Environment safety intelligent monitoring and early warning analysis method

Abstract

The invention discloses an environment safety intelligent monitoring and early warning analysis method, which relates to the field of intelligent monitoring and early warning analysis, and comprises the following steps: constructing an environment monitoring data pool; the AI video analysis model analyzes the video data in real time to generate a video analysis event; when it is monitored that a sensor generates an alarm signal or a video analysis event meeting an early warning condition is recognized, an adjacent monitoring device is automatically dispatched to focus on an event occurrence area, video data are collected and rechecked, a side AI model extracts multi-dimensional historical and real-time monitoring data associated with the event from a data pool, and the multi-dimensional historical and real-time monitoring data are stored in a database; carrying out fusion analysis on the data and the rechecked video data, and outputting an event verification result with a confidence score; and according to the confidence score, querying the preset hierarchical early warning-disposal mapping table, and dynamically executing corresponding early warning information release and field equipment linkage control, thereby realizing a closed loop from passive warning to active intelligent research, judgment and disposal, and effectively improving the real-time performance, accuracy and automation level of environmental safety monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent monitoring and early warning analysis, and in particular to an environment safety intelligent monitoring and early warning analysis method. BACKGROUND

[0002] With the rapid development of smart cities, smart power grids and industrial internet, higher requirements are put forward for the environment safety monitoring of key infrastructures such as power distribution rooms, charging stations and data centers. The traditional environment safety monitoring system mainly relies on the simple combination of manual inspection and single-point alarm devices, which has been difficult to meet the real-time, accuracy and intelligent needs.

[0003] Currently, the mainstream environment safety monitoring scheme has significant limitations in system, intelligence and real-time. First, the video, sensor and access control systems are independent of each other, forming a "data island", which leads to the inability to integrate information and the easy occurrence of "alarm storm". Second, the monitoring mainly relies on passive video recording and simple sensor alarm, lacking active and intelligent identification capabilities for complex scenarios such as fire and intrusion. Finally, alarm verification relies on manual video review, with slow response and high false alarm rate, and the disposal method is rigid and cannot perform hierarchical linkage according to the emergency level of the event.

[0004] Therefore, an environment safety intelligent monitoring and early warning analysis method is provided to solve the above problems. SUMMARY

[0005] In order to solve the above problems, the present application provides an environment safety intelligent monitoring and early warning analysis method, which builds a unified multi-source heterogeneous environment monitoring data pool and deploys edge-side agents to automatically schedule the optimal monitoring perspective when an event is triggered, and intelligently analyzes the real-time video review data and multi-dimensional historical and real-time monitoring data to generate a quantitative confidence, and finally executes hierarchical early warning and precise linkage control based on a dynamic strategy library, thereby realizing a closed loop from passive alarm to active intelligent research and disposal, and effectively improving the real-time, accuracy and automation level of environment safety monitoring.

[0006] To achieve the above purpose, the present application provides an environment safety intelligent monitoring and early warning analysis method, comprising the following steps: S1: collecting environment monitoring data to build an environment monitoring data pool; the environment monitoring data includes data of video monitoring devices, environmental sensors and intelligent terminals in the target monitoring area; S2: using an AI video analysis model to perform real-time analysis on the video data in the environment monitoring data pool, identifying preset safety events and generating video analysis events; S3: automatically triggering a device linkage rule when an alarm signal is monitored from a sensor in the data pool or a video analysis event meeting a pre-warning condition is identified; the linkage rule comprising: automatically dispatching adjacent monitoring devices to focus on the event occurrence area, collecting review video data, and an edge AI model extracting multi-dimensional historical and real-time monitoring data associated with the event from the data pool, fusing and analyzing the review video data, and outputting an event verification result with a confidence score; S4: querying a preset hierarchical pre-warning-disposal mapping table according to the confidence score, and dynamically executing corresponding pre-warning information publishing and on-site device linkage control.

[0007] Preferably, in step S1: The data of the video monitoring device includes real-time video data of the target monitoring area; The data of the environmental sensor includes smoke concentration value of a smoke detector, liquid level of a water immersion sensor, numerical value of a temperature and humidity sensor, and device surface temperature value of an infrared thermometer; The data of the intelligent terminal includes switch state of an access controller, opening and closing feedback of a fire valve, and online state data of a sound and light alarm.

[0008] Preferably, the preset security events in S2 include personnel intrusion, fire identification, personnel fall, device state anomaly, and liquid leakage.

[0009] Preferably, the AI video analysis model in S2 specifically includes: Based on a deep learning model, real-time analysis is performed on video data to identify predefined security events, and a structured analysis result containing event type, occurrence location, and confidence is output.

[0010] Preferably, the automatic dispatching of adjacent monitoring devices in S3 specifically includes: According to the physical location coordinates of the alarm event, a pre-stored spatial topology relationship library is queried to determine a candidate device set from the deployed monitoring device set; Based on the relative spatial geometric relationship between each device in the candidate device set and the alarm event, a scheduling priority score of each device is calculated; According to the scheduling priority score, the optimal monitoring device is selected and a standardized control instruction is generated to drive it to turn and focus.

[0011] Preferably, the scheduling priority score is determined based on the included angle between the visual axis of the monitoring device and the direction of the alarm point, the Euclidean distance between the monitoring device and the alarm point, and the physical occlusion state between the monitoring device and the alarm point.

[0012] Preferably, the scheduling priority score is calculated by a weighted scoring function, specifically represented as: ; in, The angle between the line of sight of the monitoring equipment and the direction of the alarm point. The distance between the monitoring equipment and the alarm point is expressed in Euclidean form. This refers to the physical obstruction between the monitoring equipment and the alarm point. If there is no obstruction, then... =0; if obscured, then =1; , , This is the normalization function; , , These are the weighting coefficients, and .

[0013] Preferably, the fusion analysis in S3 specifically includes: The video data to be reviewed is input into the side AI model to perform event review and obtain the video review confidence level. Extract N types of multi-dimensional monitoring data related to events from the data pool, and obtain the result through the corresponding data evaluation model. Support score for class data; The overall confidence score is obtained based on the video review confidence and support scores.

[0014] Preferably, the overall confidence score is expressed as follows: ; in, The weights for the confidence level of video re-judgment. For the first The weighting of the support score for class monitoring data. To determine the confidence level for video re-evaluation, For the first Support score for class data.

[0015] Preferably, the graded early warning-handling mapping table in S4 is a dynamically configurable association table, which uses event type and overall confidence score range as joint query conditions to map to the corresponding early warning level and handling instruction set.

[0016] Therefore, the present invention employs the above-mentioned intelligent monitoring and early warning analysis method for environmental safety, which has the following beneficial effects: (1) By constructing a unified spatiotemporal benchmark environmental monitoring data pool, this invention breaks down the data barriers between video, sensors and terminal devices, providing a unified data foundation for multi-source information correlation analysis and collaborative decision-making, and fundamentally solving the problems of system fragmentation and alarm storms.

[0017] (2) The application automatically schedules the optimal visual angle monitoring device for review through the event-driven intelligent linkage rule, and uses the side intelligent agent to fuse and analyze the review video and multi-dimensional historical / real-time data, upgrades the traditional review process relying on manual and lag to second-level and automatic intelligent research and judgment, and greatly reduces the invalid emergency response caused by sensor false alarm.

[0018] (3) The application innovatively outputs the event verification result with confidence score, and queries the dynamically configurable hierarchical warning-disposal mapping table based on the result, so that the system can perform differentiated warning and linkage control according to the certainty of the event. This realizes the transformation of the response strategy from one-size-fits-all to precision, and optimizes the allocation of emergency resources.

[0019] The technical solutions of the application will be further described in detail below with the help of the accompanying drawings and embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 It is a flowchart of an environmental safety intelligent monitoring and early warning analysis method in the application. DETAILED DESCRIPTION

[0021] The following detailed description of the embodiments of the application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but only represents selected embodiments of the application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the application without creative labor belong to the scope of protection of the application.

[0022] Unless otherwise defined, the technical terms or scientific terms used in the application shall be understood as the usual meaning understood by those skilled in the art in the field to which the application belongs.

[0023] The terms "comprise", "contain", "include" and similar words in the present invention mean that the elements before the word are encompassed in the elements listed after the word, and do not exclude the possibility of also encompassing other elements. The terms "in", "out", "up", "down" and the like indicate the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present invention. When the absolute position of the described object changes, the relative positional relationship may also change accordingly. In the present invention, unless otherwise explicitly specified and limited, the term "attached" and other terms should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or it can be integrated; it can be directly connected, or it can be indirectly connected through an intermediate medium; it can be the internal connection of two elements or the interaction relationship between two elements. For those skilled in the art, the specific meaning of the above terms in the present invention can be understood according to the specific circumstances.

[0024] Embodiments An environmental safety intelligent monitoring and early warning analysis method, as shown in Figure 1 The method comprises the following steps: S1: Collecting environmental monitoring data to build an environmental monitoring data pool; the environmental monitoring data includes data of video monitoring equipment, environmental sensors and intelligent terminals in a target monitoring area; The data of the video monitoring equipment includes real-time video data of the target monitoring area; The data of the environmental sensors includes smoke concentration values of smoke detectors, liquid level heights of water immersion sensors, numerical values of temperature and humidity sensors, and device surface temperature values of infrared thermometers; The data of the intelligent terminals includes switch states of access control controllers, opening and closing feedbacks of fire gate valves, and online state data of sound and light alarms.

[0025] S2: Using an AI video analysis model to perform real-time analysis on the video data in the environmental monitoring data pool, identifying preset safety events and generating video analysis events; The preset safety events include personnel intrusion, smoke and fire identification, personnel falling, device state anomaly and liquid leakage.

[0026] The AI video analysis model specifically includes: Based on a deep learning model, the video data is analyzed in real time to identify predefined safety events, and a structured analysis result containing event type, occurrence location and confidence is output.

[0027] Specifically, the deep learning model is a hybrid architecture combining convolutional neural networks and time series models. The convolutional neural network is responsible for extracting spatial features (such as texture, shape, color) from single-frame images, while the time series model is used to capture dynamic features between consecutive frames (such as smoke diffusion, personnel movement trajectory). Standardize the preprocessing of video data, including scaling the size to the model input size, normalizing the pixel value, and inputting the preprocessed image frames into the deployed deep learning model. The model performs forward propagation operations and outputs raw results containing the following information: class probability distribution: the model's confidence in each candidate region belonging to each predefined event category. Boundary box coordinates: the location of the predicted event in the image. Non-maximum suppression: remove redundant and overlapping detection boxes for the same event, and keep the one with the highest confidence. Threshold filtering: set a confidence threshold, and only output event detection results with a confidence higher than the threshold to filter out noise and uncertain predictions. The output results include structured analysis results of event type, occurrence location, and confidence. Event type: directly derived from the model output, filtered by the threshold; image coordinates: derived from the model output, after post-processing of the boundary box coordinates. The coordinates are in pixels, defining the specific area of the event in the video frame; confidence: directly derived from the model's output of the class probability value, a number between 0 and 1, intuitively reflecting the degree of confidence in the model's judgment this time.

[0028] S3: When an alarm signal is detected from a sensor in the data pool or a video analysis event that meets the early warning condition is identified, automatically trigger the device linkage rule; the linkage rule includes: automatically dispatching adjacent monitoring devices to focus on the event occurrence area, collecting review video data, and the side AI model extracting multi-dimensional historical and real-time monitoring data associated with the event from the data pool, and performing fusion analysis with the review video data to output event verification results with confidence scores; The automatic dispatching of adjacent monitoring devices specifically includes: According to the physical location coordinates of the alarm event, query the pre-stored spatial topology relationship library to determine the candidate device set from the deployed monitoring device set; Based on the relative spatial geometric relationship between each device in the candidate device set and the alarm event, calculate the scheduling priority score of each device; The scheduling priority score is determined based on the angle between the monitoring device's visual axis and the alarm point direction, the Euclidean distance between the monitoring device and the alarm point, and the physical occlusion state between the monitoring device and the alarm point.

[0029] The scheduling priority score is calculated by a weighted scoring function, specifically represented as: ; where, is the angle between the monitoring device's visual axis and the alarm point direction, to monitor the Euclidean distance between the device and the alarm point, to monitor the physical occlusion state between the device and the alarm point, if there is no occlusion, = 0; if occluded, = 1; , , is a normalization function; , , is a weight coefficient, and . Specifically, may be , with a value range of [0, 1]; , is a preset reference distance constant, which is determined according to the typical effective recognition range of the camera in the monitoring scene. For example, for a power distribution room scene, after testing, most high-definition cameras have a maximum effective recognition distance of about 25 meters when the target occupies more than 15% of the screen. Therefore, = 25 (meters).

[0030] According to the dispatch priority score, the optimal monitoring device is selected and a standardized control instruction is generated to drive it to turn and focus.

[0031] The fusion analysis specifically includes: input the review video data into the edge-side AI model for event review, to obtain a video review confidence; Specifically, the review video stream collected by the dispatch camera and focused on the alarm area is input into a special lightweight edge-side AI model for event review. The model outputs a probability value between 0 and 1, which is the video review confidence, directly reflecting the degree of confidence that the AI model believes the alarm event to be true based on the current visual evidence.

[0032] extract N types of multi-dimensional monitoring data associated with the event from the data pool, and obtain the support score of the first type of data through the corresponding data evaluation model; From the environmental monitoring data pool, extract N types of historical and real-time monitoring data that are associated in time and adjacent in space with the current alarm event. For each type of data , a preset data evaluation model is used for processing to obtain a support score.

[0033] The data evaluation model can be a rule-based logic judge or a simple regression or classification model. Its core function is to evaluate the support or negation degree of this type of data for the current alarm event.

[0034] Specifically: For temperature data, the evaluation model can calculate the temperature rise rate in the last 30 seconds near the alarm point. If the rate exceeds the threshold, a higher support score is output.

[0035] For other smoke sensing state data, the evaluation model can check whether other smoke sensors in the same fire compartment are also alarming at the same time. If most of them are alarming, a higher support score is output.

[0036] For historical video analysis events, the evaluation model can query whether there have been related abnormal events (such as flashes) in the same area within a certain period of time before the alarm, and give a decaying support score according to the time distance.

[0037] Based on the video re-determination confidence and the support score, the overall confidence score is obtained.

[0038] A weighted fusion algorithm is used to integrate the video re-determination confidence and the support scores of all multi-dimensional data to calculate the final overall confidence score. This score integrates visual evidence and other environmental evidence, and is a more comprehensive and robust assessment of the authenticity of the event.

[0039] Specifically, the overall confidence score is represented as: ; Wherein, is the weight of the video re-determination confidence, is the weight of the support score of the th monitoring data, is the video re-determination confidence, is the support score of the th data.

[0040] S4: According to the confidence score, a pre-set hierarchical warning-disposal mapping table is queried to dynamically execute corresponding warning information release and on-site device linkage control.

[0041] The hierarchical warning-disposal mapping table is a dynamically configurable association table, which takes the event type and the overall confidence score interval as the joint query condition, and maps to the corresponding warning level and disposal instruction set.

[0042] The hierarchical warning-disposal mapping table is the core strategy library for intelligent decision-making and response. Specifically, as shown in Table 1 below, the hierarchical warning-disposal mapping table for fire conditions is given. During system operation, the event verification result (such as {event type: "fire", overall confidence: 0.88}) is taken as input to query this table. For example, for a fire confidence of 0.88, it will match to "first alarm", and then the system will automatically execute all the disposal instructions corresponding to this level in turn, realizing second-level emergency response.

[0043] Table 1 Hierarchical warning-disposal mapping table

[0044] Therefore, the application adopts the above-mentioned environment safety intelligent monitoring and early warning analysis method, breaks the information island by constructing a unified multi-source data pool, realizes the event-driven automation linkage and review by using the edge side intelligent agent, and innovatively introduces the hierarchical decision mechanism based on the quantitative confidence, thereby realizing the systematic breakthrough in the three aspects of data fusion, intelligent research and accurate response, finally constructing a closed-loop intelligent monitoring system with high real-time, high reliability and low false alarm, effectively solving the industry problems of response lag, high false alarm rate and linkage rigidity of the traditional scheme.

[0045] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, but not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can still be modified or replaced by equivalents, and these modifications or equivalent replacements should not make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present application.

Claims

1. An environmental safety intelligent monitoring and early warning analysis method, characterized in that, The method comprises the following steps: S1: collecting environmental monitoring data to build an environmental monitoring data pool; the environmental monitoring data includes data of video monitoring devices, environmental sensors and intelligent terminals in a target monitoring area; S2: using an AI video analysis model to perform real-time analysis on video data in the environmental monitoring data pool, identify preset safety events and generate video analysis events; S3: when an alarm signal is detected from a sensor in the data pool or a video analysis event meeting a pre-warning condition is identified, automatically triggering a device linkage rule; the linkage rule includes: automatically dispatching adjacent monitoring devices to focus on the event occurrence area, collecting review video data, and using a side AI model to extract multi-dimensional historical and real-time monitoring data associated with the event from the data pool, and perform fusion analysis on the review video data to output an event verification result with a confidence score; S4: according to the confidence score, querying a preset hierarchical pre-warning-disposal mapping table to dynamically execute corresponding pre-warning information release and on-site device linkage control.

2. The environmental safety intelligent monitoring and early warning analysis method according to claim 1, characterized in that, In step S1: The data of the video monitoring device includes real-time video data of the target monitoring area; The data of the environmental sensor includes smoke concentration value of a smoke detector, liquid level of a water immersion sensor, value of a temperature and humidity sensor, and device surface temperature value of an infrared thermometer; The data of the intelligent terminal includes on-off state of an access controller, opening and closing feedback of a fire valve, and online state data of a sound and light alarm.

3. The environmental safety intelligent monitoring and early warning analysis method according to claim 2, characterized in that: The preset safety events in S2 include personnel intrusion, fire identification, personnel fall, device state anomaly and liquid leakage.

4. The environmental safety intelligent monitoring and early warning analysis method according to claim 3, characterized in that, The AI video analysis model in S2 specifically includes: Based on a deep learning model, real-time analysis is performed on video data to identify predefined safety events, and a structured analysis result containing event type, occurrence location and confidence is output.

5. The environmental safety intelligent monitoring and early warning analysis method according to claim 4, characterized in that, The automatic dispatching of adjacent monitoring devices in S3 specifically includes: According to the physical location coordinates of the alarm event, querying a pre-stored spatial topology relationship database to determine a candidate device set from the deployed monitoring device set; Based on the relative spatial geometric relationship between each device in the candidate device set and the alarm event, a scheduling priority score of each device is calculated; According to the scheduling priority score, the optimal monitoring device is selected and a standardized control instruction is generated to drive it to focus.

6. The environmental safety intelligent monitoring and early warning analysis method according to claim 5, characterized in that, The scheduling priority score is determined based on the angle between the monitoring device's visual axis and the alarm point direction, the Euclidean distance between the monitoring device and the alarm point, and the physical blocking state between the monitoring device and the alarm point.

7. The environmental safety intelligent monitoring and early warning analysis method according to claim 6, characterized in that, The scheduling priority score is calculated by a weighted scoring function, specifically represented as: ； wherein, is the angle between the visual axis of the monitoring device and the direction of the alarm point, is the Euclidean distance between the monitoring device and the alarm point, is the physical occlusion state between the monitoring device and the alarm point, if there is no occlusion, then = 0; if occluded, then = 1; , , is a normalization function; , , is a weight coefficient, and .

8. The environmental safety intelligent monitoring and early warning analysis method according to claim 1, characterized in that, The fusion analysis in S3 specifically includes: Inputting the review video data into the side AI model for event rejudgment to obtain a video rejudgment confidence; extracting the N-class multi-dimensional monitoring data associated with the event from the data pool, obtaining the support score of the first class data through the corresponding data evaluation model; Based on the video rejudgment confidence and support score, an overall confidence score is obtained.

9. The environmental safety intelligent monitoring and early warning analysis method according to claim 8, characterized in that, The overall confidence score is represented as: ； wherein, is a weight for video re-identification confidence, is a weight for the first class of monitoring data support score, is a video re-identification confidence, is a weight for the first class of data support score.

10. The environmental safety intelligent monitoring and early warning analysis method according to claim 9, characterized in that, The hierarchical pre-warning-disposal mapping table in S4 is a dynamically configurable association table, which maps to the corresponding pre-warning level and disposal instruction set based on the joint query conditions of event type and overall confidence score interval.

Images