Urban fire remote monitoring and early warning method, device and equipment and storage medium

Abstract

The application provides a city fire-fighting remote monitoring and early warning method, device, equipment and storage medium, the method comprises the following steps: constructing a digital twin scene containing spatial structure attributes and component medium attributes based on the building information model data of a building to be monitored, and determining a visual basic environment based on the index association of fire-fighting equipment and three-dimensional space position in the digital twin scene; identifying the space channel allowing medium flow based on the building component semantic attribute in the visual basic environment, and constructing a global space connected network; in response to a fire-fighting host alarm signal, establishing a line-of-sight association relationship between a video pixel area and a three-dimensional voxel unit based on the starting disaster source point positioned in the visual basic environment and the monitoring camera parameters; based on the line-of-sight association relationship and the global space connected network, the image area in the real-time video stream is converted and separated. The application improves the early warning accuracy and emergency disposal efficiency of fire-fighting remote monitoring.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, device, equipment, and storage medium for remote monitoring and early warning of urban fires. Background Technology

[0002] With the development of IoT and cloud computing technologies, urban fire remote monitoring systems have been widely applied in fire safety management of various buildings. Existing urban fire remote monitoring and early warning methods are typically based on a cloud-edge-device architecture. This involves deploying automatic fire alarm control panels, user information transmission devices, and various IoT sensors within the building to upload real-time operational status data and alarm signals from fire equipment to a cloud management platform. Upon receiving a fire alarm or fault signal, the platform notifies on-duty personnel via SMS, telephone, or app push notifications, and displays the location information and basic status of the alarm equipment on the management interface.

[0003] However, existing remote monitoring and early warning methods mainly rely on two-dimensional floor plans or simple text labels to identify alarm points. This approach severs the inherent connection between alarm signals, the three-dimensional spatial structure of buildings, and real-time monitoring video. In complex commercial complexes or high-rise buildings, remote monitoring personnel cannot intuitively construct the three-dimensional spatial relationships around alarm points based solely on abstract coordinates or text, nor can they automatically establish a precise mapping between alarm locations and corresponding monitoring perspectives. Consequently, when an alarm occurs, the system cannot automatically associate and parse the corresponding video content based on the three-dimensional spatial attributes of the alarm point. This often requires personnel to manually search video channels repeatedly or rely on on-site feedback for confirmation. Furthermore, due to the lack of comprehensive analysis of the three-dimensional spatial connectivity and visual context of the alarm point, it is difficult to utilize dynamic environmental features in the video to assist in judging the authenticity of sensor data. This makes it impossible to quickly eliminate false alarms caused by environmental interference such as dust and airflow through spatial and visual collaborative verification, reducing the accuracy of fire remote monitoring and the efficiency of emergency response. Summary of the Invention

[0004] This invention provides a method, device, equipment, and storage medium for remote monitoring and early warning of urban fires, which solves the problems of false alarms being difficult to eliminate and remote verification due to the lack of spatial connectivity and comprehensive analysis of visual context in the prior art, thereby improving the accuracy of early warning and the efficiency of emergency response in remote fire monitoring.

[0005] In a first aspect, the present invention provides a method for remote monitoring and early warning of urban fires, comprising: A digital twin scene containing spatial structural attributes and component media attributes is constructed based on the building information model data of the building to be monitored. The basic environment for visualization is determined based on the index association between fire-fighting equipment and three-dimensional spatial location in the digital twin scene. Based on the semantic attributes of building components in the visualized basic environment, spatial channels that allow the flow of media are identified, and a global spatial connectivity network is constructed. In response to the alarm signal from the fire alarm control panel, a line-of-sight relationship is established between the video pixel area and the three-dimensional voxel unit based on the initial disaster source point located in the visualized basic environment and the parameters of the monitoring camera. Based on line-of-sight relationships and global spatial connectivity networks, dynamic image regions in real-time video streams are transformed and separated to obtain a set of real disaster voxels that are connected to the initial disaster source point by a path. Based on the spatiotemporal diffusion trend of the real disaster voxel set and the numerical change trend of the alarm signal of the fire control panel, a comprehensive early warning command is generated and pushed to the remote monitoring terminal.

[0006] In a second aspect, the present invention also provides an urban fire remote monitoring and early warning device, applied to the urban fire remote monitoring and early warning method as described in the first aspect; the urban fire remote monitoring and early warning device includes: The basic environment construction module is used to construct a digital twin scene containing spatial structural attributes and component media attributes based on the building information model data of the building to be monitored, and to determine the visualization basic environment based on the index association between fire-fighting equipment and three-dimensional spatial location in the digital twin scene. The connectivity network construction module is used to identify spatial channels that allow the flow of media based on the semantic attributes of building components in the visualized basic environment, and to construct a global spatial connectivity network. The line-of-sight mapping module is used to respond to the alarm signal of the fire control panel and establish the line-of-sight relationship between the video pixel area and the three-dimensional voxel unit based on the starting disaster source point located in the visualization base environment and the monitoring camera parameters. The conversion and separation module is used to convert and separate dynamically changing image regions in a real-time video stream based on line-of-sight relationships and a global spatial connectivity network, to obtain a set of real disaster voxels that are connected to the initial disaster source point by a path. The early warning decision and push module is used to generate comprehensive early warning instructions based on the spatiotemporal diffusion trend of the real disaster voxel set and the numerical change trend of the alarm signal of the fire control panel, and push the comprehensive early warning instructions to the remote monitoring terminal.

[0007] Thirdly, the present invention also provides an electronic device, comprising: a memory for storing computer software programs; and a processor for reading and executing the computer software programs, thereby realizing the urban fire remote monitoring and early warning method as described above.

[0008] Fourthly, the present invention also provides a non-transitory computer-readable storage medium storing a computer software program, which, when executed by a processor, implements the urban fire remote monitoring and early warning method described above.

[0009] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the urban fire remote monitoring and early warning method as described above.

[0010] The urban fire remote monitoring and early warning method provided in this invention constructs a digital twin scene containing spatial structural attributes and component media attributes based on building information model data, and establishes an index association between equipment and three-dimensional space. This results in a visualized basic environment that can accurately represent the complex three-dimensional geometric relationships inside the building, breaking the abstract limitations of traditional two-dimensional plans on spatial information. Based on the semantic attributes of building components in this visualized basic environment, it identifies spatial channels that allow media flow and constructs a global spatial connectivity network, clarifying the logical boundaries of disaster spread in three-dimensional physical space. In response to the alarm signal from the fire control panel, it uses the visualized basic environment to locate the initial disaster source point and, combined with the monitoring camera parameters, establishes a line-of-sight association between video pixel areas and three-dimensional voxel units, realizing the alarm point location in the video frame. The automated and precise mapping in the image solves the inefficiency of manually searching for video channels. Furthermore, based on line-of-sight relationships and a global spatial connectivity network, dynamically changing image regions in the real-time video stream are transformed and separated, retaining only the set of real disaster voxels that are physically connected to the initial disaster source point. This effectively eliminates visual artifacts caused by environmental interference due to overlapping viewpoints but spatial disconnectivity. Finally, a comprehensive early warning command is generated based on the spatiotemporal diffusion trend of the real disaster voxel set and the numerical change trend of the alarm signal from the fire control panel. This achieves deep collaborative verification of alarm signals, three-dimensional spatial structure, and video visual content, solving the problems of false alarms being difficult to eliminate and remote verification being difficult due to the lack of spatial connectivity and comprehensive analysis of visual context in the background technology. This improves the early warning accuracy and emergency response efficiency of fire remote monitoring. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating the urban fire remote monitoring and early warning method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the urban fire remote monitoring and early warning device provided in an embodiment of the present invention; Figure 3 An embodiment diagram of the electronic device provided in this invention; Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0014] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0015] See Figure 1 , Figure 1 This is a flowchart illustrating the urban fire remote monitoring and early warning method provided by the present invention. In this embodiment, the executing entity of the urban fire remote monitoring and early warning method is a monitoring and early warning device. Therefore, the urban fire remote monitoring and early warning method includes: Step 10: Construct a digital twin scene containing spatial structural attributes and component media attributes based on the building information model data of the building to be monitored, and determine the basic environment for visualization based on the index association between fire-fighting equipment and three-dimensional spatial location in the digital twin scene.

[0016] Optionally, the monitoring and early warning device acquires building information model (BIM) data of the building to be monitored. This data is a digital 3D model file containing the building's geometry, spatial location, component types, and material properties. Subsequently, the BIM data is parsed to extract spatial structural attributes and component media attributes. The spatial structural attributes include the 3D coordinates, dimensions, shape, and topological connections of building components such as walls, floors, beams, and columns; the component media attributes include the material type of the components (e.g., concrete, glass, metal) and their physical barrier or transmission characteristics to heat, smoke, light, and other media.

[0017] After obtaining the spatial structural properties and component medium properties, the monitoring and early warning device reconstructs a digital twin scene of the building to be monitored in virtual space. That is, it obtains a three-dimensional virtual environment that is consistent with the physical building in terms of geometry and spatial layout.

[0018] After completing the construction of the digital twin scene, the monitoring and early warning device begins to establish an index association between fire-fighting equipment and its three-dimensional spatial location. It reads the fire-fighting facility ledger data or equipment location table of the building to be monitored, obtaining the unique identifier of each fire-fighting equipment, its type (e.g., smoke detector, manual alarm button, fire alarm control circuit equipment), and the two-dimensional plane coordinates or floor / room number of its installation location. The installation location coordinates of the fire-fighting equipment are then mapped to the corresponding three-dimensional spatial coordinates in the digital twin scene. Specifically, for each fire-fighting equipment, a corresponding virtual device node is generated in the digital twin scene, and the three-dimensional spatial coordinates of this virtual device node are set as the actual installation location coordinates of the fire-fighting equipment. Simultaneously, a bidirectional index table is established between the unique identifier of the fire-fighting equipment and the identifier of the virtual device node. Through this bidirectional index table, when a signal from a specific fire-fighting equipment is received, the monitoring and early warning device can immediately locate the corresponding virtual device node and its surrounding three-dimensional spatial area in the digital twin scene. This establishes the basic visualization environment. This basic visualization environment includes the building's three-dimensional geometry, the physical properties of its components, and a complete digital twin scene with precisely labeled three-dimensional location information for all fire-fighting equipment.

[0019] In one embodiment, a commercial complex is used as an example. The monitoring and early warning device imports the building's BIM (Building Information Model) file and parses it to obtain a three-dimensional model of the reinforced concrete structure with 6 floors above ground and 1 floor below ground, including the thickness of each floor slab, the location of the walls, and the size and material of the door and window openings (such as fireproof glass and solid brick walls).

[0020] The monitoring and early warning device reads the fire alarm control panel log and obtains the device information for the "Panhai Sanjiang Alarm Control Panel" located at the "1F-Exhibition Board" position, along with its subordinate "Manual Button_01001007". The installation location of this manual button is recorded as "behind the exhibition board on the east side of the 1st floor exhibition hall". In the 3D model of the 1st floor in the digital twin scenario, the monitoring and early warning device calculates the specific 3D coordinates (X, Y, Z) behind the exhibition board based on the exhibition hall's floor plan and floor height data. A virtual node representing "Manual Button_01001007" is generated at this coordinate, and the correspondence between its device ID and the virtual node ID is recorded. Similarly, all smoke detectors, sprinkler heads, fire hydrant buttons, and other equipment on this floor and throughout the entire building are mapped to their corresponding 3D coordinate points in the digital twin scenario, forming a complete visualization base environment.

[0021] Step 20: Identify spatial channels that allow media flow based on the semantic attributes of building components in the visualized basic environment, and construct a global spatial connectivity network.

[0022] Optionally, the monitoring and early warning device identifies spatial channels that allow the flow of hazardous media such as smoke and heat based on the semantic attributes of building components in the visualized basic environment. It distinguishes between main activity areas (first-type spatial units) and auxiliary circulation areas (second-type spatial units) within the building, establishing horizontal connectivity edges based on geometric adjacency relationships, vertical coupling edges based on vertically connected components (such as stairwells and pipe shafts), and unidirectional forced flow edges based on fluid transport equipment (such as ventilation ducts). Finally, based on these nodes and their corresponding edges, a global spatial connectivity network representing potential disaster spread paths is constructed, as detailed in steps 201 to 205.

[0023] Step 30: In response to the alarm signal from the fire alarm control panel, establish the line-of-sight relationship between the video pixel area and the three-dimensional voxel unit based on the starting disaster source point located in the visualized basic environment and the monitoring camera parameters.

[0024] Optionally, when the monitoring and early warning device receives an alarm signal from the fire alarm control panel, it parses the unique identifier of the fire equipment that triggered the alarm in the alarm signal. Using the bidirectional index table established in step 10, it queries the virtual device node corresponding to the unique identifier of the fire equipment, thereby determining the three-dimensional spatial coordinates of the initial disaster source point in the visualization environment and completing the location identification of the initial disaster source point. That is, the initial disaster source point is the mapping point in the digital twin scene of the physical location of the fire equipment that triggered the alarm signal. Considering that smoke diffusion may cause detectors far from the fire source to alarm before detectors at the fire source, to improve the coverage of video association, this embodiment also defines the three-dimensional spatial region corresponding to the spatial unit node where the initial disaster source point is located and its directly adjacent upstream node in the global spatial connectivity network as the potential source point region.

[0025] Simultaneously, the monitoring and early warning device acquires parameters of the surveillance cameras deployed within the building to be monitored, specifically including the unique identifier of the surveillance camera, the three-dimensional spatial coordinates of its installation location, the optical center coordinates of the lens, the field of view (horizontal and vertical), focal length, resolution, and the attitude angles of the lens orientation (pitch, yaw, and roll). In this embodiment, the monitoring and early warning device can traverse all online surveillance cameras or filter out a set of candidate surveillance cameras based on preset spatial proximity rules.

[0026] The monitoring and early warning device calculates the geometric line-of-sight relationship between a monitoring camera and the potential source area based on the spatial extent of the potential source area and the parameters of each monitoring camera. Specifically, the monitoring and early warning device constructs a field-of-sight cone with the optical center of the monitoring camera as the vertex and the field of view angle as the cone angle. It determines whether the three-dimensional spatial coordinates of the initial disaster source point are located within this field-of-sight cone and further verifies whether there are any building components marked as opaque or solid in the visualization base environment that obstruct the line connecting the optical center to the initial disaster source point. If the initial disaster source point is located within the field-of-sight cone and there are no solid building components obstructing the line, then the initial disaster source point is determined to be within the effective field of view of the monitoring camera. At this time, the line-of-sight vector from the optical center of the monitoring camera to the initial disaster source point is further calculated, and subsequent pixel-to-voxel mapping calculations are performed.

[0027] If the initial disaster source is not located within the field of view cone, or if there are solid building components obstructing the line of sight, it is determined that the initial disaster source is not within the effective field of view of the surveillance camera. In this case, it is marked that an effective line-of-sight relationship cannot be established between the surveillance camera and the current initial disaster source, and subsequent pixel region and voxel unit mapping calculations are not performed on the surveillance camera, or the surveillance camera is removed from the list of associated devices for the current alarm event. If, after traversing all candidate surveillance cameras, no camera with an effective line-of-sight relationship to the initial disaster source is found (e.g., the fire occurs inside the ceiling, in a pipe shaft, or other visual blind spot, or the area is not covered by a camera), the pure sensor mode early warning process will be activated: a comprehensive early warning command will be generated directly based on the numerical change trend of the alarm signal from the fire control panel and the topological path information in the global spatial connectivity network, and the confidence level of the command will be marked as medium to low confidence or requiring manual verification, to ensure basic early warning capabilities in scenarios with obstruction or no video coverage, and to avoid missed alarms.

[0028] Subsequently, for surveillance cameras determined to be within the effective field of view, the monitoring and early warning device establishes a line-of-sight relationship between video pixel regions and three-dimensional voxel units. The specific process is as follows: the two-dimensional video image captured by the surveillance camera is divided into several pixel regions, each corresponding to a rectangular block in the video image. Simultaneously, the three-dimensional space in the visualization environment is divided into several cube units of uniform size, i.e., voxel units. For each pixel region within the field of view of the surveillance camera, a ray emanating from the center point of that pixel region is constructed based on the surveillance camera's intrinsic parameter matrix (including focal length and principal point coordinates) and extrinsic parameter matrix (including rotation and translation information). The propagation path of this ray in three-dimensional space is tracked, and all voxel units traversed by the ray are recorded. To balance computational accuracy and real-time performance, this embodiment employs a "dynamic ROI refinement" strategy to discretize the three-dimensional space: first, in the global spatial connectivity network, starting from the initial disaster source point, a set of "potentially affected areas" nodes that may be affected within a preset time threshold (e.g., 5 minutes) is retrieved based on the medium flow capacity parameter. High-precision voxel discretization (e.g., 0.3m × 0.3m × 0.3m) is applied only to the 3D space within the potential influence area, while coarse-grained meshes (e.g., 1.0m × 1.0m × 1.0m) or ignored for non-potential influence areas. For ultra-large-scale buildings, an octree adaptive subdivision strategy can be combined to locally refine the mesh only around the voxels where dynamic changes are detected.

[0029] For each voxel unit that might be captured by the surveillance camera, it is projected back into the video frame to determine its corresponding pixel area range. This establishes a bidirectional line-of-sight relationship between the video pixel area and the 3D voxel unit, describing the geometric volume range of each pixel or pixel area in the video frame in 3D space, and the projection area of ​​each voxel unit in 3D space in the video frame.

[0030] In one embodiment, assuming that "Manual Button_01001007" triggers a fire alarm, the monitoring and early warning device determines the starting disaster source as the three-dimensional coordinates P_start of the exhibition panel on the first floor. Two candidate cameras installed on the ceiling of the first floor exhibition hall are retrieved: "Screen Camera_01" and "Corridor Camera_02".

[0031] First, we analyze "Screen Camera_01". Its installation coordinates are C_cam1, with a horizontal field of view of 90 degrees and a vertical field of view of 60 degrees, and the lens facing the exhibition panel. Calculations show that point P_start is located within the camera's field of view cone. Further occlusion detection reveals that the line connecting C_cam1 to P_start passes through air and is not obstructed by walls or pillars. Therefore, P_start is determined to be within the effective field of view of "Screen Camera_01". The monitoring and early warning device constructs a line-of-sight vector from C_cam1 to P_start and divides the video image into 16x9 macroscopic pixel blocks. For the first pixel block in the upper left corner, the monitoring and early warning device back-projects a ray R_1 into the three-dimensional space based on its corresponding pixel coordinates and the camera's intrinsic parameters. Detecting the intersection of ray R_1 with the first-floor exhibition space in the digital twin scene reveals that ray R_1 passes through voxel units Voxel_A, Voxel_B, and Voxel_C in sequence. Therefore, the monitoring and early warning device records that there is a line-of-sight relationship between video pixel block 1 and voxel set {Voxel_A, Voxel_B, Voxel_C}.

[0032] Next, we analyze "Corridor Camera_02". Its installation coordinates are C_cam2, with the lens facing deep into the corridor and away from the exhibition hall interior. Calculations show that point P_start is outside the camera's field of view cone (i.e., a blind spot). Alternatively, even if P_start is at the edge of the field of view cone, the monitoring and early warning device detects that the line connecting C_cam2 to P_start is blocked by a solid concrete wall. Therefore, it is determined that P_start is not within the effective field of view of "Corridor Camera_02". "Corridor Camera_02" is marked as having no effective line-of-sight association with the current alarm event, and no subsequent pixel-to-voxel mapping calculations are performed on it to save computational resources and avoid erroneous data associations. Ultimately, the monitoring and early warning device only establishes a complete line-of-sight association mapping table between video pixel areas and three-dimensional voxel units based on "Screen Camera_01".

[0033] Step 40: Based on the line-of-sight relationship and the global spatial connectivity network, the dynamically changing image regions in the real-time video stream are transformed and separated to obtain a set of real disaster voxels that are connected to the starting disaster source point by a path.

[0034] Optionally, the monitoring and early warning device processes the real-time video stream based on the established line-of-sight relationship (two-way line-of-sight relationship) and the global spatial connectivity network. First, it identifies the dynamic feature pixel set in the video frame and maps it to an initial candidate voxel set through line-of-sight relationship. Then, it uses the solid properties of building components to perform line-of-sight accessibility verification, eliminating obstructed voxels; it uses the global spatial connectivity network to retrieve source point connected voxel sets that are connected to the initial disaster source point, and takes the intersection to obtain a preliminary filtered voxel set; finally, through spatial continuity verification and medium penetration attribute filtering, it separates the real disaster voxel set that has a path connected to the initial disaster source point. The specific process is as described in steps 401 to 405.

[0035] Step 50: Based on the spatiotemporal diffusion trend of the real disaster voxel set and the numerical change trend of the alarm signal of the fire control panel, generate a comprehensive early warning command and push the comprehensive early warning command to the remote monitoring terminal.

[0036] Optionally, the monitoring and early warning device generates visual morphological sequences and sensor morphological sequences based on the spatiotemporal diffusion trend of the actual disaster voxel set and the numerical change trend of the alarm signal from the fire alarm control panel, respectively. The morphological matching degree between the two is calculated to obtain a spatiotemporal evolution consistency index, thereby determining the confidence level of the alarm event. Then, combined with core spatial positioning data, video evidence data, and topological path information, a comprehensive early warning instruction is formed, including the early warning level, three-dimensional location information, and spread path data, as detailed in steps 501 to 505. Upon receiving the comprehensive early warning instruction, it is pushed to the remote monitoring terminal.

[0037] The embodiments of the present invention realize in-depth collaborative verification of alarm signals, three-dimensional spatial structure and video visual content, which solves the problems of false alarms being difficult to eliminate and remote verification being difficult due to the lack of spatial connectivity and comprehensive analysis of visual context in the background technology, and improves the early warning accuracy and emergency response efficiency of fire remote monitoring.

[0038] Optionally, the process of steps 201 to 205 includes: Step 201: Based on the geometric data and semantic tags of building components in the visualization base environment, identify different types of spatial areas inside the building to be monitored, and generate a complete set of spatial nodes including a first type of spatial unit and a second type of spatial unit; the first type of spatial unit is the main activity area defined by the physical enclosure structure, and the second type of spatial unit is the auxiliary circulation area defined by the gaps in the building structure or the equipment mezzanine.

[0039] Optionally, the monitoring and early warning device reads the geometric data and semantic tags of building components based on the constructed visualization environment. The geometric data refers to the three-dimensional coordinates, dimensions, and shape information of the physical structures constituting the building, such as walls, floors, beams, and columns. Semantic tags are functional attribute identifiers assigned to building components, such as load-bearing walls, fire-resistant glass doors, ventilation shafts, and ceiling mezzanines. Then, based on the semantic tags, the device filters out the physical enclosing structural components that constitute the spatial boundary. These physical enclosing structural components are those that physically block the flow of media or define the spatial scope, including load-bearing walls, fire-resistant partitions, floors, and fixed partitions. Non-enclosing structural components, such as movable furniture, temporary decorations, and non-boundary internal equipment, are excluded.

[0040] Subsequently, the monitoring and early warning device performs volumetric segmentation of the building's interior space based on the topological connections of the selected solid enclosing structural components. Specifically, it identifies independent closed or semi-closed volumetric regions formed by the solid enclosing structural components in three-dimensional space. For each identified volumetric region, its geometric center coordinates, volume range, list of boundary components, and floor information are extracted.

[0041] The monitoring and early warning device determines the functional attributes of a volume area based on its semantic labels and geometric features, classifying it into either a first-class or second-class spatial unit. Specifically, if the semantic label of a volume area identifies it as a primary functional space, such as an office, lobby, computer room, or corridor, or if its geometric features meet the standards for personnel activity, such as a clear height greater than a preset threshold or the presence of doors and windows for entrances and exits, it is marked as a first-class spatial unit, referring to the main activity area defined by the physical enclosure structure. If the semantic label of a volume area identifies it as a structural void or auxiliary equipment space, such as above a suspended ceiling, under a floor, inside a pipe shaft, or in a cable trench, or if its geometric features are narrow, concealed, and lack direct personnel entrances and exits, such as a clear height lower than a preset threshold or connection only through an inspection port, it is marked as a second-class spatial unit, referring to an auxiliary circulation area defined by structural voids or equipment mezzanines.

[0042] The monitoring and early warning device generates a unique spatial node identifier for each identified first-class and second-class spatial unit, and records its geometric center coordinates, volume range, spatial type identifier, and floor information. All generated spatial nodes together form the complete set of spatial nodes.

[0043] Step 202: Based on the geometric adjacency relationship of each spatial unit in the complete set of spatial nodes and the semantic attributes of boundary components, establish horizontal connecting edges between adjacent spatial units with medium flow conditions to obtain the basic horizontal topology.

[0044] Optionally, geometric adjacency refers to two spatial units sharing at least one common boundary plane or boundary surface in the horizontal direction, and there are openings or connecting components on the common boundary that allow the flow of media; the semantic attributes of the boundary components include component type (such as fire door, ordinary door, fixed window, louver), state (such as normally open, normally closed, closed) and material permeability.

[0045] The monitoring and early warning device traverses the entire set of spatial nodes, encompassing all spatial unit nodes belonging to the same floor or within the same horizontal elevation range. For any two spatial unit nodes at the same horizontal level, it extracts their 3D geometric boundary data within the visualized base environment and determines whether a geometric adjacency exists between them. If two spatial units are completely isolated horizontally by a solid wall with no direct connecting opening, they are considered to have no geometric adjacency. If two spatial units are connected horizontally through a door, window, vent, unsealed opening, or a directly open passageway, they are considered to have a geometric adjacency.

[0046] For adjacent spatial unit nodes that are determined to have a geometric adjacency relationship, the monitoring and early warning device further identifies the horizontal flow path between units based on the semantic attributes of boundary components. If the boundary component allows hazardous media such as smoke or heat to pass through, for example, if the door is open, the window is operable or made of heat-conducting glass, or there is a ventilation opening, then it is determined that there are conditions for media flow. A horizontal connecting edge is then established between adjacent spatial unit nodes where media flow conditions exist. That is, this horizontal connecting edge is a directed or undirected line segment connecting two spatial unit nodes, representing that hazardous media such as smoke and heat can spread horizontally between the two spatial units. The starting node identifier, ending node identifier, connecting position coordinates, connecting area, and boundary component type of each horizontal connecting edge are recorded. All established horizontal connecting edges within the same floor are combined with the corresponding spatial unit nodes to obtain the basic horizontal topology of that floor. This structure is used to describe the network of potential disaster spread paths formed between different functional areas of the building on the same horizontal level through doors, windows, and other openings.

[0047] Step 203: Based on the semantic tags of vertically penetrating components in the visualization base environment, determine the cross-layer physical interface connecting the first type of spatial unit and the second type of spatial unit, and extract the spatial location and associated spatial unit identifier of each cross-layer physical interface to obtain the vertical penetration interface set.

[0048] Optionally, vertically penetrating components refer to building structures or facilities that run through multiple floors in a vertical direction, such as stairwells, elevator shafts, pipe shafts, cable shafts, atrium openings, etc.

[0049] The monitoring and early warning device reads the semantic tags of vertically connected components based on the visualized environment. It then identifies the cross-level physical interface connecting the first type of spatial unit and the second type of spatial unit based on this vertically connected component. This cross-level physical interface refers to the vertical connection opening between the first type of spatial unit (such as a room) and the second type of spatial unit (such as a pipe shaft or ceiling mezzanine), which can be determined by detecting vertical opening components on the boundary of the spatial unit. For example, a smoke exhaust vent in a room connects to a ceiling mezzanine, or an access panel in a ceiling mezzanine connects to a pipe shaft on the upper level.

[0050] For each identified cross-layer physical interface, the monitoring and early warning device extracts its spatial coordinates (X, Y, Z), the identifier of the lower-layer spatial unit it connects to, and the identifier of the upper-layer spatial unit it connects to. All extracted cross-layer physical interface information forms a vertical penetration interface set, which is used to clarify the specific location and path entry point of the disaster medium penetrating from one spatial type to another in the vertical direction.

[0051] Step 204: Based on the set of vertical penetration interfaces, establish a vertical coupling edge between the first type of spatial unit and the second type of spatial unit with vertical connectivity to obtain a cross-level connection structure.

[0052] Optionally, the monitoring and early warning device traverses the vertical penetration interface set. For each cross-layer physical interface in the vertical penetration interface set, it reads the lower-layer spatial unit identifier and the upper-layer spatial unit identifier associated with that interface. Then, based on the read spatial unit identifier, it searches for the corresponding spatial unit node in the entire set of spatial nodes and confirms its spatial type identifier, verifying whether the two spatial unit nodes involved in the cross-layer physical interface are respectively a first-type spatial unit and a second-type spatial unit. After successful verification, it is confirmed that one node is a first-type spatial unit (main activity area) and the other node is a second-type spatial unit (auxiliary circulation area), and a vertical coupling edge is established between these two specific types of spatial unit nodes. This vertical coupling edge is a directional topological connection line, the direction of which is determined by the natural diffusion law of the medium or the physical guidance of the interface components (for example, when flue gas spreads upward, the direction is from the lower first-type spatial unit to the upper second-type spatial unit; or when gravity flows downward, the direction is from the upper second-type to the lower first-type). At the same time, it records the starting node identifier, ending node identifier, associated cross-layer physical interface identifier, and circulation direction attribute of each vertical coupling edge. By summarizing all the verified and established vertical coupling edges, a cross-level connection structure is obtained, which is used to characterize the media penetration path in the vertical dimension between the main activity area and the auxiliary circulation area inside the building.

[0053] Step 205: Based on the basic horizontal topology and the cross-level connection structure, and combined with the semantic information and path data of fluid transport equipment in the visualized basic environment, construct a global spatial connectivity network.

[0054] Optionally, the monitoring and early warning device constructs a global spatial connectivity network based on the basic horizontal topology and the cross-level connection structure, combined with the semantic information and path data of fluid transport equipment in the visualized basic environment, as described in steps 2051 to 2054.

[0055] This invention, through precise identification of main activity areas and auxiliary circulation areas based on geometric data and semantic tags of building components, and the establishment of horizontal connecting edges and vertical coupling edges respectively, constructs a global spatial connectivity network that can fully characterize the complex three-dimensional spatial connectivity relationships inside the building, and clarifies all potential logical boundaries and channels for the diffusion of disaster media (such as smoke and heat) in the physical space of the building.

[0056] Optionally, the process of steps 2051 to 2054 includes: Step 2051: Based on the semantic information and path data of the fluid transport equipment in the visualized basic environment, determine the forced medium transport channel that runs through multiple spatial units, and determine each forced medium transport channel as a unidirectional forced flow edge connecting the starting spatial unit and the ending spatial unit based on the medium flow direction, thus obtaining the mechanical link structure.

[0057] Optionally, the monitoring and early warning device reads the semantic information and path data of the fluid transport equipment within the visualized basic environment. The fluid transport equipment refers to the mechanical equipment and associated piping systems used within a building for the forced transport of air, smoke, or other media, including air conditioning fans, smoke exhaust fans, fresh air systems, exhaust systems, and ventilation ducts and shafts connecting these devices. Semantic information refers to the attribute tags assigned to these devices, such as smoke exhaust fans, air supply outlets, return air outlets, and fire dampers. Path data refers to the geometric orientation, connection relationships, and cross-sectional dimensions of these devices and their piping in three-dimensional space.

[0058] The monitoring and early warning device identifies forced medium delivery channels that traverse multiple spatial units based on path data from fluid delivery equipment. These forced medium delivery channels refer to continuous pipe or duct systems driven by fluid delivery equipment, capable of transporting media across physical barriers or over long distances. It also tracks the start and end points of each ventilation duct segment to determine all spatial unit nodes it passes through.

[0059] For each identified forced media transport channel, the monitoring and early warning device determines the flow direction of the media based on the operating logic of the fluid transport equipment or a preset media flow direction identifier. If a fan is installed in the channel, the flow direction is determined based on the fan's inlet and outlet; if it is a natural ventilation duct, the flow direction is determined based on preset gravity or pressure difference logic. Each forced media transport channel is abstracted as a unidirectional forced flow edge. The starting node of this unidirectional forced flow edge is the spatial unit where the media flows in, and the ending node is the spatial unit where the media flows out. Combining all unidirectional forced flow edges yields a mechanical link structure, used to characterize the disaster media transport path driven by mechanical equipment, possessing clear directionality and high-speed diffusion capabilities.

[0060] Step 2052: Based on the basic horizontal topology, the cross-level connection structure, and the mechanical link structure, construct a hybrid directed graph containing different types of spatial unit nodes, horizontal connected edges, vertical coupled edges, and unidirectional forced flow edges.

[0061] Optionally, based on the acquired basic horizontal topology, the cross-level connection structure, and the mechanical link structure, the monitoring and early warning device uses all the aforementioned spatial unit nodes (including first-type and second-type spatial unit nodes) as vertices of the graph, and horizontal connecting edges, vertical coupling edges, and unidirectional forced flow edges as edges of the graph, constructing a unified hybrid directed graph. In this hybrid directed graph, horizontal connecting edges are typically undirected or bidirectional directed edges, representing bidirectional diffusion of the medium; vertical coupling edges are typically directed edges, representing the main permeation direction of the medium under gravity or thermal buoyancy; and unidirectional forced flow edges are strictly unidirectional directed edges, representing the medium transmission direction under mechanical drive.

[0062] Step 2053: Based on the connectivity properties of each node in the hybrid directed graph, prune isolated nodes and invalid terminal nodes to obtain a purified effective topological subgraph.

[0063] Optionally, an isolated node refers to a spatial unit node that has no edges connected to it. It typically represents a completely enclosed, unventilated, and unconnected dead space, or an erroneous node that has not been connected to other spaces due to missing data. An invalid terminal node refers to a terminal node that has only an inward edge and no outward edge, and is not part of an external building vent or safe zone, or a starting node that has only an outward edge and no inward edge and is not a disaster source. For example, some enclosed decorative cavities, although connected to the main space by tiny gaps, can be considered invalid terminals in a macro-hazard diffusion model.

[0064] The monitoring and early warning device traverses all nodes in the hybrid directed graph, identifies isolated nodes, and removes them from the graph. Simultaneously, it identifies invalid terminal nodes and removes them and their connected invalid edges from the graph based on preset connectivity thresholds or business rules. This removes nodes and edges that have no practical significance in the disaster spread simulation, reducing subsequent computation and improving network effectiveness. After pruning, the remaining nodes and edges form a purified, effective topology subgraph.

[0065] Step 2054: Based on the effective topology subgraph, add a spatial type identifier to each node, a connectivity type identifier to each edge, and a flow capacity parameter to obtain the global spatial connectivity network.

[0066] Optionally, the spatial type identifier is used to distinguish whether the node is a first-type spatial unit (main activity area) or a second-type spatial unit (auxiliary circulation area), so that different disaster diffusion parameters can be used for different types of spaces in subsequent analysis. The connectivity type identifier is used to distinguish whether the edge is a horizontally connected edge, a vertically coupled edge, or a unidirectional forced circulation edge; the circulation capacity parameter refers to the quantitative indicators that the edge allows the medium to pass through, such as the connected area, wind speed, drag coefficient, etc.

[0067] The monitoring and early warning device adds a spatial type identifier to each node in the effective topology subgraph. For each edge in the effective topology subgraph, it adds a connectivity type identifier and a flow capacity parameter. Specifically, for horizontally connected edges, the flow capacity parameter can be calculated based on the area of ​​door and window openings; for vertically coupled edges, it can be calculated based on the interface cross-sectional area and height difference; for unidirectional forced flow edges, it can be calculated based on the fan power and pipe cross-sectional area. After attribute enhancement, the resulting graph structure, containing rich semantic attributes and physical parameters, is the global spatial connectivity network. This network not only describes the connectivity relationships between spaces but also quantifies the strength and direction of connectivity.

[0068] Furthermore, to enable the global spatial connectivity network to adapt in real time to the actual physical state of the building, this embodiment adds a dynamic state association identifier to each edge of the global spatial connectivity network. This identifier is associated with the real-time status point ID in the building automation system (BAS) or fire alarm control system. For example, a horizontal connectivity edge is associated with the door magnetic status ID of a fire door, and a unidirectional forced flow edge is associated with the start / stop status ID of a smoke exhaust fan. This allows for real-time reading of these status IDs during subsequent path retrieval, retaining only edges with a connected or running status, thereby dynamically eliminating physical paths interrupted by closed fire doors or stopped fans, ensuring that the topology is consistent with the actual isolation state on site.

[0069] The embodiments of the present invention generate a high-precision, high-availability global spatial connectivity network, which not only makes up for the shortcomings of traditional methods that only focus on static physical connectivity and ignore the impact of dynamic mechanical ventilation, but also provides a data basis for estimating the speed of disaster spread by quantifying circulation capacity parameters.

[0070] Optionally, the processes of steps 401 to 405 include: Step 401: Based on the real-time video stream of the surveillance camera, identify the pixel regions in the image that exhibit dynamic changing characteristics, and generate a dynamic feature pixel set.

[0071] Optionally, the monitoring and early warning device acquires real-time video streams from surveillance cameras that have a line-of-sight relationship with the initial disaster source. This real-time video stream refers to a sequence of two-dimensional images continuously acquired and transmitted by the surveillance cameras. Each frame consists of several pixels, and each pixel contains visual information such as color and brightness. Differential processing or optical flow calculation is performed on multiple consecutive frames in the real-time video stream to identify pixel regions exhibiting dynamic changes. For example, in this embodiment, the current video frame is selected as the reference frame, and the video frames from the previous moment or several moments ago are selected as the background frame. The difference in grayscale values ​​or color vectors between the reference frame and the background frame at corresponding pixel positions is calculated. If the difference of a pixel exceeds a preset dynamic change threshold, it is determined that the pixel has undergone a significant change, which may correspond to dynamic events such as smoke diffusion, flame flickering, or object movement. Furthermore, to further eliminate false dynamic features caused by strong light reflection, direct sunlight, or overexposure of cameras in the later stages of a fire, this embodiment can also introduce a brightness saturation verification mechanism: when the brightness of a local area of ​​the image exceeds a preset saturation threshold and lasts for a certain period of time, the weight of the dynamic feature pixels in that area is reduced, or a joint judgment is made by combining the changes in chromaticity components in the color space (such as HSV) to distinguish the color features of real smoke / flames from changes in ambient light.

[0072] The monitoring and early warning device aggregates all pixels identified as having undergone significant changes, forming several connected pixel regions. For each connected pixel region, it records the set of pixel coordinates, the region center coordinates, the region area, and the average intensity of change, thus obtaining dynamic change characteristics. All identified pixel regions containing dynamic change characteristics together form a dynamic feature pixel set. This eliminates static background parts (such as walls and furniture) from the image, retaining only the visual foreground areas that may contain disaster-causing media or other dynamic interference.

[0073] Step 402: Based on the line-of-sight relationship, traverse each pixel in the dynamic feature pixel set and map the pixel to the candidate voxel unit in the three-dimensional space of the visualization base environment to obtain the initial candidate voxel set.

[0074] Optionally, the monitoring and early warning device traverses every pixel or pixel region in the dynamic feature pixel set based on the line-of-sight relationship. For each pixel, based on its coordinates in the video frame and combined with the intrinsic and extrinsic parameter matrices of the monitoring camera, a three-dimensional ray is constructed from the optical center of the camera and passes through the pixel. The propagation path of this ray in the visualization environment is tracked, and all voxel units traversed by the ray are recorded. Since a single ray may pass through multiple voxels, or a pixel region may correspond to a ray cone, all voxel units corresponding to the pixel or pixel region that are traversed by the ray are marked as candidate voxel units.

[0075] The monitoring and early warning device deduplicates and merges candidate voxel units obtained by mapping all pixels in the dynamic feature pixel set to obtain an initial candidate voxel set. This initial candidate voxel set represents the possible geometric volume range in three-dimensional space corresponding to all pixels that exhibit dynamic changes in the video image.

[0076] Step 403: Based on the solid properties of building components in the visualized basic environment, perform visual reachability verification on the initial candidate voxel set to obtain the visually reachable voxel set.

[0077] Optionally, the solid properties of building components refer to the physical characteristics of whether the component allows light to pass through. For example, concrete walls and solid wooden doors are opaque solid components, glass windows are transparent or semi-transparent components, and air is a completely transparent medium.

[0078] The monitoring and early warning device reads the solid properties of building components based on the visualized environment, then traverses the initial candidate voxel set. For each voxel, it reconstructs the line of sight from the optical center of the monitoring camera to the geometric center of that voxel. It checks whether this line of sight intersects with building components marked as opaque or solid in the visualized environment during propagation. If the line of sight is obscured by an opaque building component (such as a solid wall, column, or opaque partition) before reaching the target voxel, the voxel is determined to be unreachable and removed from the initial candidate voxel set. If the line of sight is not obscured by an opaque component, or only passes through a transparent / semi-transparent component (such as glass), the voxel is determined to be reachable and retained. After verifying all initial candidate voxels, a reachable voxel set is obtained. This set eliminates false voxels that overlap in the video projection direction but are actually obscured by physical building components and cannot be directly observed by the camera, ensuring that the remaining voxels are visually visible to the camera.

[0079] Step 404: Based on the global spatial connectivity network, search all spatial unit nodes where the starting disaster source point is located have a topological connectivity path to obtain the source point connected voxel set, and determine the intersection of the line-of-sight reachable voxel set and the source point connected voxel set as the preliminary screening voxel set.

[0080] Optionally, the initial disaster source point refers to the spatial unit node corresponding to the three-dimensional spatial location of the fire-fighting equipment that triggered the alarm, as determined in step 30.

[0081] The monitoring and early warning device is based on a global spatial connectivity network, retrieving all spatial unit nodes that have a topological connection path to the initial disaster source point. The process is as follows: In the global spatial connectivity network, starting from the spatial unit node where the initial disaster source point is located, a breadth-first search or depth-first search algorithm is used to traverse all downstream spatial unit nodes connected by horizontal connecting edges, vertical coupling edges, or unidirectional forced flow edges. The three-dimensional spatial volume occupied by these nodes reachable through network paths constitutes the source point connectivity volume set, representing all spatial areas that the disaster medium can theoretically diffuse to based on the building's physical structure and ventilation system logic.

[0082] Subsequently, the monitoring and early warning device calculates the intersection of the visually accessible voxel set and the source-connected voxel set. That is, it retains only voxel units that are both visible in the camera's field of view (visually accessible) and physically connected to the originating disaster source (source-connected), thus obtaining a preliminary voxel set. This achieves the dual purpose of using visual accessibility to eliminate visual obstruction illusions and using spatial connectivity to eliminate visually visible but physically inaccessible independent areas from the disaster source, such as smoke seen through glass in a neighboring room; if there is no connection between the two, then the disaster is not caused by this particular source.

[0083] Step 405: Based on the preliminary selected voxel set and the adjacency relationship in the global spatial connectivity network, the authenticity is verified to obtain the real disaster voxel set.

[0084] Optionally, the monitoring and early warning device performs authenticity verification based on the preliminary selected voxel set and the adjacency relationship in the global spatial connectivity network to obtain the real disaster voxel set, as described in steps 4051 to 4053.

[0085] The embodiments of the present invention are based on a multi-dimensional collaborative screening mechanism of visual dynamics, physical occlusion, and spatial connectivity, which ensures a high degree of consistency in geometric position, visual representation, and physical logic of the final generated real disaster voxel set, providing a clean and highly reliable data foundation for subsequent spatiotemporal evolution analysis and confidence assessment.

[0086] Optionally, the processes of steps 4051 to 4053 include: Step 4051: Based on the adjacency relationship in the global spatial connectivity network, perform spatial continuity verification on the initially screened voxel set to obtain a spatially continuous voxel set.

[0087] Optionally, the adjacency relationship in a global spatial connectivity network refers to the direct connection relationship established between nodes in the network through horizontal connecting edges, vertical coupling edges, or unidirectional forced flow edges. This relationship is mapped to three-dimensional space and manifests as direct contact between adjacent voxel units in geometric space or connection through a specific interface.

[0088] The monitoring and early warning device traverses every voxel in the initial screening voxel set. For each voxel to be verified, its corresponding spatial unit node in the global spatial connectivity network is retrieved, and all adjacent nodes of that node are found. It is determined whether the voxel to be verified has a direct adjacent path to the voxel where the initial disaster source point is located, or whether it has a direct adjacent relationship with other voxel units already confirmed in the initial screening voxel set. If a voxel has no direct adjacent relationship with the voxel where the initial disaster source point is located, and has no direct adjacent relationship with any other voxel unit in the initial screening voxel set, then the voxel is determined to be an isolated voxel unit. Since isolated voxel units are usually discrete points generated by video noise, light and shadow flicker, or algorithm errors, they do not conform to the characteristic of continuous diffusion of disaster media (such as smoke and flames) in physical space. Therefore, these isolated voxel units are removed from the initial screening voxel set, resulting in a spatially continuous voxel set composed of voxel units with direct adjacent relationships.

[0089] Step 4052: Based on the medium penetration attribute identifier of the voxel units in the spatial continuous voxel set, the voxel units that allow medium flow are screened to obtain the medium flow voxel set.

[0090] Optionally, the medium penetration attribute label refers to the physical attribute label assigned to each voxel unit in the visualization base environment, used to characterize whether the space occupied by the voxel allows hazardous media such as smoke and heat to pass through. For example, voxels labeled as air or the interior of ventilation ducts have the attribute of allowing medium flow; while voxels labeled as concrete walls, solid columns, or glass curtain walls have the attribute of blocking medium flow (although they may be visually transparent, they physically block smoke).

[0091] The monitoring and early warning device traverses each voxel unit in the spatial continuous voxel set, reading its medium penetration attribute identifier. If a voxel unit's medium penetration attribute identifier indicates that flow is permitted, the voxel unit is retained; if it indicates that flow is prohibited or solid resistance, the voxel unit is removed from the spatial continuous voxel set. After screening, the remaining voxel units form a medium flow voxel set. This further eliminates invalid areas that are spatially continuous and visible but actually located inside solid building components (such as voxels that have entered the wall due to ray tracing errors) or lack medium transmission capabilities.

[0092] Step 4053: Based on the medium flow voxel set, the spatial coordinates and attribute information of the voxel units therein are encapsulated to obtain the real disaster voxel set that has a path connection with the starting disaster source point.

[0093] Optionally, the monitoring and early warning device extracts the three-dimensional spatial center coordinates (X, Y, Z), volume, spatial unit identifier (e.g., "1F-Exhibition Hall"), medium type identifier (e.g., "Flue Gas"), and timestamp information of each voxel unit in the medium flow voxel set. This data is packaged and encapsulated according to a preset data structure format to form a standardized data object, thus obtaining the real disaster voxel set that has a path connection to the initial disaster source point. This real disaster voxel set not only contains the specific three-dimensional location information of the disaster occurrence but also implicitly includes its topological connectivity path information to the initial disaster source point.

[0094] The embodiments of the present invention achieve a progressive and precise filtering from visually visible to physically reachable and then to media passable, which completely solves the problems of high false alarm rate and inaccurate positioning caused by the lack of spatial topological constraints and physical attribute verification in traditional video fire detection.

[0095] Optionally, the processes of steps 501 to 505 include: Step 501: Based on the spatial distribution data of the real disaster voxel set within a continuous time window, and combined with the spatial security attributes of the nodes in the global spatial connectivity network, determine the spatial threat index of the real disaster voxel set within each time slice, and generate a path threat sequence arranged in chronological order based on the spatial threat index.

[0096] Optionally, the monitoring and early warning device acquires the spatial distribution data of the real disaster voxel set obtained in step 40 within a continuous time window. The continuous time window refers to a preset duration (e.g., 30 seconds or 60 seconds) extending backward from the alarm trigger time. This time period is divided into several equal-length time slices (e.g., one time slice per second). For each time slice, the number of voxel units contained in the real disaster voxel set at that moment, the total volume, and the spatial node range occupied by these voxel units in the global spatial connectivity network are statistically analyzed to obtain spatial distribution data.

[0097] Simultaneously, the monitoring and early warning device reads the spatial safety attributes of each node in the global spatial connectivity network. These spatial safety attributes refer to pre-set risk weight values ​​based on the fire resistance rating of building components, personnel density, evacuation difficulty, and media circulation capacity. For example, main activity areas (Category I spatial units) have higher spatial safety attribute weights due to their high population density; auxiliary circulation areas (Category II spatial units) are also set to specific high-risk values ​​due to their concealed location and the ease with which smoke can spread; while non-circulation areas such as solid walls have zero weight. This is used to characterize the potential threat to human life, building structural safety, and property when hazardous media (such as smoke and heat) spread to the spatial unit.

[0098] The monitoring and early warning device combines the spatial distribution data of real disaster voxel sets with the spatial security attributes of corresponding nodes to calculate the spatial threat index of the real disaster voxel sets within each time slice. The calculation method is as follows: the spatial security attribute weights of all spatial nodes belonging to the real disaster voxel units within the time slice are weighted and summed, and then multiplied by the total volume or quantity coefficient of the disaster voxels within the time slice to obtain a quantitative value characterizing the potential hazard level of the disaster at that moment. The spatial threat indices corresponding to all time slices are arranged in chronological order to generate a path threat sequence arranged chronologically, reflecting the evolution trend of the disaster's coverage and hazard level in three-dimensional space over time.

[0099] In one embodiment, taking the manual button fire alarm in the first-floor exhibition hall as an example, the continuous time window is set from 19:08:25 to 19:08:35, a total of 10 seconds, divided into 10 time slices (T1 to T10). At time T1 (19:08:25): the actual disaster voxel set only contains 5 voxels near the source point, located in the "main activity area of ​​the exhibition hall" (Node_1F_Hall_Main, with a spatial security attribute weight of 0.8). The spatial threat index = 5 * 0.8 = 4.0.

[0100] At time T5 (19:08:29): The real disaster voxel set has spread to 20 voxels, of which 15 are in the "main activity area of ​​the exhibition hall" and 5 have entered the "ceiling mezzanine" (Node_1F_Ceiling_Void, with a spatial safety attribute weight of 0.9, due to its concealment and rapid spread). Spatial threat index = (15*0.8) + (5*0.9) = 12.0 + 4.5 = 16.5.

[0101] At time T10 (19:08:35): the actual disaster voxel set has spread to 50 voxels and spread to the corridor. The monitoring and early warning device arranges the spatial threat indices (4.0, ..., 16.5, ...) calculated at these 10 times in sequence to form a path threat sequence [4.0, 5.2, 7.1, 9.5, 16.5, ...].

[0102] Step 502: Based on the values ​​of the corresponding time slices in the continuous time window associated with the initial disaster source in the alarm signal of the fire control panel, generate a sensor value sequence arranged in chronological order.

[0103] Optionally, the monitoring and early warning device reads historical sensor data associated with the initial disaster source from the alarm signal of the fire control panel. The fire-fighting equipment (such as smoke detectors, manual buttons, or combined detectors) corresponding to the initial disaster source will continuously report status values ​​within a continuous time window. For smoke detectors, this value is the smoke concentration value or analog voltage value; for temperature detectors, this value is the ambient temperature value.

[0104] The monitoring and early warning device extracts sensor values ​​at time points that correspond one-to-one with time slices in the path threat sequence within a continuous time window. If the sensor reporting frequency is higher than the time slice frequency, the average or maximum value within that time slice is used; if it is lower, linear interpolation is used or the last known value is retained. These extracted sensor values ​​are arranged in chronological order to generate a chronologically ordered sensor value sequence, reflecting the evolutionary trend of disaster energy release or medium concentration changes perceived at the electrical signal level.

[0105] In one embodiment, corresponding to the aforementioned 10 time slices (T1 to T10), the monitoring and early warning device extracts the smoke concentration readings of the "Display Board Wireless 4G Smoke Detector" from the monitoring and early warning host logs during the period from 19:08:25 to 19:08:35. Reading at time T1: 3.0%obs / m; Reading at time T5: 3.4%obs / m; Reading at time T10: 4.2%obs / m, etc. These values ​​are arranged to form a sensor value sequence [3.0, 3.1, 3.2, 3.3, 3.4, ..., 4.2].

[0106] Step 503: Based on the variation characteristics of values ​​between adjacent time slices in the path threat sequence, visual change pattern identification is performed on each time slice to obtain a visual pattern sequence arranged in chronological order. Based on the variation characteristics of values ​​between adjacent time slices in the sensor value sequence, sensor change pattern identification is performed on each time slice to obtain a sensor pattern sequence arranged in chronological order.

[0107] Optionally, during a real fire, the smoke and heat generated by combustion follow the law of diffusion in physical space. Their visual manifestations (such as smoke volume expansion and flame area increase) and the physical quantities detected by sensors (such as smoke concentration and temperature rise) have an inherent positive correlation over time. That is, as the fire develops, the visually perceived space occupied by the disaster usually shows a non-decreasing expansion trend, while the sensor values ​​also show a cumulative upward trend.

[0108] Therefore, the monitoring and early warning device calculates the changing characteristics of spatial threat indicators between adjacent time slices for a path threat sequence. These characteristics include the rate of change (current value minus the previous value) or the direction of change (increasing, decreasing, or remaining constant). Then, based on preset threshold rules, each time slice is visually labeled with a change pattern. For example, if the rate of change is greater than a preset positive threshold, it is labeled as rapid expansion; if the rate of change is near zero, it is labeled as stable; and if the rate of change is negative, it is labeled as contraction. All the labels for the time slices are arranged sequentially to obtain a visual pattern sequence arranged chronologically.

[0109] For the sensor value sequence, the monitoring and early warning device calculates the change characteristics of sensor values ​​between adjacent time slices. It also calculates the rate of change or direction of change, and labels each time slice with a sensing change pattern according to preset threshold rules. For example, a significant increase in value is labeled as an energy surge, a small fluctuation as background noise, and a decrease as dissipation. The labels of all time slices are arranged sequentially to obtain a chronological sequence of sensing patterns.

[0110] In one embodiment, for the generation of the visual morphological sequence, from T1 to T2, the spatial threat index increases from 4.0 to 5.2, with a change rate of 1.2, which is marked as slow growth. From T4 to T5, the spatial threat index increases from 9.5 to 16.5, with a change rate of 7.0, exceeding the threshold, which is marked as rapid expansion. The final visual morphological sequence is: [Initial, Slow Growth, Slow Growth, Accelerated Growth, Rapid Expansion, ...].

[0111] For the generation of the sensing pattern sequence, from T1 to T2, the smoke concentration increases slightly from 3.0% to 3.1%, which is marked as a steady increase. From T4 to T5, the smoke concentration increases from 3.3% to 3.4%, still showing a steady increase. If the concentration suddenly jumps from 3.4% to 4.0% at a certain moment, it is marked as a surge. The final sensing pattern sequence is: [Initial, steady increase, steady increase, steady increase, steady increase, ...].

[0112] Step 504: Based on the visual morphology sequence and the sensor morphology sequence, calculate the morphology matching degree of the two sequences in the time dimension according to the preset morphology matching rules to obtain the morphology matching score sequence.

[0113] Optionally, the monitoring and early warning device acquires preset morphological matching rules and calculates the matching degree of visual morphological sequences and sensory morphological sequences segment by segment along the time dimension. These morphological matching rules define a mapping table that defines the similarity between two morphological identifiers. For example, when the visual morphological identifier and the sensory morphological identifier are completely identical, the matching degree is assigned a value of 1; when they belong to the same direction but different rates, such as rapid expansion and slow expansion, the matching degree is assigned a value of 0.6; when both are stable, the matching degree is assigned a value of 1; when their directions are opposite, or there is no visual change but a sensory change, such as expansion and contraction, the matching degree is assigned a value of 0; and in other cases, the value is assigned a value of 0.3. Therefore, each time slice is traversed, searching for the identifier of that slice in the visual morphological sequence and the identifier of that slice in the sensory morphological sequence, and querying the corresponding matching score according to the matching rules. The matching scores of all time slices are arranged in order to obtain a morphological matching score sequence, which is an ordered numerical sequence reflecting the degree of synchronization between visual spatial evolution and sensor signal evolution at each moment.

[0114] In one embodiment, it is assumed that at time T1, the visual-initial vs. sensor-initial condition has a matching score of 1.0. At time T2, the visual-slow growth vs. sensor-smooth increase condition has a matching score of 1. At time T5, the visual-rapid expansion vs. sensor-smooth increase condition has a matching score of 0.6. The final morphological matching score sequence is: [1.0, 1.0, 1.0, 0.6, 0.6, ...].

[0115] Step 505: Based on the effective matching degree in the morphological matching score sequence, generate a spatiotemporal evolution consistency index that characterizes the consistency between visual space expansion and electrical signal energy release evolution, and generate the comprehensive early warning command based on the spatiotemporal evolution consistency index.

[0116] Optionally, the monitoring and early warning device first counts the number of time slices in the morphological matching score sequence whose values ​​are greater than or equal to a preset confidence threshold. This number is then divided by the total number of time slices within the continuous time window to obtain the effective matching rate. Subsequently, the device identifies the longest consecutive time slice in the morphological matching score sequence whose values ​​are continuously greater than or equal to the preset confidence threshold. This length is then divided by the total number of time slices to obtain the evolutionary continuity rate. This evolutionary continuity rate is a parameter characterizing the degree to which the synchronization state of visual and sensor signals remains uninterrupted on the time axis, used to eliminate false high consistency caused by occasional instantaneous matching. The effective matching rate and the evolutionary continuity rate are weighted and summed according to a preset consistency fusion weight, and the summation result is normalized to the 0-1 range to obtain the spatiotemporal evolutionary consistency index. This spatiotemporal evolutionary consistency index comprehensively reflects the quantitative scalar measure of the overall consistency between the spatial diffusion trend of the real disaster voxel set and the numerical change trend of the fire alarm signal within the continuous time window in terms of both morphological synchronization and temporal continuity.

[0117] In one embodiment, assuming a preset confidence threshold of 0.5, statistical analysis of the sequence reveals that 6 time slices have a morphological matching score greater than or equal to 0.5, with a total of 10 time slices. The calculated effective matching rate is 6 ÷ 10 = 0.6. Subsequently, assuming the device identifies the longest consecutive time slice with a value greater than or equal to 0.5 in the sequence as 7, the calculated evolutionary continuity rate is 7 ÷ 10 ≈ 0.7. The preset consistency fusion weights are 0.6 for the effective matching rate and 0.4 for the evolutionary continuity rate. A weighted summation is performed: 0.6 × 0.6 + 0.7 × 0.4 = 0.64. This result is already within the range of 0 to 1, requiring no additional normalization.

[0118] Optionally, after obtaining the spatiotemporal evolution consistency index, the monitoring and early warning device generates the comprehensive early warning command based on the spatiotemporal evolution consistency index, specifically as in steps 5051 to 5052.

[0119] Based on the mechanism of multimodal data spatiotemporal consistency analysis, this invention effectively distinguishes between real fires (where visual and sensor evolution is highly synchronized) and false alarms caused by environmental interference (where visual and sensor evolution is divergent or asynchronous). Thus, it automatically generates comprehensive early warning commands with high confidence without human intervention, solving the problems of high false alarm rates and verification difficulties caused by the inability to cross-verify traditional single-source alarms, and improving the intelligent judgment level and emergency response efficiency of fire remote monitoring systems.

[0120] Optionally, the process of steps 5051 to 5052 includes: Step 5051: Based on the spatiotemporal evolution consistency index, determine the confidence level of the current alarm event, and extract core spatial positioning data and associated video evidence data from the real disaster voxel set.

[0121] Optionally, the monitoring and early warning device acquires the spatiotemporal evolution consistency index and compares it with a preset confidence level judgment threshold range. The index is then compared with the preset confidence level judgment threshold range to determine the confidence level of the current alarm event. The preset confidence level judgment thresholds include a high confidence range, a medium confidence range, and a low confidence range. If the spatiotemporal evolution consistency index falls into the high confidence range (e.g., greater than or equal to 0.8), the current alarm event is judged to be of high confidence; if it falls into the medium confidence range (e.g., greater than or equal to 0.5 and less than 0.8), it is judged to be of medium confidence; and if it falls into the low confidence range (e.g., less than 0.5), it is judged to be of low confidence. The judgment result is recorded as a confidence level identifier.

[0122] Subsequently, the monitoring and early warning device extracts core spatial positioning data from the real disaster voxel set. This core spatial positioning data refers to a set of structured parameters that can uniquely identify the current spatial location and geometric extent of the disaster, including the disaster centroid coordinates, the disaster bounding box size, and a list of spatial unit node identifiers to which the disaster belongs. Among them, the disaster centroid coordinates refer to the three-dimensional coordinate point formed by the arithmetic mean of the spatial coordinates of all voxel units in the real disaster voxel set along the X, Y, and Z axes; the disaster bounding box size refers to the side lengths of the smallest axis-aligned cuboid containing all voxel units in the real disaster voxel set along the X, Y, and Z axes.

[0123] Simultaneously, the monitoring and early warning device also extracts related video evidence data. This video evidence data refers to video clips recorded within a continuous time window by surveillance cameras that have a line-of-sight relationship with the actual disaster voxel set, which, after cropping, retain only the visible area of ​​the disaster, along with their corresponding camera device identifiers, recording start and end timestamps, and original video stream storage indexes.

[0124] In one embodiment, it is assumed that the spatiotemporal evolution consistency index calculated in step 505 is 0.85. The monitoring and early warning device queries the preset threshold table and finds that 0.85 is greater than the high confidence threshold of 0.8, therefore determining the confidence level of the current alarm event as high confidence. Subsequently, the core spatial positioning data was extracted: the monitoring and early warning device traversed the spatial coordinates of all voxel units, calculated the arithmetic mean of the X-axis as 7.35 meters, the arithmetic mean of the Y-axis as 3.62 meters, and the arithmetic mean of the Z-axis as 2.08 meters, and obtained the disaster centroid coordinates (7.35, 3.62, 2.08); the minimum value of the X-axis direction was 6.8 meters and the maximum value was 7.9 meters, the minimum value of the Y-axis direction was 3.1 meters and the maximum value was 4.2 meters, and the minimum value of the Z-axis direction was 1.5 meters and the maximum value was 2.6 meters, and obtained the disaster bounding box size as 1.1 meters × 1.1 meters × 1.1 meters; the spatial unit node identifiers of all voxel units were extracted and deduplicated to obtain the node identifier list [S1-101, S1-102, S2-101-A]. Simultaneously, based on the line-of-sight association established in step 403, the surveillance camera associated with the aforementioned voxel set is identified as CAM-1F-EXH-03. Video clips of this camera, cropped from the target area within a continuous time window (2026-07-30T10:43:30+08:00 to 2026-07-30T10:44:00+08:00), are extracted from the video storage system and recorded with the storage index VID-20260730-104330-CAM1FEXH03-CLIP as video evidence data.

[0125] Step 5052: Based on the confidence level, the core spatial positioning data, the video evidence data, and the topological path information in the global spatial connectivity network, the comprehensive early warning instruction is obtained by encapsulation; the comprehensive early warning instruction includes the early warning level, three-dimensional location information, and spread path data.

[0126] Optionally, the monitoring and early warning device utilizes a global spatial connectivity network to retrieve the spatial unit node where the initial disaster source point is located and its downstream connected nodes. Based on the connectivity type identifier and flow capacity parameters of each node, it predicts the main spread direction and high-risk areas of the disaster medium in the near future. These predicted spread directions, high-risk node identifiers, and estimated arrival times are combined to form spread path data. Next, the monitoring and early warning device structurally encapsulates the confidence level, core spatial positioning data, spread path data, and video evidence data according to a preset data protocol format. After encapsulation, the resulting data structure is the comprehensive early warning command. This command contains all the key information needed for rapid assessment and emergency response by remote monitoring personnel.

[0127] In one embodiment, it is assumed that the node where the initial disaster source point is located is S1-101. The monitoring and early warning device takes the node where the initial disaster source point is located, S1-101, as the starting point and performs shortest path search with S1-101, S1-102, and S2-101-A in the node identifier list as the ending points respectively: the path to S1-101 is [S1-101], with a length of 0 hops; the path to S1-102 is [S1-101, S1-102], with a length of 1 hop; the path to S2-101-A is [S1-101, S2-101-A], with a length of 1 hop. The three paths are merged, deduplicated, and sorted by the number of hops to obtain the topology path information as [S1-101 (0 hops), S1-102 (1 hop), S2-101-A (1 hop)]. Subsequently, the encapsulation operation is performed: the high confidence level is written to the warning level field; the disaster centroid coordinates (7.35, 3.62, 2.08) and the disaster bounding box size of 1.1m × 1.1m × 1.1m are written to the 3D location information field; the aforementioned topological path information and video evidence data (camera identifier CAM-1F-EXH-03, storage index VID-20260730-104330-CAM1FEXH03-CLIP, recording time period 2026-07-30T10:43:30+08:00 to 2026-07-30T10:44:00+08:00) are written to the spread path data field. Finally, a complete comprehensive warning instruction is generated.

[0128] Furthermore, the urban fire remote monitoring and early warning device provided by the present invention will be described below. The urban fire remote monitoring and early warning device described below can be referred to in correspondence with the urban fire remote monitoring and early warning method described above.

[0129] Optional, refer to Figure 2 , Figure 2 This is a schematic diagram of the urban fire remote monitoring and early warning device provided by the present invention. The urban fire remote monitoring and early warning device includes: The basic environment construction module 210 is used to construct a digital twin scene containing spatial structural attributes and component medium attributes based on the building information model data of the building to be monitored, and to determine the visual basic environment based on the index association between fire-fighting equipment and three-dimensional spatial location in the digital twin scene. The network construction module 220 is used to identify spatial channels that allow the flow of media based on the semantic attributes of building components in the visualized basic environment, and to construct a global spatial network. The line-of-sight mapping module 230 is used to respond to the alarm signal of the fire control panel and establish the line-of-sight relationship between the video pixel area and the three-dimensional voxel unit based on the starting disaster source point located in the visualization base environment and the monitoring camera parameters. The conversion and separation module 240 is used to convert and separate dynamically changing image regions in a real-time video stream based on line-of-sight relationships and a global spatial connectivity network, to obtain a set of real disaster voxels that are connected to the initial disaster source point by a path. The early warning decision and push module 250 is used to generate a comprehensive early warning command based on the spatiotemporal diffusion trend of the real disaster voxel set and the numerical change trend of the alarm signal of the fire control panel, and push the comprehensive early warning command to the remote monitoring terminal.

[0130] The embodiments of the present invention realize in-depth collaborative verification of alarm signals, three-dimensional spatial structure and video visual content, which solves the problems of false alarms being difficult to eliminate and remote verification being difficult due to the lack of spatial connectivity and comprehensive analysis of visual context in the background technology, and improves the early warning accuracy and emergency response efficiency of fire remote monitoring.

[0131] Please see Figure 3 , Figure 3 An embodiment diagram of an electronic device provided in accordance with the present invention. For example... Figure 3 As shown, an electronic device 300 provided in this embodiment of the invention includes a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, it implements the processes of steps 10 to 50.

[0132] Please see Figure 4 , Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with an embodiment of the present invention is shown. Figure 4As shown, this embodiment provides a computer-readable storage medium 400 on which a computer program 311 is stored. When the computer program 311 is executed by a processor, it implements the processes of steps 10 to 50.

[0133] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the urban fire remote monitoring and early warning method provided by the above methods, which includes steps 10 to 50.

[0134] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

Claims

1. A method for remote monitoring and early warning of urban fires, characterized in that, include: A digital twin scene containing spatial structural attributes and component media attributes is constructed based on the building information model data of the building to be monitored. The basic environment for visualization is determined based on the index association between fire-fighting equipment and three-dimensional spatial location in the digital twin scene. Based on the semantic attributes of building components in the visualized basic environment, spatial channels that allow the flow of media are identified, and a global spatial connectivity network is constructed. In response to the alarm signal from the fire alarm control panel, a line-of-sight relationship is established between the video pixel area and the three-dimensional voxel unit based on the initial disaster source point located in the visualized basic environment and the parameters of the monitoring camera. Based on line-of-sight relationships and global spatial connectivity networks, dynamic image regions in real-time video streams are transformed and separated to obtain a set of real disaster voxels that are connected to the initial disaster source point by a path. Based on the spatiotemporal diffusion trend of the real disaster voxel set and the numerical change trend of the alarm signal of the fire control panel, a comprehensive early warning command is generated and pushed to the remote monitoring terminal.

2. The urban fire remote monitoring and early warning method according to claim 1, characterized in that, The method of identifying spatial channels that allow media flow based on the semantic attributes of building components in the visualized basic environment, and constructing a global spatial connectivity network, includes: Based on the geometric data and semantic tags of building components in the visualization environment, different types of spatial areas inside the building to be monitored are identified, and a complete set of spatial nodes containing a first type of spatial unit and a second type of spatial unit is generated. The first type of spatial unit is the main activity area defined by the physical enclosure structure, and the second type of spatial unit is the auxiliary circulation area defined by the gaps in the building structure or the equipment mezzanine. Based on the geometric adjacency relationship of each spatial unit in the complete set of spatial nodes and the semantic attributes of boundary components, horizontal connecting edges are established between adjacent spatial units with medium flow conditions to obtain the basic horizontal topology. Based on the semantic tags of vertically penetrating components in the visualized basic environment, the cross-layer physical interface connecting the first type of spatial unit and the second type of spatial unit is determined, and the spatial location and associated spatial unit identifier of each cross-layer physical interface are extracted to obtain the vertical penetration interface set. Based on the set of vertical penetration interfaces, a vertical coupling edge is established between the first type of spatial unit and the second type of spatial unit with vertical connectivity to obtain a cross-level connection structure. Based on the aforementioned basic horizontal topology and the aforementioned cross-level connection structure, and combined with the semantic information and path data of fluid transport equipment in the visualized basic environment, a global spatial connectivity network is constructed.

3. The urban fire remote monitoring and early warning method according to claim 2, characterized in that, Based on the basic horizontal topology and the cross-level connection structure, and combined with the semantic information and path data of fluid transport equipment in the visualized basic environment, a global spatial connectivity network is constructed, including: Based on the semantic information and path data of the fluid transport equipment in the visualized basic environment, a forced medium transport channel that runs through multiple spatial units is determined, and each forced medium transport channel is determined as a unidirectional forced flow edge connecting the starting spatial unit and the ending spatial unit based on the medium flow direction, thus obtaining the mechanical link structure. Based on the aforementioned basic horizontal topology, the aforementioned cross-level connection structure, and the aforementioned mechanical link structure, a hybrid directed graph is constructed, which includes different types of spatial unit nodes, horizontal connected edges, vertical coupled edges, and unidirectional forced flow edges. Based on the connectivity properties of each node in the hybrid directed graph, isolated nodes and invalid terminal nodes are pruned to obtain a purified effective topological subgraph. Based on the effective topological subgraph, a spatial type identifier is added to each node, and a connectivity type identifier and flow capacity parameter are added to each edge to obtain the global spatial connectivity network.

4. The urban fire remote monitoring and early warning method according to claim 1, characterized in that, The obtained set of real disaster voxels that are path-connected to the initial disaster source point includes: Based on the real-time video stream from the surveillance camera, pixel regions in the image that exhibit dynamic changing characteristics are identified, and a dynamic feature pixel set is generated. Based on the line-of-sight relationship, each pixel in the dynamic feature pixel set is traversed, and the pixel is mapped to the candidate voxel unit in the three-dimensional space of the visualization base environment to obtain the initial candidate voxel set. Based on the solid properties of building components in the visualized basic environment, the line-of-sight reachability of the initial candidate voxel set is verified to obtain the line-of-sight reachable voxel set. Based on the global spatial connectivity network, all spatial unit nodes with topological connectivity paths to the spatial unit nodes where the initial disaster source point is located are searched to obtain the source point connected voxel set. The intersection of the line-of-sight reachable voxel set and the source point connected voxel set is determined as the preliminary screening voxel set. The authenticity is verified by combining the preliminary selected voxel set with the adjacency relationships in the global spatial connectivity network to obtain the real disaster voxel set.

5. The urban fire remote monitoring and early warning method according to claim 4, characterized in that, The process of verifying the authenticity of the actual disaster voxels by combining the preliminary selected voxel set with the adjacency relationships in the global spatial connectivity network, includes: Based on the adjacency relationships in the global spatial connectivity network, the spatial continuity of the initially screened voxel set is verified to obtain a spatially continuous voxel set. Based on the medium penetration attribute identifier of the voxel units in the spatial continuous voxel set, voxel units that allow medium flow are screened to obtain a medium flow voxel set. Based on the medium flow voxel set, the spatial coordinates and attribute information of the voxel units therein are encapsulated to obtain the real disaster voxel set that has a path connection with the starting disaster source point.

6. The urban fire remote monitoring and early warning method according to claim 1, characterized in that, The comprehensive early warning command is generated based on the spatiotemporal diffusion trend of the real disaster voxel set and the numerical change trend of the fire alarm signal from the fire control panel, including: Based on the spatial distribution data of the real disaster voxel set within a continuous time window, and combined with the spatial security attributes of the nodes in the global spatial connectivity network, the spatial threat index of the real disaster voxel set within each time slice is determined, and a path threat sequence arranged in chronological order is generated based on the spatial threat index. Based on the values ​​of the corresponding time slices in the continuous time window associated with the initial disaster source in the alarm signal of the fire control panel, a sensor value sequence arranged in chronological order is generated. Based on the variation characteristics of values ​​between adjacent time slices in the path threat sequence, each time slice is marked with a visual change pattern to obtain a visual pattern sequence arranged in chronological order. Based on the variation characteristics of values ​​between adjacent time slices in the sensor value sequence, each time slice is marked with a sensor change pattern to obtain a sensor pattern sequence arranged in chronological order. Based on the visual morphology sequence and the sensor morphology sequence, the morphology matching degree of the two sequences in the time dimension is calculated according to the preset morphology matching rules to obtain the morphology matching score sequence. Based on the effective matching degree in the morphological matching score sequence, a spatiotemporal evolution consistency index is generated to characterize the consistency between visual space expansion and electrical signal energy release evolution, and the comprehensive early warning instruction is generated based on the spatiotemporal evolution consistency index.

7. The urban fire remote monitoring and early warning method according to claim 6, characterized in that, The generation of the comprehensive early warning instruction based on the spatiotemporal evolution consistency index includes: Based on the spatiotemporal evolution consistency index, the confidence level of the current alarm event is determined, and core spatial positioning data and associated video evidence data are extracted from the real disaster voxel set. The comprehensive early warning instruction is obtained by encapsulating the confidence level, the core spatial positioning data, the video evidence data, and the topological path information in the global spatial connectivity network; the comprehensive early warning instruction includes the early warning level, three-dimensional location information, and spread path data.

8. A remote monitoring and early warning device for urban fire protection, characterized in that, The method for remote monitoring and early warning of urban fire protection as described in any one of claims 1 to 7; the remote monitoring and early warning device for urban fire protection includes: The basic environment construction module is used to construct a digital twin scene containing spatial structural attributes and component media attributes based on the building information model data of the building to be monitored, and to determine the visualization basic environment based on the index association between fire-fighting equipment and three-dimensional spatial location in the digital twin scene. The connectivity network construction module is used to identify spatial channels that allow the flow of media based on the semantic attributes of building components in the visualized basic environment, and to construct a global spatial connectivity network. The line-of-sight mapping module is used to respond to the alarm signal of the fire control panel and establish the line-of-sight relationship between the video pixel area and the three-dimensional voxel unit based on the starting disaster source point located in the visualization base environment and the monitoring camera parameters. The conversion and separation module is used to convert and separate dynamically changing image regions in a real-time video stream based on line-of-sight relationships and a global spatial connectivity network, to obtain a set of real disaster voxels that are connected to the initial disaster source point by a path. The early warning decision and push module is used to generate comprehensive early warning instructions based on the spatiotemporal diffusion trend of the real disaster voxel set and the numerical change trend of the alarm signal of the fire control panel, and push the comprehensive early warning instructions to the remote monitoring terminal.

9. An electronic device, characterized in that, include: Memory, used to store computer software programs; A processor is configured to read and execute the computer software program, wherein when the processor executes the computer software program, it implements the urban fire remote monitoring and early warning method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The storage medium stores a computer software program, which, when executed by a processor, implements the urban fire remote monitoring and early warning method as described in any one of claims 1 to 7.

Images