaming media distribution module pulls and transcodes video streams from surveillance cameras and mobile drones, receives the detection results of fire targets output by the target detection module, overlays fire target identification tags on the video stream, and then sends it to a monitoring display device with the authority to receive video streams. The global parameter calculation module calculates and updates the globally unified sunlight vector within the monitoring area in real time; When the monitoring camera identifies a suspected fire target, the virtual-real mapping positioning module maps the camera's attitude and zoom parameters to the digital twin scene, and calculates the real three-dimensional geographic coordinates of the current fire target and the elevation grid cell where the fire target is located through a spatial ray projection algorithm. The directional light reception test module determines the light reception status of the monitoring camera in the area where the current fire target is located based on the solar radiation geometric constraints, and outputs the judgment result of whether the monitoring camera is in a backlight state, a reflective state, or a normal monitoring state with good lighting conditions. The optimization scheduling module selects surveillance cameras with good lighting conditions from the candidate surveillance cameras as collaborative devices for verification based on the output of the directional light receiving test module, and transmits the two types of basic adjudication data to the multi-factor fusion adjudication module; if there are no candidate surveillance cameras or candidate surveillance cameras with good lighting conditions, the central management server schedules mobile drones for verification. The multi-factor fusion adjudication module converts the two types of basic adjudication data fed back by the collaborative device and performs multi-factor weighted calculation to obtain the final adjudication comprehensive confidence level. And based on the overall confidence level of the final ruling The system can automatically eliminate alarm signals or have the central management server schedule mobile drones to perform vertical perspective verification.

[0009] Furthermore, the surveillance cameras are set up in the monitoring area within the forest area, and the monitoring field of view of each surveillance camera overlaps with the monitoring field of view of at least two adjacent surveillance cameras. Moreover, the ratio of the overlapping range of all monitoring fields of view of the current surveillance camera to the monitoring field of view of the surveillance camera is greater than 0.4.

[0010] Furthermore, the surveillance cameras are sorted according to the time sequence in which they identify the current fire target. The first surveillance camera to identify the current fire target is designated as the main surveillance camera for the current fire target, and the remaining surveillance cameras are designated as candidate surveillance cameras for the current fire target.

[0011] A forest fire false alarm prevention method based on solar radiation geometric constraints and multi-view collaboration, based on the aforementioned forest fire false alarm prevention system based on solar radiation geometric constraints and multi-view collaboration, utilizes a multi-view collaborative optimization strategy under solar radiation geometric constraints. It determines the collaborative devices for the current fire target by evaluating the light reception status of monitoring cameras around the suspected fire target, and verifies this by combining multi-factor weighted decision-making to filter false fire alarms caused by strong light interference. Specifically, it includes the following steps: Step 1. Global Calculation of Sun Ray Vector: The global parameter calculation module of the central management server obtains the current standard time and the longitude and latitude of the center position of the monitoring area in real time, and uses the Standard Solar Position Algorithm (SPA) to calculate the solar altitude angle corresponding to the monitoring area at the current time. With solar azimuth solar altitude angle With solar azimuth It is converted into a unified three-dimensional space sunlight vector, which serves as the global reference physical parameter for verifying the orientation and light reception status of all surveillance cameras; Step 2. Calculation of the main optical axis direction vector and the three-dimensional geographic coordinates of the fire target: When any monitoring camera detects a suspected fire target, a fire alarm signal is activated. The first monitoring camera to detect the current fire target is designated as the main monitoring camera for the current fire target. The digital twin engine of the central management server calculates the main optical axis direction vector of the main monitoring camera and the true three-dimensional geographic coordinates of the current fire target based on the attitude parameters and three-dimensional spatial coordinates of the main monitoring camera. With the true three-dimensional geographic coordinates as the center, a grid cell within a preset range is selected as the surrounding area to be verified. Step 3. Directional Light Reception Status Verification and Interception Condition Determination: After receiving a fire alarm signal, the forest fire prevention false alarm system implements local interception. The directional light reception verification module of the central management server verifies the directional light reception status within the surrounding area determined in Step 2, extracts the unit normal vector of the terrain surface, and calculates the theoretical reflected ray vector by combining it with the sunlight vector obtained in Step 1. The angle between the reverse vector of the calculated principal axis direction vector and the theoretical reflected ray vector is then used to determine the directional light reception status. And the angle between the principal optical axis direction vector and the opposite vector of the sunlight ray vector. The system determines whether the area surrounding the current fire target is a high-risk area for reflection or backlighting by the main monitoring camera. If so, step 4 is initiated. Otherwise, it is determined that the lighting conditions in the surrounding area are good. The system compares the original detection confidence level output by the main monitoring camera through the target detection module with a set threshold. If the original detection confidence level is greater than or equal to the set threshold, the forest fire false alarm prevention system removes the fire alarm signal interception and outputs a real fire alarm signal, completing this false alarm verification process. If the original detection confidence level is less than the set threshold, step 4 is initiated. Step 4. Collaborative Optimization and Scheduling: While maintaining the fire alarm interception state, the forest fire false alarm prevention system retrieves and verifies the monitoring cameras within the surrounding area determined in Step 2 that have passed terrain visibility verification. If a real fire is confirmed, the fire alarm signal interception is lifted and a fire alarm signal is output externally, completing this false alarm prevention verification process; if no fire is confirmed, the current fire alarm signal is directly eliminated, completing this false alarm prevention verification process.

[0012] Furthermore, in step 2, the digital twin engine calculates the principal optical axis direction vector of the surveillance camera using the following method: The central management server extracts the current real-time horizontal azimuth angle of the surveillance cameras. Real-time vertical pitch angle In addition to lens zoom parameters, the digital twin engine drives the corresponding virtual camera in the digital twin scene to maintain the same posture as the surveillance camera in the real world, based on the real-time horizontal azimuth angle of the surveillance camera in the real environment. With real-time vertical pitch angle Projecting onto a 3D digital twin Cartesian coordinate system, the spatial 3D unit direction vector of the current virtual camera is calculated, which is the current principal optical axis direction vector of the surveillance camera: The method for determining the surrounding area of ​​the current fire target is as follows: In a digital twin scenario, the three-dimensional spatial coordinates of a virtual camera correspond to those of a surveillance camera in the real environment are used. Starting from the center pixel coordinates of the fire target, a spatial ray is emitted towards the center pixel coordinates of the fire target. The digital twin engine detects the three-dimensional spatial coordinates of the first collision point between the spatial ray and the surface of the built-in three-dimensional digital elevation model. The three-dimensional spatial coordinates of the first collision point are the three-dimensional geographic coordinates of the current fire target in the real environment. Taking the elevation grid cell where the real three-dimensional geographic coordinates of the current fire target are located as the center, the entire area covered by n layers of adjacent elevation grid cells is the surrounding area of ​​the current fire target, where n is a positive integer.

[0013] Furthermore, in step 3, the process for verifying the directional light reception status is as follows: The directional light-receiving inspection module reads the three-dimensional terrain features of the surrounding area of ​​the current fire target. The digital twin engine of the central management server analyzes the unit normal vector of the three-dimensional terrain surface and uses the standard reflection vector formula to calculate the theoretical reflected ray vector of the sunlight in the surrounding area of ​​the fire target. When the angle between the reverse vector of the principal optical axis direction vector of the surveillance camera and the theoretical reflected ray vector... When the value is less than the set reflective protection threshold, the surveillance camera outputs a reflective state for the area surrounding the current fire target. When the angle between the principal optical axis direction vector of the surveillance camera and the reverse vector of the sunlight vector When the light level is below the set backlight protection threshold, the surveillance camera outputs a backlight condition for the area surrounding the current fire target. Otherwise, the surveillance camera is considered to be in normal monitoring condition with good lighting.

[0014] Furthermore, in step 4, the verification process is as follows: Other surveillance cameras whose field of view covers the real three-dimensional geographic coordinates of the current fire target are candidate surveillance cameras. The optimization scheduling module traverses all candidate surveillance cameras and selects them as collaborative devices. The target detection module obtains the original detection confidence of the candidate surveillance cameras, and the multi-factor fusion adjudication module is used for verification. The verification result of the multi-factor fusion adjudication module is used as the conclusion of the current fire alarm signal verification. If there are no other surveillance cameras or the candidate surveillance cameras in the surrounding area are all in a high-risk state of reflection or backlighting, the central management server will directly dispatch a mobile drone to the airspace directly above the current fire target to verify it from a vertical perspective, and use the fire detection results of the mobile drone as the conclusion for verifying the fire alarm signal.

[0015] Furthermore, the multi-factor fusion adjudication module performs the following review and verification process: The raw detection confidence and light avoidance index of candidate surveillance cameras are calculated using the following formula with multi-factor weighting: in, The overall confidence level of the final ruling; This represents the raw detection confidence score output by the target detection module. The light-shielding index has a range of values. ; and Let be the weighting coefficient, satisfying ,and If the highest overall confidence level of the final ruling If the confidence level is greater than or equal to the set threshold for a valid ruling, then the overall confidence level of the final ruling is directly adopted. The highest-scoring detection result from the collaborative equipment will be used as the conclusion for verifying the fire alarm signal; if the highest final decision has a comprehensive confidence level... If the confidence level is less than the set threshold for a valid decision, a mobile drone is dispatched to the airspace above the current fire target to verify the fire from a vertical perspective, and the detection result of the mobile drone is used as the conclusion for verifying the fire alarm signal.

[0016] In summary, the invention has the following beneficial effects: This invention introduces solar radiation geometric constraints as an alarm interception mechanism without relying on iterative optimization of traditional target detection algorithms. Through spatial geometric calculations of the sunlight vector, the unit normal vector of the 3D terrain surface, and the principal optical axis direction vector of the monitoring camera, it orientably determines whether the monitoring cameras in the vicinity of a suspected fire target are in a high-risk state of backlighting or reflection. This provides physical verification and interception of high-risk alarms from an optical perspective, effectively filtering false fire alarms caused by strong light interference. This invention proposes a multi-view collaborative optimization strategy based on solar radiation geometric constraints. By evaluating the light reception status of each candidate monitoring camera in the vicinity of a suspected fire target, it selects monitoring cameras with good lighting conditions as collaborative devices, replacing the traditional equal-weighted blind polling mode. This avoids invalid verification caused by multiple devices simultaneously misjudging and mutually confirming false alarms in strong light scenarios, significantly improving the reliability of multi-view collaborative verification. Furthermore, by integrating the original detection confidence and the light avoidance index into a multi-factor weighted calculation adjudication model, this invention can dispatch mobile drones to perform vertical perspective verification above the fire point in extreme cases where monitoring cameras cannot effectively determine the fire, greatly improving the agility and accuracy of the alarm. Attached Figure Description

[0017] Figure 1 This is an overall flowchart of the method of the present invention.

[0018] Figure 2 This is a system architecture block diagram of the present invention.

[0019] Figure 3 This is a flowchart of the false alarm prevention decision logic of the present invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings.

[0021] It should be noted that, for ease of description, the descriptions of direction in the following text are consistent with the directions in the accompanying drawings, but they do not limit the structure of the present invention.

[0022] like Figures 1-3As shown, this invention discloses a forest fire false alarm prevention system based on solar radiation geometric constraints and multi-view collaboration. The system includes a central management server, several surveillance cameras, and mobile drones. This invention employs a centralized physical hardware architecture that does not rely on expensive and difficult-to-maintain edge computing nodes. All streaming media retrieval and distribution, underlying core algorithm calculations, and collaborative scheduling of different types of devices are centrally processed through the central management server of the forestry command center. The central management server connects to all surveillance cameras and mobile drones via a network. The central management server includes a target detection module, a digital twin engine, a streaming media distribution module, a global parameter calculation module, a virtual-real mapping positioning module, a directional light reception verification module, an optimization scheduling module, and a multi-factor fusion adjudication module.

[0023] The target detection module identifies fire targets in video streams acquired by surveillance cameras and mobile drones, outputting the detection results and original detection confidence scores of suspected fire targets. Based on the chronological order of fire target identification, it determines the main surveillance camera currently detecting the fire target. The target detection module includes a target detection algorithm; in this embodiment, YOLO is used. Existing algorithms capable of target detection are applicable to this invention. The original detection confidence score in this invention represents the confidence value directly output by the target detection algorithm after detecting and identifying suspected fire targets in the video stream.

[0024] The digital twin engine imports a 3D digital elevation model of the monitored area, enabling spatial coordinate transformation, spatial ray projection, and terrain parameter analysis. The streaming media distribution module retrieves and transcodes video streams from monitoring cameras and mobile drones, receives fire target detection results from the target detection module, overlays fire target identification tags onto the video stream, and then sends it directionally to monitoring display devices with video stream receiving permissions. The streaming media distribution module also controls the opening and closing of the video stream. The global parameter calculation module calculates and updates the globally unified solar ray vector within the monitored area in real time. The virtual-real mapping positioning module maps the camera's attitude and zoom parameters to the digital twin scene when a suspected fire target is detected by the monitoring camera, and uses a spatial ray projection algorithm to calculate the true 3D geographic coordinates of the suspected fire target and the elevation grid cell where the fire target is located. The directional light reception verification module is based on solar geometry. The constraint judgment module determines the lighting status of the surveillance camera when monitoring the area where the current fire target is located, outputting the judgment result of whether the surveillance camera is in a backlight state, a reflective state, or a normal monitoring state with good lighting conditions. The optimization scheduling module first performs terrain visibility verification on the candidate surveillance cameras, eliminating candidate surveillance cameras that are physically obstructed by terrain from the current fire target. Then, based on the output of the directional lighting verification module, it selects surveillance cameras with good lighting conditions from the candidate surveillance cameras as collaborative devices for verification, and passes the two types of basic data for adjudication to the multi-factor fusion adjudication module. If there are no candidate surveillance cameras or candidate surveillance cameras with good lighting conditions, the central management server dispatches a mobile drone for verification. The multi-factor fusion adjudication module converts the two types of basic data for adjudication fed back by the collaborative devices and performs multi-factor weighted calculation to obtain the final adjudication comprehensive confidence level. And based on the overall confidence level of the final ruling The system can automatically eliminate alarm signals or issue waypoint commands to dispatch mobile drones to perform vertical perspective verification.

[0025] Surveillance cameras are deployed within the monitored area of ​​the forest, with each camera's field of view overlapping with the fields of view of at least two adjacent cameras. The ratio of the overlapping area to the camera's total field of view is greater than 0.4. In the outdoor physical environment, multiple visible light PTZ cameras with staggered spatial locations and partially overlapping fields of view are deployed as surveillance cameras, ensuring that the monitored area within the forest is covered by two or more cameras. All surveillance cameras establish network communication with the central management server via a wired network or a high-bandwidth wireless bridge.

[0026] In terms of logical role allocation, the surveillance cameras of this invention are sorted according to the time sequence in which suspected fire targets are identified, and dynamically divided into main surveillance cameras and candidate surveillance cameras in terms of functional roles. Specifically, the target detection module in the central management server performs target detection on the video streams returned by all surveillance cameras. When a suspected fire target of smoke or flame is identified, the surveillance camera that first identifies the current fire target within the current fire alarm signal verification period is designated as the main surveillance camera, while other surveillance cameras whose monitoring field of view can cover the current fire target are designated as candidate surveillance cameras.

[0027] This invention also discloses a forest fire false alarm prevention method based on solar geometric constraints and multi-view collaboration. Based on the aforementioned forest fire false alarm prevention system based on solar geometric constraints and multi-view collaboration, this method utilizes a multi-view collaborative optimization strategy under solar geometric constraints. It determines the collaborative devices for the current fire target by evaluating the light reception status of monitoring cameras around the suspected fire target, and verifies this by combining multi-factor weighted decision-making to filter out false fire alarms caused by strong light interference. Specifically, it includes the following steps: Step 1. Global Calculation of Sun Ray Vector: To establish a unified physical light-receiving benchmark for the entire monitoring area, the global parameter calculation module in the central management server first performs a global calculation of the sun ray vector. This module obtains the current standard time and the longitude and latitude of the center location of the entire monitoring area in real time, and uses the Standard Solar Position Algorithm (SPA) to calculate the solar altitude angle corresponding to the monitoring area at the current moment. With solar azimuth The global parameter calculation module calculates the solar altitude angle. With solar azimuth It is converted into a unified three-dimensional space sunlight ray vector, which serves as the global reference physical parameter for verifying the directional lighting status of all surveillance cameras.

[0028] The specific conversion method for sunlight ray vectors is as follows: The global parameter calculation module uses standard astronomical algorithms to calculate the solar altitude angle based on the current UTC time and the longitude and latitude of the center of the monitored area. With solar azimuth solar altitude angle With solar azimuth Converting angles to three-dimensional unit vectors: vector This refers to the uniform sunlight vector across the entire monitored area at the current moment.

[0029] Step 2. Calculation of the main optical axis direction vector and the 3D geographic coordinates of the fire target: When any monitoring camera within the monitoring area detects a suspected fire target, a fire alarm signal is activated. The first monitoring camera to detect the current fire target is designated as the main monitoring camera for that target. The central management server immediately retrieves the real-time horizontal azimuth, vertical tilt, and lens zoom parameters of the main monitoring camera. The digital twin engine of the central management server drives the corresponding virtual camera in the digital twin scene to rotate synchronously based on the main monitoring camera's attitude parameters and 3D geographic coordinates, ensuring that the virtual monitoring camera maintains a completely consistent attitude with the real-world main monitoring camera. This is based on the real-time horizontal azimuth of the monitoring camera. Real-time vertical pitch angle Determine the current main optical axis direction vector of the main monitoring camera.

[0030] The central management server reads the parameters of the monitoring cameras in real time based on their current position and orientation. The digital twin engine obtains the principal optical axis direction vectors of the main monitoring camera and candidate monitoring cameras in the same way. The specific implementation method of the digital twin engine to obtain the principal optical axis direction vectors of the monitoring cameras is as follows: The central management server retrieves the real-time horizontal azimuth angle of the surveillance cameras through their underlying hardware communication interfaces or standard control protocols. With real-time vertical pitch angle And lens zoom parameters; the digital twin engine calculates the current real-time horizontal azimuth angle of the surveillance camera. With real-time vertical pitch angle Projecting onto a 3D digital twin Cartesian coordinate system, calculate the spatial 3D unit direction vector of the current virtual camera: vector This refers to the current main optical axis direction vector of the surveillance camera. The target detection module in the central management server locks the two-dimensional pixel coordinates of the fire target in the surveillance image. In the digital twin scenario, this corresponds to the three-dimensional spatial coordinates of the virtual camera corresponding to the main surveillance camera of the current fire target. Starting from the principal optical axis, a spatial ray is emitted along the principal optical axis direction vector towards the center pixel coordinates of the fire target. The digital twin engine uses a built-in spatial ray casting algorithm to detect the first collision point between the spatial ray and the surface of the built-in 3D digital elevation model. The 3D spatial coordinates of the first collision point in the digital twin scene are the 3D geographic coordinates of the current fire target in the real environment. Taking the elevation grid cell where the current fire target's 3D spatial coordinates in the digital twin scene are located as the center, the entire area covered by n adjacent elevation grid cells is the surrounding area of ​​the current fire target, where n is a positive integer and is adjusted according to the fire target's position and monitoring accuracy requirements during the monitoring process.

[0031] Step 3. Directional Light Reception Status Verification and Interception Condition Determination: To eliminate strong light interference, the forest fire false alarm prevention system based on solar radiation geometric constraints and multi-view collaboration first implements local interception after receiving a fire alarm signal. The directional light reception verification module in the central management server verifies the directional light reception status of the surrounding area determined in Step 2. The directional light-receiving inspection module reads the 3D terrain features of the area surrounding the current fire target and directly parses the unit normal vector of the 3D terrain surface of the surrounding area from the digital twin engine. Based on the law of reflection, the theoretical reflected ray vector of the sunlight in the surrounding area is calculated using the standard reflection vector formula. The calculation formula is as follows: in, Let be the vector of the incident sunlight. Let be the unit normal vector of the terrain surface. The calculated theoretical reflected ray vector, This represents the vector dot product operation.

[0032] The directional light reception inspection module performs the following judgments on the surveillance camera to be inspected for light reception: Calculate the included angle for determining high risk of glare. : When the reverse vector of the main optical axis direction vector of the surveillance camera ( ) and the theoretical reflected ray vector The included angle When the reflection protection threshold is less than the set threshold, the surveillance camera will be in a reflective state in the area surrounding the current fire target. The area surrounding the current fire target is the high-risk area for reflection of the surveillance camera.

[0033] Calculate the angle for determining high risk of backlighting : When the principal optical axis direction vector of the surveillance camera The vector opposite to the vector of sunlight ( The angle between ) When the light level is below the set backlight protection threshold, the output monitoring camera will be in a backlight state for the area surrounding the current fire target, and the area surrounding the current fire target will be the high-risk backlight zone for the aforementioned monitoring camera; otherwise, the output monitoring camera will be in a normal monitoring state with good lighting conditions.

[0034] In the process of identifying high-risk backlighting areas, terrain visibility verification is introduced to eliminate invalid backlighting scenes caused by mountain obstruction. The terrain visibility verification operation in the digital twin scene is as follows: Starting from the 3D spatial coordinates of the surveillance camera, a spatial ray is emitted in the opposite direction of the sunlight vector. The digital twin engine, combined with the 3D digital elevation model, detects whether there are mountains, highlands, or other terrain features obstructing the propagation path of the emitted spatial ray; only when the angle between the principal optical axis direction vector and the opposite vector of the sunlight vector is within the specified range is the terrain visibility verified. Only when the backlight protection threshold is lower than the set threshold and there is no terrain obstruction between the surveillance camera and the sun can the elevation grid cell be determined to be in a high-risk backlight zone. Other situations are not considered high-risk backlight zones.

[0035] If the area surrounding the current fire target is in a high-risk area for reflection or backlighting from the main monitoring camera, then step 4 is initiated; otherwise, the lighting conditions around the current fire target are considered to be good. If the lighting conditions are good and the original detection confidence level output by the target detection module is greater than or equal to the set threshold, then the forest fire false alarm prevention system based on solar geometric constraints and multi-view collaboration will deactivate the fire alarm interception and output a real fire alarm signal, completing this false alarm verification process; if the original detection confidence level is less than the set threshold, then step 4 is initiated.

[0036] Step 4. Collaborative Optimization and Scheduling: While maintaining the fire alarm interception state, the forest fire false alarm prevention system retrieves and verifies the monitoring cameras within the surrounding area determined in Step 2 that have passed terrain visibility verification. If a real fire is confirmed, the fire alarm signal interception is lifted and a fire alarm signal is output externally, completing this false alarm prevention verification process; if no fire is confirmed, the current fire alarm signal is directly eliminated, completing this false alarm prevention verification process.

[0037] Other surveillance cameras whose field of view can cover the actual three-dimensional geographic coordinates of the current fire target are selected as candidate surveillance cameras. The optimization and scheduling module traverses all candidate surveillance cameras and first performs terrain visibility verification: taking the three-dimensional spatial coordinates corresponding to the candidate surveillance camera in the digital twin scene as the starting point, a spatial ray is emitted towards the three-dimensional spatial coordinates of the current fire target. The digital twin engine detects whether there is a collision point between the spatial ray and the surface of the three-dimensional digital elevation model on the propagation path of the spatial ray. If there is a collision point, it is determined that there is terrain occlusion, and the current candidate surveillance camera does not have effective observation conditions and is eliminated. If there is no collision point, the terrain visibility verification is passed and the subsequent light reception status evaluation process is entered.

[0038] For candidate surveillance cameras that pass the terrain visibility verification, the illumination status of the surrounding area of ​​the fire target is verified based on the method in step 3. Surveillance cameras with good lighting conditions are selected from the candidate cameras that pass the terrain visibility verification as the cooperating devices of the main surveillance camera. For each cooperating device, two types of basic decision-making data are collected: (1) Call the target detection module to identify the real-time video stream of the collaborative equipment and obtain the original detection confidence of the collaborative equipment for the current fire target; (2) Based on the light-receiving state verification method in step 3, the angle between the reverse vector of the principal optical axis direction vector of the collaborative device and the theoretical reflected ray vector is obtained. The angle between the principal optical axis direction vector and the opposite vector of the sunlight ray vector. It serves as the basic parameter for calculating the light avoidance index.

[0039] If any candidate surveillance cameras have good lighting conditions and can be used as collaborative devices, then all the basic data for both types of decisions for each collaborative device will be input into the multi-factor fusion decision module for verification. Conversely, if none of the candidate surveillance cameras have good lighting conditions, then a mobile drone will be directly dispatched to the airspace directly above the current fire target to conduct verification from a vertical perspective, and the detection results of the mobile drone will be used as the final judgment for the fire alarm signal verification.

[0040] When the optimization and scheduling module fails to find other surveillance cameras that can serve as collaborative devices, or when all candidate surveillance cameras in the surrounding area are in a high-risk state of reflection or backlighting, a mobile drone is dispatched to the airspace above the current fire target to verify the fire from a vertical perspective. The fire detection results of the mobile drone are then directly output as the conclusion for verifying the fire alarm signal.

[0041] The final conclusion of the fire alarm signal verification will be executed with a unified response: if a real fire is confirmed, the fire alarm signal interception will be lifted and the fire alarm signal will be output externally, completing this false alarm verification process; if no fire is confirmed, the current fire alarm signal will be directly eliminated.

[0042] The multi-factor weighted fusion adjudication module performs the following verification process: The multi-factor fusion adjudication module receives two types of basic adjudication data, and calculates the included angle... and Converted to a light-avoidance index, and then combined with the original detection confidence levels of each coordinating device for multi-factor weighted calculation, the final comprehensive confidence level of the decision of each coordinating device is obtained. The formula for calculating multi-factor weighted average is: in, The overall confidence level of the final ruling; This represents the raw detection confidence score output by the target detection module. The light avoidance index, with a value range of [0,1], is used to quantify the lighting conditions when collaborative equipment observes fire targets. A larger value indicates better lighting conditions. The calculation formula is as follows: and Let be the weighting coefficient, satisfying ,and .

[0043] Forest fire false alarm prevention systems based on solar radiation geometric constraints and multi-view collaboration prioritize the use of data with higher overall confidence in the final decision. The conclusion of the collaborative device, if the highest final decision has a comprehensive confidence level If the confidence level is greater than or equal to the set threshold for a valid ruling, then the overall confidence level of the final ruling is directly adopted. The highest-scoring detection result from the collaborative equipment will be used as the conclusion for verifying the fire alarm signal; if the highest final decision has a comprehensive confidence level... If the fire level is below the set threshold for valid decision confidence, a mobile drone is dispatched to the airspace above the current fire target to verify the fire from a vertical perspective. The detection result of the mobile drone is used as the conclusion for verifying the fire alarm signal. A unified response is executed based on the accepted final conclusion: if a real fire is confirmed, the alarm signal interception is lifted and a fire alarm signal is output externally, completing this false alarm verification process; if no fire is confirmed, the current fire alarm signal is directly eliminated.

[0044] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A forest fire false alarm prevention system based on solar radiation geometric constraints and multi-view collaboration, characterized in that, It includes a central management server, several surveillance cameras, and mobile drones. The central management server connects all the surveillance cameras and mobile drones via a network. The central management server includes a target detection module, a digital twin engine, a streaming media distribution module, a global parameter calculation module, a virtual-real mapping and positioning module, a directional light reception verification module, an optimization scheduling module, and a multi-factor fusion adjudication module, among which: The target detection module identifies fire targets in the video streams acquired by surveillance cameras and mobile drones, outputs the detection results and original detection confidence of suspected fire targets, and determines the main surveillance camera and candidate surveillance cameras of the current fire target. The digital twin engine imports a three-dimensional digital elevation model of the monitored area to achieve spatial coordinate transformation, spatial ray projection, and terrain parameter analysis. The streaming media distribution module pulls and transcodes video streams from surveillance cameras and mobile drones, receives the detection results of fire targets output by the target detection module, overlays fire target identification tags on the video stream, and then sends it to a monitoring display device with the authority to receive video streams. The global parameter calculation module calculates and updates the globally unified sunlight vector within the monitoring area in real time; When the monitoring camera identifies a suspected fire target, the virtual-real mapping positioning module maps the camera's attitude and zoom parameters to the digital twin scene, and calculates the real three-dimensional geographic coordinates of the current fire target and the elevation grid cell where the fire target is located through a spatial ray projection algorithm. The directional light reception test module determines the light reception status of the monitoring camera in the area where the current fire target is located based on the solar radiation geometric constraints, and outputs the judgment result of whether the monitoring camera is in a backlight state, a reflective state, or a normal monitoring state with good lighting conditions. The optimization scheduling module selects surveillance cameras with good lighting conditions from the candidate surveillance cameras as collaborative devices for verification based on the output of the directional light receiving test module, and transmits the two types of basic adjudication data to the multi-factor fusion adjudication module; if there are no candidate surveillance cameras or candidate surveillance cameras with good lighting conditions, the central management server schedules mobile drones for verification. The multi-factor fusion adjudication module converts the two types of basic adjudication data fed back by the collaborative device and performs multi-factor weighted calculation to obtain the final adjudication comprehensive confidence level. And based on the overall confidence level of the final ruling The system can automatically eliminate alarm signals or have the central management server schedule mobile drones to perform vertical perspective verification.

2. The forest fire false alarm prevention system based on solar radiation geometric constraints and multi-view collaboration according to claim 1, characterized in that, The surveillance cameras are set up in the monitoring area within the forest area, and the monitoring field of view of each surveillance camera overlaps with the monitoring field of view of at least two adjacent surveillance cameras. The ratio of the overlapping range of all monitoring fields of view of the current surveillance camera to the monitoring field of view of the surveillance camera is greater than 0.

4.

3. The forest fire false alarm prevention system based on solar radiation geometric constraints and multi-view collaboration according to claim 2, characterized in that, The surveillance cameras are sorted according to the time sequence in which they identify the current fire target. The first surveillance camera to identify the current fire target is designated as the main surveillance camera for the current fire target, and the remaining surveillance cameras are designated as candidate surveillance cameras for the current fire target.

4. A method for preventing false forest fire alarms based on solar radiation geometric constraints and multi-view collaboration, based on the forest fire false alarm prevention system based on solar radiation geometric constraints and multi-view collaboration as described in any one of claims 1 to 3, characterized in that, A multi-view collaborative optimization strategy based on solar radiation geometric constraints is employed. By evaluating the light reception status of surveillance cameras around suspected fire targets, the collaborative devices for the current fire target are determined. Verification is then performed using a multi-factor weighted decision-making process to filter out false fire alarms caused by strong light interference. The specific steps include: Step 1. Global Calculation of Sun Ray Vector: The global parameter calculation module of the central management server obtains the current standard time and the longitude and latitude of the center position of the monitoring area in real time, and uses the Standard Solar Position Algorithm (SPA) to calculate the solar altitude angle corresponding to the monitoring area at the current time. With solar azimuth solar altitude angle With solar azimuth It is converted into a unified three-dimensional space sunlight vector, which serves as the global reference physical parameter for verifying the orientation and light reception status of all surveillance cameras; Step 2. Calculation of the main optical axis direction vector and the three-dimensional geographic coordinates of the fire target: When any monitoring camera detects a suspected fire target, a fire alarm signal is activated. The first monitoring camera to detect the current fire target is designated as the main monitoring camera for the current fire target. The digital twin engine of the central management server calculates the main optical axis direction vector of the main monitoring camera and the true three-dimensional geographic coordinates of the current fire target based on the attitude parameters and three-dimensional spatial coordinates of the main monitoring camera. With the true three-dimensional geographic coordinates as the center, a grid cell within a preset range is selected as the surrounding area to be verified. Step 3. Directional Light Reception Status Verification and Interception Condition Determination: After receiving a fire alarm signal, the forest fire prevention false alarm system implements local interception. The directional light reception verification module of the central management server verifies the directional light reception status within the surrounding area determined in Step 2, extracts the unit normal vector of the terrain surface, and calculates the theoretical reflected ray vector by combining it with the sunlight vector obtained in Step 1. The angle between the reverse vector of the calculated principal axis direction vector and the theoretical reflected ray vector is then used to determine the directional light reception status. And the angle between the principal optical axis direction vector and the opposite vector of the sunlight ray vector. The system determines whether the area surrounding the current fire target is a high-risk area for reflection or backlighting by the main monitoring camera. If so, step 4 is initiated. Otherwise, it is determined that the lighting conditions in the surrounding area are good. The system compares the original detection confidence level output by the main monitoring camera through the target detection module with a set threshold. If the original detection confidence level is greater than or equal to the set threshold, the forest fire false alarm prevention system removes the fire alarm signal interception and outputs a real fire alarm signal, completing this false alarm verification process. If the original detection confidence level is less than the set threshold, step 4 is initiated. Step 4. Collaborative Optimization and Scheduling: While maintaining the fire alarm interception state, the forest fire false alarm prevention system retrieves and verifies the monitoring cameras within the surrounding area determined in Step 2 that have passed terrain visibility verification. If a real fire is confirmed, the fire alarm signal interception is lifted and a fire alarm signal is output externally, completing this false alarm prevention verification process; if no fire is confirmed, the current fire alarm signal is directly eliminated, completing this false alarm prevention verification process.

5. The forest fire false alarm prevention method based on solar radiation geometric constraints and multi-view collaboration according to claim 4, characterized in that, In step 2, the digital twin engine calculates the principal optical axis direction vector of the surveillance camera as follows: The central management server extracts the current real-time horizontal azimuth angle of the surveillance cameras. Real-time vertical pitch angle In addition to lens zoom parameters, the digital twin engine drives the corresponding virtual camera in the digital twin scene to maintain the same posture as the surveillance camera in the real world, based on the real-time horizontal azimuth angle of the surveillance camera in the real environment. With real-time vertical pitch angle Projecting onto a 3D digital twin Cartesian coordinate system, the spatial 3D unit direction vector of the current virtual camera is calculated, which is the current principal optical axis direction vector of the surveillance camera: The method for determining the surrounding area of ​​the current fire target is as follows: In a digital twin scenario, the three-dimensional spatial coordinates of a virtual camera correspond to those of a surveillance camera in the real environment are used. Starting from the center pixel coordinates of the fire target, a spatial ray is emitted towards the center pixel coordinates of the fire target. The digital twin engine detects the three-dimensional spatial coordinates of the first collision point between the spatial ray and the surface of the built-in three-dimensional digital elevation model. The three-dimensional spatial coordinates of the first collision point are the three-dimensional geographic coordinates of the current fire target in the real environment. Taking the elevation grid cell where the real three-dimensional geographic coordinates of the current fire target are located as the center, the entire area covered by n layers of adjacent elevation grid cells is the surrounding area of ​​the current fire target, where n is a positive integer.

6. The forest fire false alarm prevention method based on solar radiation geometric constraints and multi-view collaboration according to claim 4, characterized in that, In step 3, the process for verifying the directional light reception status is as follows: The directional light-receiving inspection module reads the three-dimensional terrain features of the surrounding area of ​​the current fire target. The digital twin engine of the central management server analyzes the unit normal vector of the three-dimensional terrain surface and uses the standard reflection vector formula to calculate the theoretical reflected ray vector of the sunlight in the surrounding area of ​​the fire target. When the angle between the reverse vector of the principal optical axis direction vector of the surveillance camera and the theoretical reflected ray vector... When the value is less than the set reflective protection threshold, the surveillance camera output will be in a reflective state for the area surrounding the current fire target. When the angle between the principal optical axis direction vector of the surveillance camera and the reverse vector of the sunlight vector When the light level is below the set backlight protection threshold, the surveillance camera outputs a backlight condition for the area surrounding the current fire target. Otherwise, the surveillance camera is considered to be in normal monitoring condition with good lighting.

7. The forest fire false alarm prevention method based on solar radiation geometric constraints and multi-view collaboration according to claim 4, characterized in that, In step 4, the verification process is as follows: Other surveillance cameras whose field of view covers the real three-dimensional geographic coordinates of the current fire target are candidate surveillance cameras. The optimization scheduling module traverses all candidate surveillance cameras and selects them as collaborative devices. The target detection module obtains the original detection confidence of the candidate surveillance cameras, and the multi-factor fusion adjudication module is used for verification. The verification result of the multi-factor fusion adjudication module is used as the conclusion of the current fire alarm signal verification. If there are no other surveillance cameras or the candidate surveillance cameras in the surrounding area are all in a high-risk state of reflection or backlighting, the central management server will directly dispatch a mobile drone to the airspace directly above the current fire target to verify it from a vertical perspective, and use the fire detection results of the mobile drone as the conclusion for verifying the fire alarm signal.

8. The forest fire false alarm prevention method based on solar radiation geometric constraints and multi-view collaboration according to claim 7, characterized in that, The multi-factor fusion adjudication module performs the following verification process: The raw detection confidence and light avoidance index of candidate surveillance cameras are calculated using the following formula with multi-factor weighting: in, The overall confidence level of the final ruling; This represents the raw detection confidence score output by the target detection module. The light-shielding index has a value range of [0,1]. and Let be the weighting coefficient, satisfying ,and If the highest overall confidence level of the final ruling If the confidence level is greater than or equal to the set threshold for a valid ruling, then the overall confidence level of the final ruling is directly adopted. The highest-scoring detection result from the collaborative equipment will be used as the conclusion for verifying the fire alarm signal; if the highest final decision has a comprehensive confidence level... If the confidence level is less than the set threshold for a valid decision, a mobile drone is dispatched to the airspace above the current fire target to verify the fire from a vertical perspective, and the detection result of the mobile drone is used as the conclusion for verifying the fire alarm signal.

Images