Video stream intelligent analysis method and system in smart fire monitoring platform

Abstract

The present application relates to the technical field of computer vision and video analysis, in particular to a video stream intelligent analysis method and system in a smart fire monitoring platform; the method comprises: acquiring multiple monitoring video streams of the same target monitoring scene, identifying combustible components and fixed reference objects, and constructing a component reference linking table; generating directional migration edges based on the component migration relationship between the combustible components, establishing an ignition migration three-dimensional network; establishing a fixed reference object time-sharing blanking table for the blanking state data of the fixed reference objects in consecutive time slices, determining a target reference object, establishing a component blanking succession table, generating a sequential correspondence relationship and calculating a migration consistency value and a blanking succession consistency value; generating a target fire condition judgment result based on the foregoing results and outputting the result to the smart fire monitoring platform. The present application realizes the structured identification, sequential correspondence analysis and unified fire condition judgment of the fire condition evolution process in multiple monitoring video streams.

Description

Technical Field

[0001] This invention relates to the field of computer vision and video analysis technology, specifically to a video stream intelligent analysis method and system in a smart fire monitoring platform. Background Technology

[0002] With the continuous application of video surveillance, computer vision, and intelligent analysis technologies in fire protection scenarios, smart fire monitoring platforms have gradually evolved from simple video access and viewing to a technical system that combines multiple monitoring video streams for fire identification, status analysis, and platform alarms. In actual deployment, monitoring points typically cover computer rooms, power distribution rooms, storage areas, equipment mezzanines, corridors, and areas with dense components. By analyzing target objects, local area changes, and continuous time sequence information in the monitoring videos, the system can automatically determine the fire occurrence process, thereby improving the continuity and automation of fire monitoring. In the process of fire analysis based on multiple video streams in the intelligent fire monitoring platform, there are still problems such as difficulty in timely identification of early fires and difficulty in uniformly determining the evolution relationship of fires under multiple cameras. Specifically, in the early stages of a fire, such as in the computer room, power distribution room, warehouse shelving area, adjacent areas of shielding components, and areas connected between components, the fire often does not first form a smoke or flame target with clear boundaries and stable shape. Instead, it first manifests as local visibility changes, boundary fading, obscuring expansion, and continuous disappearance of fixed reference areas. This results in the monitoring video showing the temporal changes of local reference objects rather than the directly identifiable flame body, which can easily cause a lag in the identification of the real fire. On the other hand, when multiple cameras jointly cover the same target monitoring scene, the impact of the same fire on different cameras usually has differences in perspective, time, and component position. The local disappearance phenomenon appearing in different video streams often corresponds to different combustible components and different time slices, making it difficult to uniformly restore the ignition migration relationship between various combustible components and their corresponding temporal sequence relationship. This can easily lead to unstable fire judgment, inaccurate positioning, and inconsistent alarm timing. Summary of the Invention

[0003] The purpose of this invention is to provide a video stream intelligent analysis method and system in a smart fire monitoring platform to solve the problems mentioned in the background art, such as the difficulty in timely identification of early fires and the difficulty in uniformly determining the evolution relationship of fires under multiple cameras.

[0004] To achieve the above objectives, the technical solution of the present invention is: a video stream intelligent analysis method in a smart fire monitoring platform, comprising: S1. Acquire multiple monitoring video streams of the same target monitoring scene, identify flammable components and fixed reference objects in the multiple monitoring video streams, assign a unique number to the same flammable component in different cameras, perform attachment labeling processing on the fixed reference object and the corresponding flammable component, and construct a component reference attachment table. Among them, the fixed reference object is a reference object in the target monitoring scene whose position remains fixed, which can be continuously identified in the video stream and used to characterize local visibility changes; the component reference hooking table is a hooking relationship data table that shows the correspondence between the fixed reference object and the combustible component. S2. Under a pre-defined unified spatial coordinate system, determine the component migration relationship between each combustible component based on all combustible components, generate directional migration edges between combustible components with component migration relationships, and establish an ignition migration three-dimensional network. Among them, the directional migration edge is a directed connection edge that represents the migration direction relationship between two combustible components; the ignition migration three-dimensional network is composed of combustible components and directional migration edges, and is used to represent the network structure of ignition migration relationship in the target monitoring scene; S3. Divide the multi-channel monitoring video stream into continuous time slices, extract the blanking status data of the fixed reference object in each time slice, establish a time-sharing blanking table for the fixed reference object, determine the combustible component and directional migration edge corresponding to the target reference object based on the time-sharing blanking table of the fixed reference object and the component reference connection table, and establish a component blanking continuity table for the target reference object; extract the directional migration sequence of the target reference object from the ignition migration three-dimensional network, extract the time slice continuity sequence from the component blanking continuity table, and establish the sequence correspondence between the directional migration sequence and the time slice continuity sequence based on the component reference connection table, and calculate the migration consistency value and blanking continuity consistency value; Among them, the fixed reference object time-sharing blanking table is a time-series data table that represents the blanking change state of the fixed reference object within a continuous time slice; the component blanking continuity table is a data table that represents the blanking continuity relationship of the combustible component corresponding to the target reference object within a continuous time slice; the target reference object is the reference object among the fixed reference objects that participates in the current fire situation judgment; S4. Based on the migration consistency value, the blanking continuity consistency value and the blanking record corresponding to the target reference object, perform fire judgment on the target monitoring scene, obtain the target fire judgment result, and feed the target fire judgment result back to the smart fire monitoring platform in real time. Among them, the blanking record corresponding to the target reference object is the continuous time slice blanking change record corresponding to the target reference object in the fixed reference object time-sharing blanking table, which is used to represent the time sequence record data of the blanking change process of the target reference object in the continuous time slice.

[0005] Preferably, in S1, the combustible component is a fixed entity in the target monitoring scene that can form an ignition and transmission path when a fire occurs; the fixed reference object is used to characterize the local visibility changes of the adjacent area of ​​the corresponding combustible component; the selection rules for the fixed reference object are as follows: it maintains its position unchanged in multiple monitoring video streams, has continuous and identifiable boundary features, is located in the surface area of ​​the combustible component, the boundary adjacent area, or the connected area between components, has a comparable visibility state in continuous time slices, and can establish a stable correspondence with the same combustible component in the corresponding images of different cameras.

[0006] Preferably, in step S1, the attachment annotation process is an association annotation method based on spatial adjacency and boundary correspondence, used to associate and mark the position of a fixed reference object with the corresponding combustible component to construct a component reference attachment table. The component reference attachment table is used to record the correspondence between the fixed reference object and the combustible component, and its data structure includes reference object identifier, component identifier, attachment position, and attachment relationship. The component reference attachment table is also used to characterize the boundary correspondence and connected area association between the fixed reference object and the combustible component. The attachment relationship rules are as follows: when the fixed reference object is located in the surface area or boundary adjacency area of ​​a single combustible component, the fixed reference object is attached to the combustible component; when the fixed reference object is located in the adjacent boundary areas of two or more combustible components, the fixed reference object is attached to the corresponding combustible components respectively; when the fixed reference object is located in the connected area between two combustible components, the fixed reference object is attached to the combustible components on both sides of the connected area and the connected area position is marked in the attachment position.

[0007] Preferably, in step S2, the preset unified spatial coordinate system is a coordinate reference system that uniformly represents the positions of flammable components and fixed reference objects in multiple monitoring video streams, used to establish a unified spatial position correspondence for the same flammable component in different cameras; the component migration relationship is the directed association relationship between various flammable components on the ignition transmission path, used to represent the adjacent transmission order between flammable components; the specific rules for determining the component migration relationship are as follows: under the preset unified spatial coordinate system, first determine whether there is a spatial adjacency relationship or a connected region association relationship between two flammable components, then determine whether there is a boundary correspondence relationship and an obstruction discontinuity relationship between two flammable components, and finally determine the migration direction based on the position of the connected region between the two flammable components. When the aforementioned conditions are met, it is determined that there is a component migration relationship between the two flammable components; wherein, the obstruction discontinuity relationship is a relationship in which there is obstruction between two flammable components but a migration direction judgment can still be formed.

[0008] Preferably, in S2, the method for generating directional migration edges specifically involves: determining the starting combustible component with a component migration relationship as the edge starting point, determining the ending combustible component with a migration direction corresponding to the starting combustible component as the edge ending point, establishing a unidirectional connection between the edge starting point and the edge ending point to generate directional migration edges; the ignition migration three-dimensional network is a directed network structure composed of combustible components as nodes and directional migration edges as edges; the ignition migration relationship refers to the component migration relationship where the starting combustible component points to the ending combustible component along the migration direction; the specific structure of the ignition migration three-dimensional network is a directed graph storage structure that uses the unique number of the combustible component as the node index, the unique number of the directional migration edge as the edge index, and is organized according to the association order of nodes and edges; the specific method for establishing the ignition migration three-dimensional network is: generating node records based on the unique number of each combustible component, generating directional migration edge records based on the migration relationship of each component, establishing an association between each directional migration edge and its corresponding edge starting point and edge ending point, and uniformly mapping the node records corresponding to the same combustible component in different cameras to obtain the ignition migration three-dimensional network.

[0009] Preferably, in step S3, the blanking state data is time-series data used to characterize the degree of change in the visible state of a fixed reference object within a corresponding time slice, used to record the blanking change process of the fixed reference object within consecutive time slices; the fixed reference object time-sharing blanking table is a time-series record table established based on the association between fixed reference objects and time slices, used to record the blanking change records of each fixed reference object within consecutive time slices and support the selection of target reference objects; the data structure of the fixed reference object time-sharing blanking table is specifically a time-series table structure, including camera identifier, reference object identifier, time slice identifier, blanking state value, and blanking change identifier; the specific method for establishing the fixed reference object time-sharing blanking table is as follows: the multiple monitoring video streams are divided according to consecutive time slices, the blanking state data of the corresponding time slice is extracted for each fixed reference object in each camera, and the data is collected according to the camera identifier, reference object identifier, and time slice identifier to generate the fixed reference object time-sharing blanking table.

[0010] Preferably, in step S3, the target reference is a reference object among the fixed references that exhibits a blanking change within a continuous time slice and can identify the corresponding combustible component and directional migration edge. It is used to participate in the establishment of the component blanking continuation table and the generation of sequential correspondence. The specific selection rule for the target reference is as follows: determine the fixed reference with a continuous time slice blanking change record in the fixed reference time-sharing blanking table, determine the corresponding combustible component based on the component reference attachment table, determine the corresponding directional migration edge based on the ignition migration three-dimensional network, and select the fixed reference that simultaneously satisfies the continuous time slice blanking change, has an attachment relationship, and has a corresponding directional migration edge as the target reference.

[0011] Preferably, in step S3, the component blanking continuation table is a continuation relationship table formed by mapping the blanking records of the target reference object to the corresponding combustible component and directional migration edge. It is used to characterize the blanking continuation relationship between the combustible component and directional migration edge corresponding to the target reference object in continuous time slices. The data structure of the component blanking continuation table is specifically a continuation relationship table structure, including component identifier, directional migration edge identifier, reference object identifier, start time slice identifier, end time slice identifier, blanking state value, and continuation order identifier. The specific rules and methods for establishing the component blanking continuation table are as follows: read the continuous time slice blanking records of the target reference object in the fixed reference object time-sharing blanking table, map each blanking record to the corresponding combustible component based on the component reference attachment table, map each blanking record to the corresponding directional migration edge based on the ignition migration three-dimensional network, sort the mapped records according to the order of time slices, and continuously merge the blanking records belonging to the same combustible component or the same directional migration edge in adjacent time slices to form the component blanking continuation table.

[0012] Preferably, in S3, the directional migration order is the arrangement order of the combustible components and directional migration edges corresponding to the target reference object along the migration direction in the ignition migration three-dimensional network, and the time slice succession order is the arrangement order of the combustible components and directional migration edges corresponding to the target reference object in the component blanking succession table according to the time slice sequence; the sequential correspondence between the directional migration order and the time slice succession order is the sequential matching relationship of the same combustible components and the same directional migration edges in the two arrangement orders, used to characterize the consistency between the spatial migration process and the time succession process; the specific rule method for establishing the sequential correspondence is: based on the component reference hooking table, the combustible components and directional migration edges in the directional migration order are matched with the combustible components and directional migration edges in the time slice succession order, and the sequential matching relationship between adjacent sequential records is generated according to the migration direction and the time slice sequence. The migration consistency value is a numerical value characterizing the degree of consistency between the directional migration sequence and the time slice succession sequence in the migration direction of combustible components, used to characterize the matching degree of spatial migration relationships; the blanking succession consistency value is a numerical value characterizing the degree of continuity between the time slice succession sequence and the directional migration sequence within consecutive time slices, used to characterize the matching degree of the blanking advancement process; the specific methods for calculating the migration consistency value and the blanking succession consistency value are as follows: count the number of combustible components and directional migration edge corresponding records with consistent sequential positions in the sequential correspondence relationship, and calculate the normalized ratio with the total number of corresponding records in the directional migration sequence to obtain the migration consistency value; count the number of combustible components and directional migration edge corresponding records that maintain continuity in consecutive time slices in the sequential correspondence relationship, and calculate the normalized ratio with the total number of corresponding records in the time slice succession sequence to obtain the blanking succession consistency value.

[0013] On the other hand, the present invention provides a video stream intelligent analysis system in a smart fire monitoring platform, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the video stream intelligent analysis method in the smart fire monitoring platform described above.

[0014] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects: 1. In this invention, based on the ignition migration three-dimensional network, the fixed reference object time-sharing blanking table, and the sequential correspondence between the directional migration sequence and the time slice sequence, the scattered local blanking changes in multiple monitoring video streams can be correlated with the ignition migration relationship between combustible components. It no longer relies solely on the appearance of smoke or the shape of flames in a single image for judgment. Thus, even when a stable smoke and fire target has not yet formed in the early stage of a fire, the evolution of the fire can still be identified based on the continuous blanking changes of fixed reference objects, thereby improving the effectiveness and stability of early fire judgment. 2. In this invention, by uniformly recording the blanking change process of fixed reference objects under different cameras, and calculating the migration consistency value and blanking continuity consistency value in combination with the migration direction relationship between combustible components, a unified analysis of dispersed and asynchronous local changes under multiple cameras can be achieved. This can distinguish the difference between the real ignition migration process and irrelevant local obscuring and short-term image disturbance, thereby improving the consistency of fire judgment results, positioning accuracy and alarm reliability in multi-channel video stream scenarios. Attached Figure Description

[0015] Figure 1 This is a flowchart of an embodiment of the present invention. Detailed Implementation

[0016] Example 1, as Figure 1 As shown, the intelligent video stream analysis method in the intelligent fire monitoring platform proposed in this invention has the following specific implementation steps: S1. Acquire multiple monitoring video streams of the same target monitoring scene, identify flammable components and fixed reference objects in the multiple monitoring video streams, assign a unique number to the same flammable component in different cameras, perform attachment labeling processing on the fixed reference object and the corresponding flammable component, and construct a component reference attachment table. Among them, the fixed reference object is a reference object in the target monitoring scene whose position remains fixed, which can be continuously identified in the video stream and used to characterize local visibility changes; the component reference hooking table is a hooking relationship data table that shows the correspondence between the fixed reference object and the combustible component. S2. Under a pre-defined unified spatial coordinate system, determine the component migration relationship between each combustible component based on all combustible components, generate directional migration edges between combustible components with component migration relationships, and establish an ignition migration three-dimensional network. Among them, the directional migration edge is a directed connection edge that represents the migration direction relationship between two combustible components; the ignition migration three-dimensional network is composed of combustible components and directional migration edges, and is used to represent the network structure of ignition migration relationship in the target monitoring scene; S3. Divide the multi-channel monitoring video stream into continuous time slices, extract the blanking status data of the fixed reference object in each time slice, establish a time-sharing blanking table for the fixed reference object, determine the combustible component and directional migration edge corresponding to the target reference object based on the time-sharing blanking table of the fixed reference object and the component reference connection table, and establish a component blanking continuity table for the target reference object; extract the directional migration sequence of the target reference object from the ignition migration three-dimensional network, extract the time slice continuity sequence from the component blanking continuity table, and establish the sequence correspondence between the directional migration sequence and the time slice continuity sequence based on the component reference connection table, and calculate the migration consistency value and blanking continuity consistency value; Among them, the fixed reference object time-sharing blanking table is a time-series data table that represents the blanking change state of the fixed reference object within a continuous time slice; the component blanking continuity table is a data table that represents the blanking continuity relationship of the combustible component corresponding to the target reference object within a continuous time slice; the target reference object is the reference object among the fixed reference objects that participates in the current fire situation judgment; S4. Based on the migration consistency value, the blanking continuity consistency value and the blanking record corresponding to the target reference object, perform fire judgment on the target monitoring scene, obtain the target fire judgment result, and feed the target fire judgment result back to the smart fire monitoring platform in real time. Among them, the blanking record corresponding to the target reference object is the continuous time slice blanking change record corresponding to the target reference object in the fixed reference object time-sharing blanking table, which is used to represent the time sequence record data of the blanking change process of the target reference object in the continuous time slice.

[0017] In this embodiment S1, the combustible component is a fixed entity in the target monitoring scene that can form an ignition and transmission path when a fire occurs; the fixed reference object is used to characterize the local visibility changes of the adjacent area of ​​the corresponding combustible component; the selection rules for the fixed reference object are as follows: it maintains its position unchanged in multiple monitoring video streams, has continuous and identifiable boundary features, is located in the surface area of ​​the combustible component, the boundary adjacent area, or the connected area between components, has a comparable visibility state in continuous time slices, and can establish a stable correspondence with the same combustible component in the corresponding images of different cameras.

[0018] In this embodiment S1, for multiple monitoring video streams accessed in the same target monitoring scene, video preprocessing is first performed to form standardized video data that can be used for subsequent identification of combustible components and fixed reference objects. The video preprocessing includes time alignment processing, image cropping processing, frame extraction processing, brightness normalization processing, contrast equalization processing, noise suppression processing, slight jitter correction processing, occlusion area removal processing, and invalid image filtering processing. Among them, time alignment processing is used to align video frames captured by different cameras according to a unified timeline, so that the images from different cameras at the same time can be analyzed together; image cropping processing is used to remove the monitoring interface border, timestamp overlay area, platform watermark area, and irrelevant background area, retaining the effective monitoring area of ​​the target monitoring scene; frame extraction... The preprocessing steps are as follows: Image frame sequences that meet the analysis frequency requirements are extracted from the continuous video stream; brightness normalization and contrast equalization are used to eliminate grayscale shifts caused by exposure differences between different cameras; noise suppression is used to reduce the interference of compression noise, fixed noise, and local speckle on boundary recognition; slight jitter correction is used to align the images with the small displacements generated by fixed cameras during long-term operation; occlusion removal is used to mark lens obstructions, fixed obstructions, and long-term invisible areas; invalid image filtering is used to remove completely black frames, severely blurred frames, and blank frames without effective monitoring content. After completing the above preprocessing, the retained image frame sequences are grouped according to camera identifier, time identifier, and scene identifier to form a multi-channel monitoring video dataset belonging to the same target monitoring scene.

[0019] In this embodiment S1, combustible component identification and fixed reference object identification are performed on the preprocessed multi-channel monitoring video dataset. Combustible component identification uses fixed entity objects in the target monitoring scene as the identification object. First, the boundary of the entity contour in each camera image is extracted. Then, candidate regions are screened based on boundary continuity, structural integrity, positional stability, and scene semantic attributes. Candidate regions that meet the characteristics of fixed entities and are located in the possible fire transmission path are marked as combustible component candidates. Subsequently, the observation results of each camera on the same area are combined to cross-verify the contour, boundary, adjacent position, and connectivity of the same entity under different viewpoints. Candidates that pass the verification are confirmed as combustible components. Fixed reference object identification uses reference objects that can continuously represent local visibility changes within a continuous time slice as the identification object. First, edge segments and corner points are extracted from each camera image. Stable visual elements such as outlines, text borders, sign borders, cabinet door edges, door frame edges, frame corners, vent edges, and stable textures on equipment surfaces are selected as fixed reference objects based on rules that ensure their fixed position, continuous and identifiable boundaries, comparable visibility within consecutive time slices, and stable correspondence with the surface area of ​​combustible components, adjacent boundary areas, or connected areas between components. During the identification process, the camera position, image coordinate range, adjacent combustible component identifiers, boundary type, and surface area type of each fixed reference object are recorded. Repeated observations of the same fixed reference object in adjacent cameras are uniformly verified to ensure that the definition of a fixed reference object is based on a stable object rather than a short-term moving target, a temporarily occluded target, or a random light and shadow area. After completing the identification of combustible components and fixed reference objects, a set of combustible components and a set of fixed reference objects are formed.

[0020] In this embodiment S1, cross-camera unique numbering processing is performed on the identified combustible components, and the selection rules for fixed reference objects are further implemented. The cross-camera unique numbering processing is carried out in the manner of "first local identification, then global merging". First, a temporary identifier is assigned to each combustible component in each camera. Then, the temporary identifiers of combustible components in the same spatial position, with the same boundary structure and consistent connectivity with the same adjacent components in different cameras are merged and assigned a unified unique number. During the merging process, the component boundary contour, relative positional relationship, distribution of adjacent fixed reference objects, and position of the connected area between components are used as consistency judgment criteria, so that the observation results of the same combustible component in multiple cameras correspond to the same number. The selection rules for fixed reference objects are implemented step by step. The verification process involves first verifying positional stability, then boundary continuity, then comparability of visible states, and finally, verifying the stability of the correspondence with combustible components. For fixed reference objects located on the surface of combustible components, priority is given to retaining objects with clear and stable edge lines, contour corners, or local textures over a long period. For fixed reference objects located in the adjacent areas of combustible component boundaries, priority is given to retaining linear or angular objects that can characterize changes in boundary visibility. For fixed reference objects located in the connected areas between components, priority is given to retaining objects that are near connected openings, transition boundaries, or connected paths and are continuously visible across multiple time slices. Candidate objects that do not meet the requirements for stable identification in continuous time slices, have significant positional drift, severely missing boundaries, or cannot establish a stable correspondence with a specific combustible component are not included in the fixed reference object set.

[0021] In this embodiment S1, the attachment annotation process is an association annotation method based on spatial adjacency and boundary correspondence. It is used to associate and mark the position of fixed reference objects with corresponding combustible components to construct a component reference attachment table. The component reference attachment table is used to record the correspondence between fixed reference objects and combustible components. Its data structure includes reference object identifier, component identifier, attachment position, and attachment relationship. The component reference attachment table is also used to characterize the boundary correspondence and connected area association between fixed reference objects and combustible components. The attachment relationship rules are as follows: when the fixed reference object is located in the surface area or boundary adjacency area of ​​a single combustible component, the fixed reference object is attached to the combustible component; when the fixed reference object is located in the adjacent boundary areas of two or more combustible components, the fixed reference object is attached to the corresponding combustible components respectively; when the fixed reference object is located in the connected area between two combustible components, the fixed reference object is attached to the combustible components on both sides of the connected area and the connected area position is marked in the attachment position.

[0022] In this embodiment S1, the annotation process is based on the spatial relationship between the set of combustible components, the set of fixed reference objects, and multiple monitoring videos. Specifically, it adopts a method of first locating, then associating, and then marking. First, for each fixed reference object, its image coordinate range, boundary contour range, and adjacent area range in the corresponding camera image are extracted, and the unique number information of the combustible components within a preset distance around the fixed reference object is read. Second, an association candidate set is established based on the spatial adjacency relationship and boundary correspondence relationship between the fixed reference object and the combustible component. Among them, the spatial adjacency relationship is used to determine the relative proximity between the fixed reference object and the combustible component in the image plane and scene space, and the boundary correspondence relationship is used to determine whether the boundary of the fixed reference object forms a stable correspondence with the surface boundary, outer edge boundary, or connected area boundary of the combustible component. Next, relationship marking is performed on each set of fixed reference objects and combustible components in the candidate set of associations. Relationship marking includes three types: surface area attachment, boundary adjacent area attachment, and connected area attachment. For surface area attachment, the fixed reference object is required to be located in the internal area of ​​the surface of a single combustible component or near the surface edge. For boundary adjacent area attachment, the fixed reference object is required to be located outside the boundary of the combustible component or in the boundary transition area, and to maintain stable adjacency with the boundary. For connected area attachment, the fixed reference object is required to be located in the connecting opening, transition channel, or boundary area between two combustible components, and to be able to reflect the local visibility changes of the combustible components on both sides. After completing the relationship marking, the corresponding combustible component number, attachment position, and attachment relationship are written for each fixed reference object, forming the original attachment annotation record required for the subsequent construction component reference attachment table.

[0023] In this embodiment S1, the construction of the component reference attachment table is based on the original attachment annotation record, and a table structure is used to store the correspondence between fixed reference objects and combustible components. The component reference attachment table includes at least four types of fields: reference object identifier, component identifier, attachment position, and attachment relationship. The reference object identifier uniquely points to a specific fixed reference object in the set of fixed reference objects, the component identifier uniquely points to a combustible component that has been uniquely numbered, the attachment position indicates the specific attachment location or area type of the fixed reference object relative to the combustible component, and the attachment relationship indicates whether the fixed reference object and the combustible component are attached to a surface area, a boundary adjacent area, or a connected area. During the construction process, the... When a fixed reference object corresponds to only one combustible component, a single connection record is written to the component reference connection table. When a fixed reference object corresponds to multiple combustible components, an independent connection record is written for each correspondence to maintain the decomposability of connection relationships during subsequent time-series analysis. When a fixed reference object is located in a connected region between two combustible components, a connected region position mark is written in the connection position field, and a connected region connection mark is written in the connection relationship field to distinguish it from simple surface region connection or boundary adjacent region connection. When the same fixed reference object is repeatedly observed across cameras, the component reference connection table retains a unified reference object identifier and the corresponding component identifier.

[0024] In this embodiment S1, the attachment rules of the component reference attachment table are executed according to the location type of the fixed reference object and the corresponding number of combustible components. When the fixed reference object is located on the surface area or boundary adjacent area of ​​a single combustible component, the fixed reference object is attached to the combustible component, and marked as surface area attachment or boundary adjacent area attachment according to the specific location. When the fixed reference object is located in the adjacent boundary area of ​​two or more combustible components at the same time, the fixed reference object is attached to each corresponding combustible component, and multiple attachment records are formed in the component reference attachment table. Each attachment record retains its own component identifier and attachment position, so that subsequent processing can... It is sufficient to distinguish the independent correspondence between the fixed reference object and different combustible components; when the fixed reference object is located in the connected area between two combustible components, the fixed reference object is simultaneously attached to the combustible components on both sides of the connected area, and the location of the connected area is marked in the attachment position and the connected area attachment is marked in the attachment relationship, so as to infer the transmission effect of the connected area on the visibility change based on the disappearance change of the fixed reference object; in addition, for fixed reference objects located in the boundary overlap area, the occlusion transition area or the local field of view boundary area, the attachment object is determined first based on the unique number of the combustible component, the continuity of the boundary contour and the stability of the adjacency relationship during the attachment annotation process.

[0025] In this embodiment S2, the preset unified spatial coordinate system is a coordinate reference system that uniformly represents the positions of flammable components and fixed reference objects in multiple monitoring video streams. It is used to establish a unified spatial position correspondence for the same flammable component in different cameras. The component migration relationship is the directed association between various flammable components on the ignition transmission path, used to represent the adjacent transmission order between flammable components. The specific rules for determining the component migration relationship are as follows: Under the preset unified spatial coordinate system, first determine whether there is a spatial adjacency relationship or a connected region association relationship between two flammable components, then determine whether there is a boundary correspondence relationship and an obstruction discontinuity relationship between two flammable components, and finally determine the migration direction based on the position of the connected region between the two flammable components. When the aforementioned conditions are met, it is determined that there is a component migration relationship between the two flammable components. Among them, the obstruction discontinuity relationship is a relationship in which there is obstruction between two flammable components but a migration direction judgment can still be formed.

[0026] In this embodiment S2, a pre-defined unified spatial coordinate system is established through a common reference area of ​​the multi-camera field of view. First, multiple scene reference objects that maintain fixed positions under long-term operating conditions and can be repeatedly observed by different cameras are selected within the target monitoring scene. The boundary contour points, corner positions, center positions, and adjacent boundary directions of each scene reference object are extracted as reference marker sets. Then, according to the camera identifiers, the pixel positions and boundary direction information of each reference marker set in the corresponding video frame are recorded. Corresponding registration is performed on the reference markers with the same name in different cameras, and the mapping parameters between pixel positions and scene entity positions are calculated. All combustible components and fixed reference objects in the same target monitoring scene are uniformly mapped to the same coordinate reference system. The pre-defined unified spatial coordinate system is established using the selected reference origin, reference direction, and reference scale within the target monitoring scene as coordinates. The baseline origin can be set at a fixed corner of the target monitoring scene, the intersection of ground lines, the starting position of the equipment array, or the starting position of the cabinet row. The baseline direction can be set according to the wall extension direction, equipment arrangement direction, the channel center extension direction, or the structural boundary direction. The baseline scale is set according to fixed distance markers in the scene, standard equipment dimensions, component spacing, or pre-measured distance. When performing coordinate transformation on the outline points of combustible components and fixed reference objects in each camera image, the mapping parameters corresponding to the camera are read first, and then the pixel coordinates of the outline points are converted into position coordinates in a unified spatial coordinate system. Multiple observation results of the same combustible component in different cameras are merged and registered according to position overlap, boundary correspondence, and adjacency consistency. Multiple observation results of the same fixed reference object in different cameras are uniformly registered according to position consistency and outline continuity.

[0027] In this embodiment S2, when determining the migration relationship between combustible components based on all combustible components, the unified number, outer contour range, boundary orientation, distribution of adjacent components, and location of connected regions of all combustible components are first read in a preset unified spatial coordinate system. For any two combustible components, a pairwise relationship determination is performed, and the determination process is divided into four steps: position proximity determination, boundary facing determination, connected region existence determination, and migration direction determination. Position proximity determination is completed by calculating the minimum boundary distance, center distance, and boundary projection overlap range of two combustible components. When the minimum boundary distance is within a preset adjacent range or there is a continuous intermediate region between two combustible components, the two combustible components are marked as a pair of position proximity components. Boundary facing determination is completed by reading the normal direction, boundary extension direction, and adjacent boundary length of the relative boundaries of two combustible components. When two combustible components have relative boundaries and the relative boundaries are close to each other, the boundary facing determination is completed. When maintaining a continuous correspondence, the two combustible components are marked as a boundary corresponding component pair. The existence of a connected region is determined by reading the connected region connection record in the component reference connection table, the location of the fixed reference object, and the range of the connected opening. When there is a connected region, boundary gap region, opening region, or transition channel region marked by the fixed reference object between two combustible components, the two combustible components are marked as a connected component pair. The migration direction is determined by comparing the relative position of the two combustible components in a unified spatial coordinate system, the direction of the connected region, the orientation of the boundary opening, and the arrangement order of the combustible components along the scene layout direction. The combustible component located on the starting side of the connected region, the input side of the boundary opening, or the pre-arrangement side is recorded as the starting combustible component, and the combustible component located on the ending side of the connected region, the output side of the boundary opening, or the post-arrangement side is recorded as the ending combustible component. The migration direction information is recorded between the corresponding component pairs.

[0028] In this embodiment S2, the component migration relationship is stored in a directed association form during recording. Each component migration relationship record corresponds to a set of starting combustible components, ending combustible components, location proximity determination results, boundary correspondence determination results, connected region existence determination results, and migration direction determination results. When two combustible components simultaneously meet the preset quantity requirements in the location proximity determination, boundary correspondence determination, and connected region existence determination, a component migration relationship record is written between the two combustible components. When there are multiple connected regions or multiple boundary correspondence areas between two combustible components, each connected region and each boundary correspondence area is determined and registered, and multiple candidate records are merged based on the consistency of the migration direction. The merged records retain unified identifiers for the starting and ending combustible components, along with the corresponding boundary area range and connected area location. For combustible component pairs located at corners, turning points, or obstruction intervals, their contour turning points, boundary polylines, and distribution of adjacent fixed reference objects in a unified spatial coordinate system are first read. Then, the migration direction is confirmed in conjunction with the arrangement order of adjacent components. Component pairs that meet the conditions of positional continuity and directional continuity are recorded as component pairs with component migration relationships. The recording results of component migration relationships are sorted and stored according to the order of the starting and ending combustible component numbers. Each record retains the relationship number, starting component number, ending component number, and corresponding relationship determination item.

[0029] In this embodiment S2, the method for generating directional migration edges is as follows: the starting combustible component with a component migration relationship is determined as the edge starting point, and the ending combustible component with a migration direction corresponding to the starting combustible component is determined as the edge ending point. A one-way connection is established between the edge starting point and the edge ending point to generate a directional migration edge. The ignition migration three-dimensional network is a directed network structure composed of combustible components as nodes and directional migration edges as edges. The ignition migration relationship refers to the component migration relationship where the starting combustible component points to the ending combustible component along the migration direction. The specific structure of the ignition migration three-dimensional network is a directed graph storage structure that uses the unique number of the combustible component as the node index, the unique number of the directional migration edge as the edge index, and is organized according to the association order of nodes and edges. The specific method for establishing the ignition migration three-dimensional network is as follows: node records are generated based on the unique number of each combustible component, directional migration edge records are generated based on the migration relationship of each component, each directional migration edge is associated with its corresponding edge starting point and edge ending point, and the node records corresponding to the same combustible component in different cameras are uniformly mapped to obtain the ignition migration three-dimensional network.

[0030] In this embodiment S2, the directional migration edge is generated based on the recorded component migration relationships. First, all component migration relationship records are read, and the starting and ending combustible components in each record are numbered and verified. The starting combustible component is taken as the edge start point, and the ending combustible component is taken as the edge end point. The corresponding directional migration edge is generated in a unidirectional connection manner. The edge attributes of the directional migration edge include edge number, edge start component number, edge end component number, edge direction identifier, boundary corresponding area identifier, and connected area identifier. The edge number is generated by combining the starting and ending combustible component numbers. The edge direction identifier is written according to the direction from the start point to the end point in a unified spatial coordinate system. The boundary corresponding area identifier... The boundary region corresponding to the boundary that triggers the directional migration edge is used to record the range of the boundary region corresponding to the boundary that triggers the directional migration edge. The connected region identifier is used to record the location of the connected region that triggers the directional migration edge. When there are multiple candidate migration relationship records between the same pair of starting combustible components and ending combustible components, the boundary region corresponding to each candidate record and the connected region are merged and registered, and a unified directional migration edge is retained. When the same combustible component points to multiple ending combustible components, multiple directional migration edges are generated respectively, and the corresponding edge direction identifier and region identifier are retained in the edge record respectively. When multiple starting combustible components point to the same ending combustible component, directional migration edges with each starting combustible component as the edge starting point are generated respectively, and the edge records are kept independently stored.

[0031] In this embodiment S2, the ignition migration three-dimensional network is established according to a directed graph storage structure, and its storage basis consists of a node table, an edge table, and a node-edge association table. The node table stores records of each combustible component in the order of its unique number. Each node record includes the component number, its position in a unified spatial coordinate system, the component outline range, the boundary direction, and the adjacent region identifier. The edge table stores records of each edge in the order of its unique number of the directional migration edge. Each edge record includes the edge number, the component number of the edge's starting point, the component number of the edge's ending point, the edge direction identifier, the boundary corresponding region identifier, and the connected region identifier. The node-edge association table records the set of incoming edges and outgoing edges connected to each node in the order of their node numbers. The set consists of an inbound edge set representing the set of directional migration edge numbers pointing to the current node, and an outbound edge set representing the set of directional migration edge numbers originating from the current node. During the writing process, all combustible component node records are first constructed based on the node table, and then all directional migration edge records are constructed based on the edge table. Subsequently, each edge record is written to the corresponding node's outbound edge set and inbound edge set according to the edge start and end point numbers. When multiple observation results corresponding to the same combustible component from different cameras have been merged into the same component number under a unified spatial coordinate system, only one node record is retained in the corresponding node table, and both the edge table and the node-edge association table use this unified component number for association storage.

[0032] In this embodiment S2, when establishing the ignition migration three-dimensional network, the unique numbering results of all combustible components and the migration relationship records of all components are read first to generate the initial records of the node table and edge table, and then the network legality verification process is performed. The legality verification process includes duplicate edge verification, isolated node verification, and direction conflict verification. Duplicate edge verification is used to identify duplicate directional migration edge records where the edge start point and edge end point are completely consistent. Isolated node verification is used to identify combustible component records that have not established an inbound edge association or outbound edge association with any directional migration edge. Direction conflict verification is used to identify the situation where there are opposite direction edge records between the same pair of combustible components. In duplicate edge verification, the edge start point and edge end point are the same and Multiple records with overlapping boundary regions are merged into one edge record; in isolated node verification, the retained isolated nodes are written into the node table but not registered in the node edge association table for outgoing or incoming edge sets; in direction conflict verification, the migration direction determination results and corresponding region identifiers between the same component pairs are read, and one edge record that satisfies the direction determination results of the unified spatial coordinate system is retained; after verification, the node table, edge table, and node edge association table are written into the ignition migration three-dimensional network database in a unified format. Each node record and each edge record in the network database can be directly indexed by a unique number, and each edge record can be traced back to the corresponding edge start point and edge end point through the node edge association table.

[0033] In this embodiment S3, the blanking state data is time-series data used to characterize the degree of change in the visible state of a fixed reference object within a corresponding time slice, and is used to record the blanking change process of the fixed reference object within consecutive time slices; the fixed reference object time-sharing blanking table is a time-series record table established based on the association between the fixed reference object and the time slice, used to record the blanking change records of each fixed reference object within consecutive time slices and support the selection of target reference objects; the data structure of the fixed reference object time-sharing blanking table is specifically a time-series table structure, including camera identifier, reference object identifier, time slice identifier, blanking state value, and blanking change identifier; the specific method for establishing the fixed reference object time-sharing blanking table is as follows: the multiple monitoring video streams are divided according to consecutive time slices, the blanking state data of the corresponding time slice is extracted for each fixed reference object in each camera, and the data is collected according to the camera identifier, reference object identifier, and time slice identifier to generate the fixed reference object time-sharing blanking table.

[0034] In this embodiment S3, when segmenting multiple monitoring video streams into continuous time slices, the camera identifier and time identifier formed in S1 for each monitoring video stream are first read. A unified time axis alignment process is then performed on the video frames of each camera within the same target monitoring scene, converting the frame timestamps in each video stream into time positions on the same time axis. Time slices are divided according to a preset time length, with each time slice corresponding to a time slice identifier. Each monitoring video stream retains the video frame group corresponding to that time slice within the same time slice. In cases of frame rate differences, video streams with frame rates higher than the analysis frequency undergo equal-interval frame extraction, while video streams with frame rates lower than the analysis frequency undergo adjacent frame preservation processing, ensuring that each time slice has a consistent frame rate. Fixed reference object observation frames are used for blanking state extraction. During time-slicing, the camera identifier, time-slice identifier, frame sequence number, fixed reference object identifier, and intra-frame position coordinates are retained within the same time-slice. Video frame groups are archived according to the camera identifier and time-slice identifier. For fixed reference object observation frames that cross the boundaries of two time-slices, they are assigned to the time-slice to which their timestamp belongs. For video streams with short-term frame drops, valid observation frames of the same fixed reference object in adjacent time-slices are recorded with padding marks. The padding marks are only used to indicate the source status of the fixed reference object observation data in that time-slice. After continuous time-slicing is completed, each fixed reference object corresponds to a set of observation frame sequences arranged according to the time-slice identifier in each video stream.

[0035] In this embodiment S3, when extracting the hidden surface disappearance state data of a fixed reference object in each time slice, the image coordinate range, boundary contour range, and camera identifier of each fixed reference object are first read according to the fixed reference object set generated in S1. The reference region corresponding to the fixed reference object is located in the video frame group of each time slice. The reference region location is performed by contour matching, edge alignment, and local texture matching. Contour matching is used to find the outer contour position of the fixed reference object in the current frame. Edge alignment is used to compare the degree of preservation between the current boundary and the reference boundary. Local texture matching is used to determine the identifiable state of the internal texture or boundary texture of the fixed reference object in the current time slice. The hidden surface disappearance state data includes the hidden surface disappearance state value and the hidden surface disappearance change indicator. The hidden surface disappearance state value is determined by the degree of preservation of the visible area of ​​the reference region. The degree of edge integrity preservation and local contrast preservation are normalized to obtain the blanking change indicator, which is used to record whether there is a change in the visible state of the current time slice relative to the previous time slice. When extracting blanking state data, the reference visible state of the fixed reference object in the reference time slice is read first, and then the current visible state in the current time slice is read. The two are compared by region overlap, edge preservation, and grayscale contrast to obtain region preservation record, edge preservation record, and contrast preservation record, respectively. Then, the three types of records are written into the blanking state data of the current time slice. When there are multiple valid observation frames of the same fixed reference object in the same time slice, the blanking state data of multiple valid observation frames are aggregated within the time slice, and the stable record of consecutive valid frames in the same time slice is taken as the blanking state data corresponding to that time slice.

[0036] In this embodiment S3, when establishing the fixed reference object time-sharing blanking table, a time series table structure is established using the camera identifier, reference object identifier, and time slice identifier as a joint index. The blanking status data of each fixed reference object in each time slice is written into the corresponding record. The data fields of the fixed reference object time-sharing blanking table include the camera identifier, reference object identifier, time slice identifier, blanking status value, and blanking change identifier. The camera identifier is used to point to the video stream source of the observation frame of the fixed reference object, the reference object identifier is used to point to the fixed reference object that has been identified and completed the attachment annotation in S1, the time slice identifier is used to indicate the position of the continuous time slice to which the record belongs, and the blanking status value is used to record the fixed reference object in the corresponding time. The degree of visible state change within a time slice, and the blanking change marker are used to record whether there is a continuous blanking change between adjacent time slices; the fixed reference object time-sharing blanking table is grouped according to the reference object marker and sorted according to the time slice marker. Each reference object marker corresponds to a time series record chain. Each record in the time series record chain retains the blanking state value and blanking change marker within that time slice; when the same fixed reference object is observed by multiple cameras, the fixed reference object time-sharing blanking table retains the records under each camera marker and merges them through the reference object marker; when no valid observation frame is obtained in a certain time slice, a missing detection marker is written to the corresponding record, and the recording order of the preceding and following time slices is retained.

[0037] In this embodiment S3, the target reference object is a reference object among the fixed reference objects that exhibits a blanking change within a continuous time slice and can identify the corresponding combustible component and directional migration edge. It is used to participate in the establishment of the component blanking continuation table and the generation of sequential correspondence. The specific selection rules for the target reference object are as follows: determine the fixed reference object with a continuous time slice blanking change record in the fixed reference object time-sharing blanking table, determine the corresponding combustible component based on the component reference attachment table, determine the corresponding directional migration edge based on the ignition migration three-dimensional network, and select the fixed reference object that simultaneously satisfies the continuous time slice blanking change, has an attachment relationship, and has a corresponding directional migration edge as the target reference object.

[0038] In this embodiment S3, when determining the target reference object based on the fixed reference object time-sharing blanking table, the time series record chain in the fixed reference object time-sharing blanking table is first read according to the reference object identifier, and the blanking change identifier and blanking state value of the same fixed reference object in continuous time slices are checked; when the fixed reference object has a blanking change identifier in no less than a preset number of continuous time slices, and the corresponding blanking state value forms a continuous change record according to the time slice order, the fixed reference object is recorded in the candidate reference object set; for each fixed reference object in the candidate reference object set, the component reference connection table is read again to check whether the fixed reference object exists. The system records the attachments to combustible components. Candidate references with attachment records retain their reference identifier, corresponding component identifier, attachment position, and attachment relationship. Candidate references without attachment records are not included in the target reference list. When reading the component reference attachment table, if a candidate reference has an attachment relationship with only one combustible component, then that combustible component is identified as the combustible component corresponding to the candidate reference. If a candidate reference has an attachment relationship with two or more combustible components, then the system reads the corresponding component identifiers for each component and retains the attachment position and attachment relationship to confirm whether the candidate reference corresponds to a directional migration edge.

[0039] In this embodiment S3, when determining the combustible component and directional migration edge corresponding to the target reference object, the component identifier corresponding to the candidate reference object is first obtained according to the component reference attachment table, and then the node record and edge record containing the component identifier are retrieved in the ignition migration three-dimensional network. For a candidate reference object that is only attached to a single combustible component, the outgoing edge set and incoming edge set of the combustible component in the ignition migration three-dimensional network are read, and the directional migration edge directly connected to the combustible component is retrieved. For a candidate reference object that is attached to the connected area between two combustible components, it is read whether there is a directional migration edge between the two combustible components. If there is a directional migration edge, the directional migration edge is determined. The directional migration edge is the corresponding directional migration edge for the candidate reference object. For candidate reference objects attached to the adjacent boundary areas of two or more combustible components, the directional migration edges between adjacent combustible components are retrieved one by one according to the attachment position and component migration relationship record. The directional migration edge that matches the attachment position is determined as the corresponding directional migration edge. When a candidate reference object simultaneously satisfies the requirements of continuous time slice blanking change, has attachment record, and can be determined as a corresponding directional migration edge, the candidate reference object is determined as the target reference object. The target reference object record includes the reference object identifier, camera identifier, corresponding component identifier, corresponding directional migration edge identifier, start time slice identifier, and blanking change record index.

[0040] In this embodiment S3, the component blanking continuation table is a continuation relationship table formed by mapping the blanking records of the target reference object to the corresponding combustible components and directional migration edges. It is used to characterize the blanking continuation relationship between the combustible components and directional migration edges corresponding to the target reference object in continuous time slices. The data structure of the component blanking continuation table is specifically a continuation relationship table structure, including component identifier, directional migration edge identifier, reference object identifier, start time slice identifier, end time slice identifier, blanking state value, and continuation order identifier. The specific rules and methods for establishing the component blanking continuation table are as follows: read the continuous time slice blanking records of the target reference object in the fixed reference object time-sharing blanking table, map each blanking record to the corresponding combustible component based on the component reference attachment table, map each blanking record to the corresponding directional migration edge based on the ignition migration three-dimensional network, sort the mapped records according to the order of time slices, and continuously merge the blanking records belonging to the same combustible component or the same directional migration edge in adjacent time slices to form the component blanking continuation table.

[0041] In this embodiment S3, when establishing the component blanking continuation table for the target reference object, the continuous time slice blanking records of the target reference object in the fixed reference object time-sharing blanking table are first read. Each blanking record is mapped to the corresponding combustible component according to the attachment relationship in the component reference attachment table. For target reference objects with attachment relationship of surface area attachment, their blanking records are written into the component blanking record item of the corresponding combustible component. For target reference objects with attachment relationship of boundary adjacent area attachment, their blanking records are written into the boundary blanking record item of the corresponding combustible component. For target reference objects with attachment relationship of connected area attachment, their blanking records are simultaneously written into the component blanking record items of combustible components on both sides of the connected area, and the directional migration edge between the combustible components on both sides in the ignition migration three-dimensional network is read. The hidden record is written to the edge hidden record item of the corresponding directional migration edge. The component hidden record continuation table is stored in a continuation relationship table structure. Its fields include component identifier, directional migration edge identifier, reference object identifier, start time slice identifier, end time slice identifier, hidden record status value, and continuation order identifier. Among them, the component identifier is used to record the combustible component corresponding to the target reference object, the directional migration edge identifier is used to record the directional migration edge corresponding to the target reference object, the reference object identifier is used to record the target reference object participating in the mapping, the start time slice identifier and end time slice identifier are used to record the time range to which the continuous hidden record belongs, the hidden record status value is used to record the hidden record status data within the time range, and the continuation order identifier is used to record the arrangement position of the record in the hidden record sequence corresponding to the same target reference object.

[0042] In this embodiment S3, the rules for establishing the component blanking continuation table are executed according to four processes: mapping, sorting, merging, and registration. The mapping process is used to write the blanking record of the target reference object in the fixed reference object time-sharing blanking table into the corresponding combustible component and the corresponding directional migration edge. The sorting process is used to arrange the blanking records of the same target reference object according to the order of the time slice identifiers. The merging process is used to merge consecutive blanking records belonging to the same combustible component or the same directional migration edge in adjacent time slices into the same continuation segment. The registration process is used to write the continuation segment into the component blanking continuation table. When performing the merging process, two adjacent blanking records are read first. The recorded component identifier, directional migration edge identifier, and time slice identifier are grouped into the same continuation segment when two adjacent records belong to the same component identifier or the same directional migration edge identifier and have consecutive time slice identifiers. The termination time slice identifier is updated in the continuation segment. When the component identifier or directional migration edge identifier corresponding to two adjacent records changes, the current continuation segment ends and the next continuation segment is established. When the target reference object corresponds to multiple combustible components, continuation segment records are formed according to each combustible component and each directional migration edge. The component blanking continuation table is stored in groups according to the target reference object identifier, and the records in each group are arranged according to the continuation order identifier.

[0043] In this embodiment S3, the directional migration order is the arrangement order of the combustible components and directional migration edges corresponding to the target reference object along the migration direction in the ignition migration three-dimensional network. The time slice continuation order is the arrangement order of the combustible components and directional migration edges corresponding to the target reference object in the component blanking continuation table according to the time slice sequence. The sequential correspondence between the directional migration order and the time slice continuation order is the sequential matching relationship of the same combustible components and the same directional migration edges in the two arrangement orders, which is used to characterize the consistency between the spatial migration process and the time continuation process. The specific rule method for establishing the sequential correspondence is as follows: based on the component reference hooking table, the combustible components and directional migration edges in the directional migration order are matched with the combustible components and directional migration edges in the time slice continuation order, and the sequential matching relationship between adjacent sequence records is generated according to the migration direction and the time slice sequence. The migration consistency value is a numerical value characterizing the degree of consistency between the directional migration sequence and the time slice succession sequence in the migration direction of combustible components, used to characterize the matching degree of spatial migration relationships; the blanking succession consistency value is a numerical value characterizing the degree of continuity between the time slice succession sequence and the directional migration sequence within consecutive time slices, used to characterize the matching degree of the blanking advancement process; the specific methods for calculating the migration consistency value and the blanking succession consistency value are as follows: count the number of combustible components and directional migration edge corresponding records with consistent sequential positions in the sequential correspondence relationship, and calculate the normalized ratio with the total number of corresponding records in the directional migration sequence to obtain the migration consistency value; count the number of combustible components and directional migration edge corresponding records that maintain continuity in consecutive time slices in the sequential correspondence relationship, and calculate the normalized ratio with the total number of corresponding records in the time slice succession sequence to obtain the blanking succession consistency value.

[0044] In this embodiment S3, when extracting the directional migration order of the target reference object from the ignition migration three-dimensional network, the combustible component identifier and directional migration edge identifier corresponding to the target reference object are first read. The corresponding node record and edge record are then retrieved from the node table, edge table, and node-edge association table of the ignition migration three-dimensional network. For a target reference object corresponding to a single directional migration edge, the directional migration order is directly determined according to the edge's start and end points. For a target reference object corresponding to multiple directional migration edges, the adjacency relationship in the directed graph is read according to the connection direction between the edge's start and end points, and the directional migration order is formed according to the connection order of the directional migration edges. When multiple combustible components are connected by continuous directional migration edges, starting from the initial combustible component corresponding to the target reference object, the edge set is read sequentially. The directional migration edge is then read, and the corresponding endpoint component is read. The component identifier and the directional migration edge identifier are arranged in the reading order. The directional migration sequence is saved in the form of a sequence record table. Each sequence record includes the sequence position identifier, component identifier, directional migration edge identifier, edge start component identifier, and edge end component identifier. When extracting the time slice continuation sequence from the component blanking continuation table, the component blanking continuation table record corresponding to the target reference object is read first. The record is sorted according to the order of continuation sequence identifier and time slice identifier. The sorted component identifier, directional migration edge identifier, and time slice range are written into the time slice continuation sequence record. The time slice continuation sequence is saved in the form of a continuation record table. Each continuation record includes the continuation position identifier, component identifier, directional migration edge identifier, start time slice identifier, end time slice identifier, and blanking status value.

[0045] In this embodiment S3, when establishing the sequential correspondence between the directional migration sequence and the time slice continuation sequence, the attachment record corresponding to the target reference object in the component reference attachment table is read first to obtain the component identifier and attachment position corresponding to the target reference object. Then, the sequence record in the directional migration sequence and the continuation record in the time slice continuation sequence are read. For each sequence record in the directional migration sequence, it is checked whether there is a continuation record with the same component identifier or the same directional migration edge identifier in the time slice continuation sequence. If there is the same component identifier or the same directional migration edge identifier, the sequence record and the continuation record are written into the sequential correspondence relationship. If the same sequence record corresponds to multiple continuation records, they are arranged according to the order of the time slice identifiers. The earliest sequential record is retained, and the remaining sequential records are written to the duplicate sequential record field. If the same sequential record corresponds to multiple sequential records, the sequential record that matches the target reference position is selected according to the attachment position and attachment relationship in the component reference attachment table. The sequential correspondence is stored in the form of a correspondence table. Each record includes a sequential position identifier, a sequential position identifier, a component identifier, a directional migration edge identifier, an attachment position, a sequential matching identifier, and a sequential matching identifier. The sequential matching identifier is used to record whether the position in the directional migration sequence corresponds to the position in the time slice sequential sequence. The sequential matching identifier is used to record whether the sequential records between adjacent time slices maintain the same arrangement relationship as the directional migration sequence.

[0046] In this embodiment S3, when calculating the migration consistency value, firstly, the number of records with the sequential matching identifier in the sequential correspondence is counted, then the total number of sequential records in the directed migration sequence is counted, and the normalized ratio of the number of records with matching status to the total number of sequential records is calculated to obtain the migration consistency value; when the directed migration sequence includes combustible component records and directed migration edge records, both combustible component records and directed migration edge records are included in the total number of sequential records; when the directed migration sequence corresponding to a certain target reference object only includes a single combustible component and a single directed migration edge, the migration consistency value is calculated according to the matching status of the single sequential record and the corresponding successor record. Consistency value; When calculating the blanking continuity consistency value, first count the number of records in the sequential correspondence where the continuity matching identifier is in a matched state and the time slices are consecutive, then count the total number of adjacent continuity records in the time slice continuity sequence, and calculate the normalized ratio of the number of continuity matching records to the total number of adjacent continuity records to obtain the blanking continuity consistency value; when there is a missing test identifier in the time slice continuity sequence, the corresponding missing test record is not included in the total number of adjacent continuity records; when there are multiple time slice continuity sequences corresponding to the same target reference object, calculate the migration consistency value and blanking continuity consistency value corresponding to each camera separately, and then group and record them according to the camera identifier.

[0047] In this embodiment S4, the blanking record corresponding to the target reference object is jointly provided by the fixed reference object time-sharing blanking table and the component blanking continuation table. The blanking record corresponding to the target reference object includes the target reference object identifier, camera identifier, time slice identifier, blanking status value, blanking change identifier, corresponding combustible component identifier, corresponding directional migration edge identifier, and continuation sequence identifier. When reading the blanking record corresponding to the target reference object, firstly, the blanking status value and blanking change identifier within the continuous time slice are extracted from the fixed reference object time-sharing blanking table according to the target reference object identifier. Then, the corresponding combustible component identifier, corresponding directional migration edge identifier, start time slice identifier, and end time slice identifier are extracted from the component blanking continuation table according to the same target reference object identifier. And the sequence identifier; when the same target reference object is observed by multiple cameras, the blanking record corresponding to each camera is read according to the camera identifier, and the time slice range corresponding to each blanking record is retained; when the same target reference object is attached to multiple combustible components at the same time, the corresponding combustible component identifier and directional migration edge identifier are read according to the attachment relationship in the component reference attachment table; when there is a missing measurement identifier in the blanking record of the target reference object, the time slice position corresponding to the missing measurement identifier is read, and the data source status corresponding to the missing measurement identifier is retained when calculating the target fire judgment result; the blanking records corresponding to the target reference object are arranged in ascending order according to the time slice identifier to form a time sequence record set participating in the generation of the target fire judgment result.

[0048] In this embodiment S4, when generating the target fire judgment result, the migration consistency value and the blanking continuity consistency value are read first, and then the blanking record corresponding to the target reference object is read. The migration consistency value is used to record the degree of matching between the directional migration sequence and the time slice continuity sequence in the direction of combustible component migration. The blanking continuity consistency value is used to record the degree of continuity matching between the time slice continuity sequence and the directional migration sequence in continuous time slices. The blanking record corresponding to the target reference object is used to record the blanking state change of the target reference object in continuous time slices. When processing the blanking record corresponding to the target reference object, the blanking state value in continuous time slices is read according to the time slice identifier. The start time slice identifier and end time slice identifier of the continuous blanking change are written into the judgment record, and the blanking state value within the time range is written into the judgment record. When generating the target fire situation assessment result, the migration consistency value, the blanking continuity consistency value, and the blanking record corresponding to the target reference object are jointly assessed. When the migration consistency value meets the preset migration assessment condition, the blanking continuity consistency value meets the preset continuity assessment condition, and there is a continuous time slice blanking change in the blanking record corresponding to the target reference object, the fire situation assessment status in the target fire situation assessment result is written into the fire situation status identifier. When the migration consistency value, the blanking continuity consistency value, or the blanking record corresponding to the target reference object does not meet the corresponding assessment condition, the fire situation assessment status in the target fire situation assessment result is written into the non-fire situation status identifier or the review status identifier. The target fire situation assessment result is registered according to the target monitoring scene identifier, the target reference object identifier, the corresponding combustible component identifier, and the corresponding directional migration edge identifier.

[0049] In this embodiment S4, the target fire situation judgment result is stored in a structured record format. The target fire situation judgment result includes the target monitoring scene identifier, camera identifier, target reference object identifier, corresponding combustible component identifier, corresponding directional migration edge identifier, start time slice identifier, end time slice identifier, migration consistency value, blanking continuity consistency value, blanking record index, and fire situation judgment status. The target monitoring scene identifier is used to record the monitoring area to which the current judgment result belongs; the camera identifier is used to record the video source corresponding to the blanking record of the target reference object; the target reference object identifier is used to record the fixed reference object participating in the current judgment; the corresponding combustible component identifier is used to record the combustible component attached to the target reference object; and the corresponding directional migration edge identifier is used to record the combustible component associated with the target reference object. The migration direction, start time slice identifier, and end time slice identifier are used to record the time range corresponding to the continuous blanking changes of the target reference object. The migration consistency value and blanking continuity consistency value are used to record the sequential calculation results. The blanking record index is used to associate the corresponding records in the fixed reference object time-sharing blanking table and the component blanking continuity table. The fire situation judgment status is used to record the status type of the current target fire situation judgment result. When multiple target reference objects correspond to the same combustible component or the same directional migration edge, corresponding target fire situation judgment result records are generated respectively, and grouped and registered according to the combustible component identifier and the directional migration edge identifier. When multiple cameras correspond to target reference objects that belong to the same combustible component, their respective target fire situation judgment result records are retained according to the camera identifier.

[0050] In this embodiment S4, when the target fire situation judgment result is fed back to the intelligent fire monitoring platform, the target fire situation judgment result is first written into the fire situation judgment record database of the intelligent fire monitoring platform, and then the target fire situation judgment result is associated with the target monitoring scene, camera, combustible component, fixed reference object, and directional migration edge. The data written into the fire situation judgment record database includes the target monitoring scene identifier, camera identifier, target reference object identifier, corresponding combustible component identifier, corresponding directional migration edge identifier, start time slice identifier, end time slice identifier, migration consistency value, blanking continuity consistency value, and fire situation judgment status. During the writing process, the target monitoring scene record is determined according to the target monitoring scene identifier, and according to... The camera identification identifies the corresponding video source. The system associates the hidden record in the time-sharing hidden record table of the fixed reference object with the target reference object identification. It also associates the node and edge records in the ignition migration three-dimensional network with the corresponding combustible component identification and the corresponding directional migration edge identification. When the fire status is identified as a fire status, the smart fire monitoring platform generates a fire status record in the corresponding target monitoring scene. When the fire status is identified as a review status, the smart fire monitoring platform generates a review record in the corresponding target monitoring scene. When the fire status is identified as a non-fire status, the smart fire monitoring platform generates a non-fire record in the corresponding target monitoring scene. Each record retains the hidden record index corresponding to the target reference object.

[0051] Example 2: The intelligent video stream analysis system in the intelligent fire monitoring platform proposed in this invention is applied to the intelligent video stream analysis method in the intelligent fire monitoring platform proposed in Example 1. It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the intelligent video stream analysis method in the intelligent fire monitoring platform in Example 1.

[0052] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A video stream intelligent analysis method in a smart fire monitoring platform, characterized in that, Includes the following steps: S1. Acquire multiple monitoring video streams of the same target monitoring scene, identify flammable components and fixed reference objects in the multiple monitoring video streams, assign a unique number to the same flammable component in different cameras, perform attachment labeling processing on the fixed reference object and the corresponding flammable component, and construct a component reference attachment table. Among them, the fixed reference object is a reference object in the target monitoring scene whose position remains fixed, which can be continuously identified in the video stream and used to characterize local visibility changes; the component reference hooking table is a hooking relationship data table that shows the correspondence between the fixed reference object and the combustible component. S2. Under a pre-defined unified spatial coordinate system, determine the component migration relationship between each combustible component based on all combustible components, generate directional migration edges between combustible components with component migration relationships, and establish an ignition migration three-dimensional network. Among them, the directional migration edge is a directed connection edge that represents the migration direction relationship between two combustible components; the ignition migration three-dimensional network is composed of combustible components and directional migration edges, and is used to represent the network structure of ignition migration relationship in the target monitoring scene; S3. Divide the multi-channel monitoring video stream into continuous time slices, extract the blanking state data of the fixed reference object in each time slice, establish a time-sharing blanking table for the fixed reference object, determine the combustible component and directional migration edge corresponding to the target reference object based on the time-sharing blanking table for the fixed reference object and the component reference connection table, and establish a component blanking continuity table for the target reference object; extract the directional migration sequence of the target reference object from the ignition migration three-dimensional network, extract the time slice continuity sequence from the component blanking continuity table, and establish the sequence correspondence between the directional migration sequence and the time slice continuity sequence based on the component reference connection table, and calculate the migration consistency value and blanking continuity consistency value; Among them, the fixed reference object time-sharing blanking table is a time-series data table that represents the blanking change state of the fixed reference object within a continuous time slice; the component blanking continuity table is a data table that represents the blanking continuity relationship of the combustible component corresponding to the target reference object within a continuous time slice; the target reference object is the reference object among the fixed reference objects that participates in the current fire situation judgment; S4. Based on the migration consistency value, the blanking continuity consistency value and the blanking record corresponding to the target reference object, perform fire judgment on the target monitoring scene, obtain the target fire judgment result, and feed the target fire judgment result back to the smart fire monitoring platform in real time. Among them, the blanking record corresponding to the target reference object is the continuous time slice blanking change record corresponding to the target reference object in the fixed reference object time-sharing blanking table, which is used to represent the time sequence record data of the blanking change process of the target reference object in the continuous time slice.

2. The intelligent video stream analysis method in the intelligent fire monitoring platform according to claim 1, characterized in that: In S1, the combustible component is a fixed entity in the target monitoring scene that can form an ignition and transmission path when a fire occurs; the fixed reference object is used to characterize the local visibility changes of the adjacent area of ​​the corresponding combustible component; the selection rules for the fixed reference object are as follows: it maintains its position in multiple monitoring video streams, has continuous and identifiable boundary features, is located in the surface area of ​​the combustible component, the boundary adjacent area, or the connected area between components, has a comparable visibility state in continuous time slices, and can establish a stable correspondence with the same combustible component in the corresponding images of different cameras.

3. The intelligent video stream analysis method in the intelligent fire monitoring platform according to claim 2, characterized in that: In S1, the attachment labeling process is an association labeling method based on spatial adjacency and boundary correspondence, which is used to associate and mark the position of fixed reference objects with corresponding combustible components in order to construct a component reference attachment table. The component reference attachment table is used to record the correspondence between fixed reference objects and combustible components. Its data structure includes reference object identifier, component identifier, attachment position, and attachment relationship. The component reference attachment table is also used to characterize the boundary correspondence and connected region association between fixed reference objects and combustible components. The attachment relationship rules are as follows: When a fixed reference object is located on the surface area or boundary adjacent area of ​​a single combustible component, the fixed reference object is attached to that combustible component; when a fixed reference object is located in the adjacent boundary areas of two or more combustible components, the fixed reference object is attached to the corresponding combustible components respectively; when a fixed reference object is located in the connected region between two combustible components, the fixed reference object is attached to the combustible components on both sides of the connected region and the connected region position is marked in the attachment position.

4. The intelligent video stream analysis method in the intelligent fire monitoring platform according to claim 3, characterized in that: In S2, the preset unified spatial coordinate system is a coordinate reference system that uniformly represents the position of flammable components and fixed reference objects in multiple monitoring video streams, and is used to establish a unified spatial position correspondence for the same flammable component in different cameras; The component migration relationship is the directed association between various combustible components on the ignition transmission path, used to characterize the adjacent transmission order between combustible components. The specific rules for determining the component migration relationship are as follows: Under a preset unified spatial coordinate system, first determine whether there is a spatial adjacency relationship or a connected region association relationship between two combustible components, then determine whether there is a boundary correspondence relationship or an obstruction discontinuity relationship between two combustible components, and finally determine the migration direction based on the position of the connected region between the two combustible components. When the aforementioned conditions are met, it is determined that there is a component migration relationship between the two combustible components. Among them, the obstruction discontinuity relationship is a relationship in which there is obstruction between two combustible components but a migration direction can still be determined.

5. The intelligent video stream analysis method in the intelligent fire monitoring platform according to claim 4, characterized in that: In S2, the method for generating directional migration edges is as follows: the starting combustible component with a component migration relationship is determined as the edge starting point, the ending combustible component with a migration direction corresponding to the starting combustible component is determined as the edge ending point, and a one-way connection is established between the edge starting point and the edge ending point to generate a directional migration edge. The ignition migration three-dimensional network is a directed network structure composed of combustible components as nodes and directional migration edges as edges. The ignition migration relationship refers to the component migration relationship from the starting combustible component to the ending combustible component along the migration direction. The specific structure of the ignition migration three-dimensional network is a directed graph storage structure that uses the unique number of the combustible component as the node index, the unique number of the directional migration edge as the edge index, and is organized according to the association order of nodes and edges. The specific method for establishing the ignition migration three-dimensional network is as follows: node records are generated based on the unique number of each combustible component, directional migration edge records are generated based on the migration relationship of each component, each directional migration edge is associated with its corresponding starting point and ending point, and the node records corresponding to the same combustible component in different cameras are uniformly mapped to obtain the ignition migration three-dimensional network.

6. The intelligent video stream analysis method in the intelligent fire monitoring platform according to claim 5, characterized in that: In S3, the blanking state data is time-series data used to characterize the degree of change in the visible state of a fixed reference object within a corresponding time slice, and is used to record the blanking change process of the fixed reference object within consecutive time slices. The fixed reference object time-sharing blanking table is a time-series record table established based on the association between the fixed reference object and the time slice, used to record the blanking change records of each fixed reference object within consecutive time slices and to support the selection of target reference objects. The data structure of the fixed reference object time-sharing blanking table is specifically a time-series table structure, including camera identifier, reference object identifier, time slice identifier, blanking state value, and blanking change identifier. The specific method for establishing the fixed reference object time-sharing blanking table is as follows: the multiple monitoring video streams are divided according to consecutive time slices, the blanking state data of the corresponding time slice is extracted for each fixed reference object in each camera, and the data is collected according to the camera identifier, reference object identifier, and time slice identifier to generate the fixed reference object time-sharing blanking table.

7. The intelligent video stream analysis method in the intelligent fire monitoring platform according to claim 6, characterized in that: In S3, the target reference is a reference object among the fixed references that exhibits a blanking change within a continuous time slice and can identify the corresponding combustible component and directional migration edge. It is used to participate in the establishment of the component blanking continuation table and the generation of sequential correspondence. The specific selection rules for the target reference are as follows: determine the fixed reference with a continuous time slice blanking change record in the fixed reference time-sharing blanking table, determine the corresponding combustible component based on the component reference attachment table, determine the corresponding directional migration edge based on the ignition migration three-dimensional network, and select the fixed reference that simultaneously satisfies the continuous time slice blanking change, has an attachment relationship, and has a corresponding directional migration edge as the target reference.

8. The intelligent video stream analysis method in the intelligent fire monitoring platform according to claim 7, characterized in that: In S3, the component blanking continuation table is a continuation relationship table formed by mapping the blanking records of the target reference object to the corresponding combustible components and directional migration edges. It is used to characterize the blanking continuation relationship of the combustible components and directional migration edges corresponding to the target reference object in continuous time slices. The data structure of the component blanking continuation table is specifically a continuation relationship table structure, including component identifier, directional migration edge identifier, reference object identifier, start time slice identifier, end time slice identifier, blanking state value, and continuation order identifier. The specific rules and methods for establishing the component blanking continuation table are as follows: read the continuous time slice blanking records of the target reference object in the fixed reference object time-sharing blanking table, map each blanking record to the corresponding combustible component based on the component reference attachment table, map each blanking record to the corresponding directional migration edge based on the ignition migration three-dimensional network, sort the mapped records according to the order of time slices, and continuously merge the blanking records belonging to the same combustible component or the same directional migration edge in adjacent time slices to form the component blanking continuation table.

9. The intelligent video stream analysis method in the intelligent fire monitoring platform according to claim 8, characterized in that: In S3, the directional migration order is the arrangement order of the combustible components and directional migration edges corresponding to the target reference object along the migration direction in the ignition migration three-dimensional network, and the time slice continuation order is the arrangement order of the combustible components and directional migration edges corresponding to the target reference object in the component blanking continuation table according to the time slice sequence; the sequential correspondence between the directional migration order and the time slice continuation order is the sequential matching relationship of the same combustible components and the same directional migration edges in the two arrangement orders, which is used to characterize the consistency between the spatial migration process and the time continuation process; the specific rule method for establishing the sequential correspondence is: based on the component reference hooking table, the combustible components and directional migration edges in the directional migration order are matched with the combustible components and directional migration edges in the time slice continuation order, and the sequential matching relationship between adjacent sequential records is generated according to the migration direction and the time slice sequence relationship. The migration consistency value is a numerical value that characterizes the degree of consistency between the directional migration sequence and the time slice succession sequence in the migration direction of combustible components, and is used to characterize the matching degree of spatial migration relationship; the blanking succession consistency value is a numerical value that characterizes the degree of continuity between the time slice succession sequence and the directional migration sequence within consecutive time slices, and is used to characterize the matching degree of the blanking advancement process; the specific method for calculating the migration consistency value and the blanking succession consistency value is as follows: count the number of combustible components and corresponding records of directional migration edges with consistent sequential positions in the sequential correspondence relationship, and calculate the normalized ratio with the total number of corresponding records in the directional migration sequence to obtain the migration consistency value; The number of records corresponding to combustible components and directional migration edges that are maintained in the continuous sequence of statistical order is calculated, and the normalized ratio is calculated with the total number of corresponding records in the sequence of time slices to obtain the blanking continuity consistency value.

10. A video stream intelligent analysis system in a smart fire monitoring platform, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes a computer program to implement the intelligent video stream analysis method in the intelligent fire monitoring platform as described in any one of claims 1-9.

Images