AI-based multi-target video behavior real-time analysis method and system

Abstract

This application provides an AI-based method and system for real-time analysis of multi-target video behavior. The method first receives real-time video stream data and real-time radiation field data; then, it performs spatiotemporal synchronous fusion of the real-time video stream data and real-time radiation field data to form a multimodal monitoring data stream; next, it dynamically identifies the matching status between personnel's work behavior and preset radiation safety standards to obtain a matching status identifier reflecting the degree of compliance of individual personnel's work behavior; then, it judges in real-time the radiation safety risk behaviors existing in multi-person collaborative operations and generates judgment results; finally, based on the judgment results, it automatically generates early warning instructions for radiation safety risk behaviors and sends them to the nuclear power plant's central control system. The technical solution provided by this application not only realizes full-process, automated real-time monitoring and proactive early warning of radiation safety risks in nuclear power plant personnel collaborative operations, but also improves the intelligence level and response efficiency of safety management.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method and system for real-time analysis of multi-target video behavior based on AI. Background Technology

[0002] In the field of safety monitoring of high-risk industrial facilities such as nuclear power plants, there is an urgent need for real-time, precise, and intelligent analysis of the behavior of workers. Such scenarios require the system to be able to track multiple targets simultaneously, not only to identify their individual actions, but also to understand the collaborative work patterns between personnel. Especially in high-radiation areas, it is necessary to dynamically assess whether their behavior complies with operating procedures in order to prevent the risk of collective radiation exposure caused by improper interaction, thereby realizing the transformation from passive alarm to proactive early warning.

[0003] One existing technical solution is a single-target video analysis system based on fixed rule thresholds. Such systems typically use computer vision technology to detect and track each worker individually, and preset static rules such as dwell time and activity area. When the behavior of a single target violates a rule, such as entering a restricted area or staying for an extended period, the system will trigger an independent alarm.

[0004] However, the existing solution has obvious limitations. Its main shortcomings are the single analytical perspective and the rigidity of the rule settings. Since the system only makes isolated judgments on a single target, it cannot effectively perceive and analyze the collaborative risks generated by the spatial relationships and behavioral interactions between multiple people, such as the surge in close-range radiation exposure caused by the passing of tools. At the same time, relying on fixed thresholds cannot adapt to the dynamically changing radiation field at the work site and complex temporary collaborative tasks, making it difficult to make accurate early warnings of real systemic safety risks caused by multi-target interactions, resulting in a high false alarm and false negative rate. Summary of the Invention

[0005] This application provides an AI-based method and system for real-time analysis of multi-target video behavior, which addresses the problems in existing technologies where the single analytical perspective and rigid rules prevent effective perception of collaborative work among multiple personnel and make it difficult to accurately warn of systemic radiation safety risks generated in dynamic interactive scenarios.

[0006] Firstly, this application provides an AI-based real-time multi-target video behavior analysis method, including:

[0007] The system receives real-time video stream data from a radiation-resistant camera and real-time radiation field data from a radiation dose sensor, wherein the real-time video stream data includes multi-person movement information and the real-time radiation field data includes dynamic radiation intensity distribution information.

[0008] The real-time video stream data and the real-time radiation field data are spatiotemporally fused to form a multimodal monitoring data stream;

[0009] Based on the multimodal monitoring data stream, the matching status between personnel's work behavior and the preset radiation safety specifications is dynamically identified to obtain a matching status identifier that reflects the degree of compliance of individual personnel's work behavior.

[0010] Based on the matching status identifier, the radiation safety risk behaviors existing in multi-person collaborative operations are judged in real time and the judgment result is generated.

[0011] Based on the judgment result, an early warning instruction for the radiation safety risk behavior is automatically generated and sent to the nuclear power plant's central control system.

[0012] Optionally, the real-time video stream data and the real-time radiation field data are spatiotemporally fused to form a multimodal monitoring data stream, and artificial intelligence technology is used to establish a correlation model between personnel movement and radiation exposure during the fusion process, including:

[0013] Extract video data segments containing personnel outlines and directions of movement from the real-time video stream data, and simultaneously extract radiation intensity readings corresponding to the timestamps of the video data segments from the radiation field data.

[0014] Establish a location mapping relationship between the personnel location information in each video data segment and the spatial region corresponding to the radiation intensity reading at the same time;

[0015] Based on the location mapping relationship, video data segments at the same time are combined with radiation intensity readings to form personnel radiation-related data units;

[0016] Based on multiple personnel radiation-related data units generated within a continuous time frame, a multimodal monitoring data stream is constructed, arranged in chronological order and containing both visual motion information and radiation environment information.

[0017] Optionally, based on the multimodal monitoring data stream, the matching status between personnel's work behavior and preset radiation safety specifications is dynamically identified to obtain a matching status identifier reflecting the degree of compliance of individual personnel's work behavior, including:

[0018] Behavioral pattern segments of each person are identified from the multimodal monitoring data stream. These behavioral pattern segments include the person's movement trajectory, the type of work action performed by the person, and radiation intensity information of the person's location.

[0019] Each person's behavioral pattern fragment is compared item by item with the corresponding clauses in the preset radiation safety regulations, which include the upper limit of allowed stay time, the limit of movement range, and the cumulative radiation exposure dose threshold for different work areas.

[0020] Generate behavioral compliance information for each person based on the comparison results;

[0021] Based on the compliance information of each person's behavior within a continuous time period, a matching status identifier reflecting the degree of compliance of a single person's work behavior is generated.

[0022] Optionally, based on the matching status identifier, the radiation safety risk behaviors existing in multi-person collaborative operations are judged in real time and a judgment result is generated, including:

[0023] Analyze the matching status indicators of each person to identify combinations of people who are simultaneously in high-radiation areas within a preset time window;

[0024] Determine the spatial distance between each person in the personnel group and detect the work actions that each person is performing in order to obtain the work action types and spatial distribution relationships of the personnel group;

[0025] Determine whether the types of work actions and spatial distribution of the personnel group exhibit any interactive patterns that could lead to an escalation of radiation exposure;

[0026] When it is determined that the interactive mode that triggers the escalation of radiation exposure exists, the radiation safety risk behavior is confirmed by combining the real-time radiation intensity information of the radiation area where the personnel group is located.

[0027] Based on the confirmed radiation safety risk behaviors, a judgment result is generated that includes the risk type, the identification of the personnel involved, and the risk level.

[0028] Optionally, determining whether the types of work actions and spatial distribution relationships of the personnel group exhibit an interaction pattern that could escalate radiation exposure includes:

[0029] Define a set of interaction modes related to radiation safety, including tool passing mode, close-range collaboration mode, and area cross-operation mode;

[0030] The relative movement trends between individuals in the personnel group are detected, as well as the positional changes between individuals and equipment in the personnel group.

[0031] The detected job action types and their spatial distribution are matched with the interaction patterns.

[0032] When the matching results show that the relationship between the type of work action and the spatial distribution matches any of the interaction modes, the cumulative radiation exposure dose of the personnel combination under the interaction mode is calculated within a preset time period.

[0033] The growth rate of the cumulative radiation exposure dose is compared with a preset dose growth rate threshold;

[0034] If the comparison results show that the growth rate of the cumulative radiation exposure dose exceeds the dose growth rate threshold, then an interaction mode that triggers an escalation of radiation exposure is confirmed.

[0035] Optionally, based on the judgment result, an early warning instruction for the radiation safety risk behavior is automatically generated, and the early warning instruction is sent to the nuclear power plant's central control system, including:

[0036] Based on the risk type and risk level contained in the judgment result, select the corresponding early warning instruction template from the preset instruction template library;

[0037] Fill the personnel identification and risk location information from the judgment results into the corresponding fields of the selected early warning instruction template to form a structured early warning instruction;

[0038] Add timestamps and priority markers to structured warning commands to obtain structured warning commands with timestamps and priority markers;

[0039] The structured early warning commands with timestamps and priority markers are transmitted to the security early warning interface of the nuclear power plant's central control system through the internal security communication network of the nuclear power plant.

[0040] After receiving the structured early warning command with timestamp and priority marker, the safety early warning interface of the nuclear power plant's central control system distributes the structured early warning command to the corresponding safety monitoring terminal according to the priority marker of the structured early warning command.

[0041] Optionally, the personnel identification and risk location information from the judgment result can be filled into the corresponding fields of the selected early warning instruction template to form a structured early warning instruction, including:

[0042] Identify the predefined field categories in the selected early warning instruction template, wherein the field categories include personnel identification field, geographical location field, risk description field, and time information field;

[0043] Extract the corresponding data content from the judgment result, wherein the corresponding data content includes the personnel number involved, the risk occurrence area number, the risk description field filled with the risk type and risk level text description, and the time when the risk behavior was identified;

[0044] According to the format defined in the early warning instruction template, fill in the personnel number involved in the personnel identification field, the risk occurrence area number in the geographical location field, the risk description field with a text description of the risk type and risk level, and the time information field with the time when the risk behavior was identified.

[0045] Logical validation is performed on the personnel identification field, geographical location field, risk description field, and time information field after the data content is entered, so that the correspondence between personnel identification and geographical location conforms to the preset personnel access rules for the nuclear power plant area.

[0046] The personnel identification field, geographical location field, risk description field, and time information field that have passed logical verification are combined and encapsulated to form a structured early warning instruction that conforms to the communication protocol of the nuclear power plant's central control system.

[0047] Secondly, this application provides an AI-based real-time multi-target video behavior analysis system, including:

[0048] The receiving module is used to receive real-time video stream data from a radiation-resistant camera and real-time radiation field data from a radiation dose sensor, wherein the real-time video stream data includes multi-person movement information and the real-time radiation field data includes dynamic radiation intensity distribution information.

[0049] The fusion module is used to perform spatiotemporal synchronous fusion of the real-time video stream data and the real-time radiation field data to form a multimodal monitoring data stream;

[0050] The identification module is used to dynamically identify the matching status between personnel's work behavior and preset radiation safety specifications based on the multimodal monitoring data stream, so as to obtain a matching status identifier that reflects the degree of compliance of individual personnel's work behavior.

[0051] The judgment module is used to judge the radiation safety risk behaviors existing in multi-person collaborative operations in real time based on the matching status identifier and generate the judgment result;

[0052] The generation module is used to automatically generate early warning instructions for the radiation safety risk behaviors based on the judgment results, and send the early warning instructions to the nuclear power plant's central control system.

[0053] Thirdly, this application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are to be invoked and executed by the processing component to implement the AI-based real-time video behavior analysis method as described in the first aspect above.

[0054] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements an AI-based real-time video behavior analysis method as described in the first aspect.

[0055] This application constructs a unified multimodal monitoring data stream by receiving and fusing video, temporal, and frequency data from radiation-resistant cameras and field data from radiation dose sensors, achieving panoramic and synchronous perception of the environment and personnel behavior in high-risk areas of nuclear power plants. Based on the fused data, this method dynamically identifies the matching status of each person's behavior with safety regulations and further comprehensively analyzes the collaborative operation patterns among multiple personnel. This enables the accurate identification of systemic radiation safety risks caused by personnel interactions that cannot be detected by independent analysis of a single target. Ultimately, it automatically generates and sends early warning commands, achieving a leap from passive alarm to proactive early warning, and improving the intelligence level and real-time response capability of nuclear power plant safety monitoring.

[0056] Furthermore, by predefined structured warning instruction templates and automatically filling in key information from the judgment results into the corresponding fields, the standardization and consistency of the generated warning instructions in terms of format and content are ensured. Logical validation of the filled information effectively guarantees the accuracy and rationality of the warning instruction content, for example, avoiding misreporting personnel outside the authorized area as risk sources. The final generated structured instructions, conforming to a specific communication protocol, can be seamlessly recognized and efficiently processed by the nuclear power plant's central control system, greatly improving the reliability of warning information flow and the efficiency of inter-system collaboration.

[0057] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 A flowchart of a multi-target video behavior real-time analysis method based on AI provided in this application is shown;

[0060] Figure 2 This paper presents a schematic diagram of the structure of a multi-target video behavior real-time analysis system based on AI provided in this application.

[0061] Figure 3 A schematic diagram of the structure of a computing device provided in this application is shown. Detailed Implementation

[0062] To enable those skilled in the art to better understand the present application, the technical solution of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0063] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0064] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0065] Figure 1 A flowchart of an AI-based real-time video behavior analysis method is provided for this application, as shown below. Figure 1 As shown, the method includes:

[0066] Step 101: Receive real-time video stream data from a radiation-resistant camera and real-time radiation field data from a radiation dose sensor, wherein the real-time video stream data includes multi-person movement information and the real-time radiation field data includes dynamic radiation intensity distribution information.

[0067] In the above scheme, real-time video stream data refers to a continuous sequence of images generated by a radiation-resistant camera, which contains the movement information of multiple workers in the picture and is used to record the activities of the workers in the monitored area.

[0068] A radiation dose sensor is a measuring device installed in the environment to detect the radiation level at various points in space;

[0069] Real-time radiation field data is dynamic radiation intensity distribution information continuously collected by radiation dose sensors, used to reflect the real-time changes in radiation intensity at different locations within the monitoring area;

[0070] Multi-person motion information is dynamic information extracted from real-time video stream data, including the movement trajectory, speed, and body movements of multiple individuals.

[0071] Dynamic radiation intensity distribution information is a spatial radiation dose map that changes over time and is obtained from real-time radiation field data. It is used to characterize the real-time spatial distribution of the radiation field.

[0072] In this scheme, firstly, multiple radiation-resistant cameras deployed in different corners of the nuclear power plant building synchronously acquire video images of the monitored area at a rate of 25 frames per second, forming a continuous real-time video stream. Secondly, a network of radiation dose sensors adjacent to the physical locations of the cameras is controlled to detect the radiation intensity at their respective points at the same sampling frequency, generating real-time radiation field data with a unified time reference. Subsequently, each frame of video image and each set of radiation readings is timestamped with millisecond precision. Then, based on the preset position correspondence between the cameras and sensors, the video frames and radiation readings acquired at the same time are initially spatially correlated. Finally, the system packages the video data with timestamps and spatial markers with the radiation data, completing the reception and preliminary organization of the raw data.

[0073] For example, during the refueling overhaul of Unit B at Nuclear Power Plant A, workers enter the reactor building lobby (Area B) to perform equipment maintenance. Radiation-resistant cameras C1 and C2, located on the ceiling of Area B, begin operating, continuously filming the work process of multiple personnel in the lobby and generating real-time video stream data. Simultaneously, radiation dose sensors S1 to S5, installed on the walls and equipment of Area B, collect gamma-ray and neutron radiation intensity in the area at specific frequencies, generating real-time radiation field data. The system receives these two types of data and establishes a correlation between a frame image captured by camera C1 at time T1 and the radiation reading measured by sensor S1 at time T1, marking the approximate radiation level within the image's field of view.

[0074] This step, by simultaneously receiving visual monitoring data and physical environmental radiation data, provides two key information sources for subsequent analysis: dynamic personnel behavior and changes in the radiation field. This lays the foundation for multimodal data analysis and ensures the real-time and comprehensiveness of the environmental information upon which behavioral analysis depends.

[0075] Step 102: The real-time video stream data and the real-time radiation field data are spatiotemporally fused to form a multimodal monitoring data stream.

[0076] Optionally, step 102 may specifically include the following steps:

[0077] Step 1021: Extract video data segments containing personnel outlines and movement directions from the real-time video stream data, and simultaneously extract radiation intensity readings corresponding to the time stamps of the video data segments from the radiation field data;

[0078] Step 1022: Establish a position mapping relationship between the personnel location information in each video data segment and the spatial region corresponding to the radiation intensity reading at the same time.

[0079] Step 1023: Based on the location mapping relationship, combine the video data segments at the same time with the radiation intensity readings to form a personnel radiation correlation data unit;

[0080] Step 1024: Based on multiple personnel radiation correlation data units generated within a continuous time period, construct a multimodal monitoring data stream arranged in chronological order, which simultaneously contains visual motion information and radiation environment information.

[0081] In the above scheme, multimodal monitoring data stream refers to integrating information from different sources into a unified data sequence arranged in chronological order;

[0082] Video data segments are short sets of images extracted from a continuous video stream, containing the outlines of specific people and their directions of movement;

[0083] A radiation intensity reading is a value of radiation intensity measured by a radiation dose sensor at a specific time and location.

[0084] Personnel location information refers to the coordinates of each person in a two-dimensional image determined from video frames;

[0085] The location mapping relationship is the rule for establishing the correspondence between the pixel coordinates of personnel and the coverage area of ​​the real-world radiation sensor;

[0086] The personnel radiation correlation data unit is a data structure that binds a person's movement state at the same moment with the radiation value of their location.

[0087] Visual motion information describes a person's posture and movement trajectory;

[0088] Radiation environment information characterizes the radiation intensity level of the area where a person is located.

[0089] In this scheme, firstly, the computer program analyzes the real-time video stream and identifies the contours of all people in each frame. A tracking algorithm calculates the movement direction of each person between consecutive frames, generating a video data segment for each person. Simultaneously, based on the time stamps recorded on these video data segments, the program selects all radiation intensity readings collected at the same time from the real-time radiation field data. Secondly, the system uses pre-calibrated parameters of the camera and radiation sensor to establish a coordinate transformation model. This model converts the location information of the people in the video data segments (such as image coordinates) into a real-world geographic coordinate system, thereby determining which spatial area the person is located within, thus establishing a positional mapping relationship between the person's location and the spatial source of the radiation readings. Then, the system... Based on the location mapping relationship established in step 1022, video data segments (containing personnel visual motion information) and radiation intensity readings (containing radiation environment information) representing the same time and spatial area are bundled together to create a personnel radiation-related data unit. This data unit is like a data packet, clearly recording "who did what, when, and where, and what the radiation value was at that time and place". Finally, the system arranges and links a large number of personnel radiation-related data units generated at multiple consecutive moments (e.g., 25 moments generated within one second) in chronological order, ultimately constructing a coherent multimodal monitoring data stream. This data stream is like a timeline containing visual and radiation information, completely recording the behavior of each person and the changes in their radiation environment over time.

[0090] Following the specific implementation of the previous scheme, in the scenario of the refueling and overhaul of Unit B at Nuclear Power Plant A, after receiving data from the C1 radiation-resistant camera and the S1 to S5 radiation dose sensors, the system begins to fuse the data. The program first identifies the outlines of personnel P1 and P2 from the video frame of the C1 camera at time T1, and determines that P1 is moving eastward, generating a video data segment containing this information. Simultaneously, it extracts the radiation intensity reading from the S1 sensor at time T1. Next, based on calibration information, the system calculates that personnel P1 is within the monitoring area of ​​sensor S1 at time T1, thus associating P1's movement with the S1 reading. Then, the system combines the video segment of P1's movement at time T1 with the S1 reading into a personnel radiation correlation data unit. The system continues this process, concatenating the data units generated at times T1, T2, T3, etc., about P1, P2, etc., in chronological order, ultimately forming a multimodal monitoring data stream describing the continuous changes in the radiation field and personnel operations within Area B.

[0091] This step precisely links the previously independent video surveillance information and radiation field information in time and space, creating a more comprehensive data view. This allows subsequent analysis to simultaneously consider the dynamic behavior of personnel and the real-time radiation risk in their area, providing a reliable data foundation for accurately assessing the safety status under collaborative operations.

[0092] Step 103: Based on the multimodal monitoring data stream, dynamically identify the matching status between personnel's work behavior and preset radiation safety standards to obtain a matching status identifier that reflects the degree of compliance of individual personnel's work behavior.

[0093] Optionally, step 103 may specifically include the following steps:

[0094] Step 1031: Identify the behavior pattern segment of each person from the multimodal monitoring data stream. The behavior pattern segment includes the person's movement trajectory, the type of work action of the person, and the radiation intensity information of the person's location.

[0095] Step 1032: The behavior pattern fragments of each person are compared item by item with the corresponding clauses in the preset radiation safety regulations. The preset radiation safety regulations include the upper limit of the allowed stay time, the limit of the range of motion, and the cumulative radiation exposure dose threshold for different work areas.

[0096] Step 1033: Generate behavioral compliance information for each person based on the comparison results;

[0097] Step 1034: Based on the compliance information of each person's behavior within a continuous time period, generate a matching status identifier that reflects the degree of compliance of the individual person's work behavior.

[0098] In the above scheme, personnel work behavior refers to the various operational activities carried out by staff within the radiation area;

[0099] The matching status of preset radiation safety regulations refers to the degree of compliance obtained by comparing actual behavior with safety regulations;

[0100] The matching status identifier is a quantified result label used to intuitively represent the degree of compliance of an individual's behavior;

[0101] A behavioral pattern segment is a set of information extracted from continuous data that describes a person’s complete activities within a specific time period. It includes the person’s movement trajectory (i.e., the path of movement), the type of work action (such as walking, bending over, operating equipment, etc.), and the radiation intensity information of the person’s location (i.e., the radiation value at that location at that time).

[0102] Pre-defined radiation safety regulations are a set of explicit rules that include the maximum allowable stay time (maximum stay time), movement range restrictions (allowed range of activities), and cumulative radiation exposure dose thresholds (maximum total radiation that can be tolerated over a period of time) for different work areas (such as high radiation areas and low radiation areas).

[0103] Behavioral compliance information is generated through comparison and reflects the specific results of whether a behavior conforms to the various clauses of the norm.

[0104] In this scheme, firstly, the computer program scans the multimodal monitoring data stream and uses a behavior segmentation algorithm to cut the continuous data stream into several short-term behavior pattern segments, each focusing on a single individual, based on the identifiers of different personnel. Each behavior pattern segment is extracted, containing the individual's movement trajectory over a period of time, the identified main work actions (such as determining whether it is carrying or inspection through skeletal key point analysis), and radiation intensity information corresponding to each location of the individual obtained from the fused data. Secondly, the system compares each behavior pattern segment obtained in step 1031 with the preset radiation safety regulations in the database item by item. For example, the system will determine whether the area where the individual is located matches their permissions and calculate whether their current stay time in the area exceeds the limits set for that area. The system first sets an upper limit for dwell time, then analyzes whether the range of motion (e.g., arm extension) exceeds the limit, and accumulates the radiation exposure dose received during this work cycle to see if it approaches or exceeds the cumulative radiation exposure dose threshold. Next, based on the comparison results of each item in step 1032 (e.g., "dwell time not exceeded," "range of motion exceeded the limit," "cumulative dose safe"), the system performs a comprehensive assessment to generate information on the person's behavioral compliance during this time period. This information quantifies the overall compliance of their behavior with safety regulations. Finally, the system performs trend analysis not based on a single instant but on multiple behavioral compliance data generated over a continuous period (e.g., the past five minutes), ultimately generating a comprehensive matching status label for the person that reflects the overall compliance level of their recent work behavior. This label can be a simple level (e.g., "Good," "Caution," "Warning"), thus providing a dynamically updated compliance snapshot.

[0105] Following the specific implementation of the previous scheme, in step 102, in the scenario of Area B of Nuclear Power Plant A, the system has formed a multimodal monitoring data stream containing the behaviors of personnel P1 and P2. The program first extracts the behavior pattern segment of personnel P1 in the past 30 seconds from this multimodal monitoring data stream, showing that P1's movement trajectory is from equipment A to equipment B, the type of operation is mainly handheld tool inspection, and the radiation intensity information of its location is provided by sensors S1 and S2, and the value is within the allowable range; then, the system compares this behavior pattern segment with the preset radiation safety specifications of the "equipment inspection area" in Area B: the dwell time (30 seconds) does not exceed the upper limit (10 minutes), the action range (inspection action) complies with the regulations, and the cumulative radiation dose is far below the threshold; the comparison result shows that all are in compliance, so the behavior compliance information of P1 is generated as "fully compliant"; the system integrates the information of P1's continuous "fully compliant" over the past few minutes, and finally generates the current matching status identifier of P1 as "safe".

[0106] This step involves meticulously extracting behavioral fragments from the fused data and automatically and multidimensionally comparing them with detailed safety rules. Ultimately, it generates a dynamically updated compliance identifier for each individual, enabling continuous and quantitative assessment of the safety of individual work behaviors. This transforms complex safety regulations into real-time monitorable indicators, providing precise individual behavioral data for subsequent assessment of group collaboration risks.

[0107] Step 104: Based on the matching status identifier, determine the radiation safety risk behaviors existing in multi-person collaborative operations in real time and generate the judgment result.

[0108] Optionally, step 104 may specifically include the following steps:

[0109] Step 1041: Analyze the matching status identifier of each person to identify combinations of people who are simultaneously in high radiation areas within a preset time window;

[0110] Step 1042: Determine the spatial distance between each person in the personnel group and detect the work actions that each person is performing, so as to obtain the work action type and spatial distribution relationship of the personnel group;

[0111] Step 1043: Determine whether the types of work actions and spatial distribution of the personnel group have an interactive pattern that could lead to an escalation of radiation exposure.

[0112] Step 1043 may specifically include the following steps:

[0113] Define a set of interaction modes related to radiation safety, including tool passing mode, close-range collaboration mode, and area cross-operation mode; detect the relative movement trends between personnel in the personnel group, as well as the positional changes between personnel and equipment in the personnel group; match the detected operation action types and spatial distribution relationships with the interaction modes; when the matching results show that the operation action type and spatial distribution relationship conform to any of the interaction modes, calculate the cumulative radiation exposure dose of the personnel group in the interaction mode within a preset time period; compare the growth rate of the cumulative radiation exposure dose with a preset dose growth rate threshold; when the comparison results show that the growth rate of the cumulative radiation exposure dose exceeds the dose growth rate threshold, it is confirmed that there is an interaction mode that triggers an escalation of radiation exposure.

[0114] Step 1044: When it is determined that the interaction mode that triggers the escalation of radiation exposure exists, the radiation safety risk behavior is confirmed by combining the real-time radiation intensity information of the radiation area where the personnel group is located.

[0115] Step 1045: Based on the confirmed radiation safety risk behaviors, generate a judgment result that includes the risk type, the identification of the personnel involved, and the risk level.

[0116] In the above scheme, radiation safety risk behavior refers to unsafe interactive behaviors that may increase collective radiation exposure when multiple people work together.

[0117] The judgment result is the system's conclusive output on the identified risky behaviors;

[0118] The preset time window is a continuous time period set by the system;

[0119] A group of people in a high-radiation area refers to a group of people who are simultaneously in a high-radiation area within a preset time window.

[0120] The type of work action refers to the category of operations that personnel are performing;

[0121] Spatial distribution describes the relative positions and distances between people;

[0122] The interactive pattern of escalating radiation exposure refers to a specific collaborative approach that leads to an abnormal increase in the cumulative radiation dose to personnel.

[0123] The interaction modes related to radiation safety are several predefined typical high-risk collaboration scenarios, such as tool transfer mode, close-range collaboration mode, and cross-regional operation mode.

[0124] Relative motion trend refers to the movement of people towards or away from each other;

[0125] The positional change relationship describes the dynamic distance between personnel and key equipment;

[0126] Cumulative radiation exposure dose is the total amount of radiation a person receives within a specific time period;

[0127] The preset dose increase rate threshold is a specified safe limit for the rate of dose increase;

[0128] Real-time radiation intensity information is the instantaneous radiation level at a person's current location.

[0129] In this solution, firstly, the system continuously monitors the matching status identifiers of all personnel. When it detects that within a preset time window (e.g., the past 30 seconds), two or more personnel are simultaneously located in high-radiation areas, these personnel are marked as a group requiring close monitoring. Secondly, the system performs a detailed analysis of this group: determining spatial distribution relationships by calculating the coordinate distances between personnel, and identifying the specific work actions each person is performing using behavioral recognition technology, thereby comprehensively understanding the group's operational status. Next, the system further assesses whether this group poses a risk: several radiation safety-related interaction modes are pre-set, such as tool passing and close-range collaboration; the system detects the relative movement trends between personnel (whether they are getting closer to each other) and their interactions with others. The system analyzes the location changes of equipment; it matches the observed types and spatial distribution of work actions with preset patterns; if the match is successful, it calculates the cumulative radiation exposure dose of personnel under that interaction pattern in the recent period; it compares the rate of increase of this dose with a preset dose rate of increase threshold; if the rate of increase is too fast, it confirms the existence of an interaction pattern that could escalate radiation exposure; then, when an interaction pattern that could escalate radiation exposure is confirmed, the system combines the real-time radiation intensity information of the area where the personnel group is located for final confirmation to determine whether this constitutes a radiation safety risk behavior that requires immediate intervention; finally, based on the confirmed radiation safety risk behavior, the system generates a structured judgment result, which clearly records the specific type of risk, the personnel involved, and the current risk level.

[0130] Following the specific implementation of the previous scheme, in step 103, in the scenario of Area B of Nuclear Power Plant A, the matching status identifiers of personnel P1 and P2 both indicate that they have been working continuously in the high-radiation area of ​​Area B. The system detects this situation and identifies them as a group of personnel in a high-radiation area. Analysis reveals that P1 and P2 are less than 1 meter apart and are both working facing the same equipment, exhibiting a close-range collaboration mode. The system calculates that the cumulative radiation exposure dose of this group due to close collaboration has increased rapidly in the past 2 minutes, exceeding the threshold. Combined with the high real-time radiation readings in the area, the system confirms the existence of radiation safety risk behavior and finally generates a judgment result: the risk type is "unauthorized close-range collaboration", involving personnel P1 and P2, and the risk level is "high".

[0131] This step identifies potentially high-risk collaborative work groups from individual compliance status and deeply analyzes the correlation between their interaction patterns and radiation exposure. This enables accurate assessment of systemic radiation risks arising from multi-person collaborative work, elevating safety monitoring from individual behavior assessment to group interaction risk assessment, and providing a clear decision-making basis for timely intervention.

[0132] Step 105: Based on the judgment result, an early warning instruction for the radiation safety risk behavior is automatically generated, and the early warning instruction is sent to the nuclear power plant's central control system.

[0133] Optionally, step 105 may specifically include the following steps:

[0134] Step 1051: Based on the risk type and risk level contained in the judgment result, select the corresponding early warning instruction template from the preset instruction template library;

[0135] Step 1052: Fill the personnel identification and risk location information from the judgment result into the corresponding fields of the selected early warning instruction template to form a structured early warning instruction;

[0136] Step 1052 may specifically include the following steps:

[0137] Identify predefined field categories in the selected early warning instruction template, including personnel identification, geographic location, risk description, and time information fields. Extract corresponding data content from the judgment result, including the personnel number involved, the risk occurrence area number, a text description of the risk type and risk level in the risk description field, and the time when the risk behavior was identified. Following the format defined in the early warning instruction template, fill in the personnel number involved in the personnel identification field, the risk occurrence area number in the geographic location field, the text description of the risk type and risk level in the risk description field, and the time when the risk behavior was identified in the time information field. Perform logical validation on the personnel identification, geographic location, risk description, and time information fields after inserting the data content to ensure that the correspondence between personnel identification and geographic location conforms to the preset nuclear power plant area personnel access rules. Combine and encapsulate the logically validated personnel identification, geographic location, risk description, and time information fields to form a structured early warning instruction that conforms to the communication protocol of the nuclear power plant central control system.

[0138] Step 1053: Add timestamps and instruction priority markers to the structured warning instructions to obtain structured warning instructions with timestamps and priority markers;

[0139] Step 1054: Transmit structured early warning instructions with timestamps and priority markers to the security early warning interface of the nuclear power plant's central control system through the internal security communication network of the nuclear power plant.

[0140] Step 1055: After receiving the structured early warning instruction with timestamp and priority mark, the safety early warning interface of the nuclear power plant central control system distributes the structured early warning instruction to the corresponding safety monitoring terminal according to the priority mark of the structured early warning instruction.

[0141] In the above scheme, the early warning command is a standardized command generated by the system to alert users to risks;

[0142] The central control system of a nuclear power plant is the core platform for receiving and processing safety information from the entire plant.

[0143] Risk type refers to the type of risk (such as timeout for close collaboration);

[0144] Risk levels indicate the severity of a risk (e.g., high, medium, low).

[0145] The default instruction template library is a database that stores various standard instruction formats;

[0146] Warning instruction templates are blank instruction frames of a specific format in the library;

[0147] Personnel identification is a unique number assigned to each person in the system;

[0148] The risk location information is the specific area number where the risky behavior occurred;

[0149] Structured warning commands are warning commands that are filled with specific information and have a standardized format.

[0150] Predefined field categories are information categories specified in the template (such as who, where, what risk, when);

[0151] The personnel identification field is where you enter the personnel number;

[0152] The geolocation field is used to enter the area code;

[0153] The risk description field is where the risk type and level are described;

[0154] The time information field is used to record the location of when the risk occurred;

[0155] The corresponding data content is the specific information extracted from the judgment result (such as personnel number P001); the personnel number is the unique code of the personnel (such as P001).

[0156] The risk zone number is a unique code for the zone (e.g., Zone-B-1).

[0157] The "Risk Description" field requires you to enter the specific nature of the risk when you enter the "Risk Type" field.

[0158] The textual description of the risk level is a textual explanation of the risk level (such as "high risk").

[0159] The moment when a risky behavior is identified is the specific point in time when the system discovers the risk;

[0160] The pre-defined personnel access rules for nuclear power plant areas are a safety system that stipulates which personnel are authorized to enter which areas;

[0161] The communication protocol of the central control system of a nuclear power plant is the format convention that must be followed when exchanging data between systems;

[0162] A timestamp is a precise time stamp attached when an instruction is created;

[0163] Instruction priority marking is an indicator of the urgency of an instruction based on its risk level;

[0164] The safety early warning interface is the program entry point for the central control system to receive early warning information.

[0165] The corresponding security monitoring terminal is a specific workstation or screen that needs to receive the instruction, determined according to the risk type and level.

[0166] In this solution, firstly, the system reads the risk type (e.g., "unauthorized close collaboration") and risk level (e.g., "high") from the judgment result. Then, based on these two key pieces of information, it searches for and selects the most matching warning instruction template (e.g., "high-risk - personnel behavior" template) from a pre-set instruction template library. Secondly, the system begins to populate the template: it first identifies several predefined field categories in the selected template, mainly including a personnel identification field that requires personnel ID, a geographic location field that requires area ID, a risk description field that requires describing risk details, and a time information field that requires recording the time of occurrence. Then, it extracts the corresponding data content from the judgment result, such as the personnel ID involved (P001, ...). The system first inputs data into the risk zone (Zone-B-1), then enters the risk type and level ("Unauthorized close collaboration, high risk"), and the precise time the risk was identified. Following the template definition, the system fills in these data points into the corresponding fields. After completion, the system performs a logical check to verify that the entered personnel number has the necessary permissions to enter the designated zone, ensuring compliance with nuclear power plant personnel access rules and preventing logical errors such as "unauthorized personnel in a restricted area." After verification, the system combines these fields and encapsulates them into a structured warning instruction that conforms to the communication protocol requirements of the nuclear power plant's central control system. Next, the system adds a current timestamp to this structured warning instruction and determines the risk level (P002). For example, a "high" warning command is assigned a corresponding instruction priority tag (such as "emergency"), thus obtaining a complete structured warning command with a timestamp and priority tag. Then, the system reliably transmits this tagged structured warning command to the nuclear power plant's central control system's externally accessible security warning interface, which is specifically used to receive alarms, through the nuclear power plant's dedicated, highly reliable security communication network. Finally, after successfully receiving this structured warning command, the nuclear power plant's central control system's security warning interface immediately parses the instruction priority tag it carries and, based on this priority, automatically distributes the structured warning command to the corresponding safety monitoring terminal currently responsible for handling such emergency situations, such as the large screen in the main control room or the dedicated workstation of the safety engineer, thus completing the final delivery of the warning information.

[0167] Following the specific implementation of the previous solution, in step 104, in the scenario of Area B of Nuclear Power Plant A, the system generates a high-risk judgment result regarding personnel P1 and P2 for "unauthorized close collaboration". Based on this, the system selects the "High Risk - Personnel Behavior" template, fills in the personnel numbers P001 and P002, area number B-1, risk description "unauthorized close collaboration, high risk" and occurrence time into the template, and after verifying that P1 and P2 are authorized to work in Area B-1, it encapsulates it into a structured early warning instruction. The system adds a current timestamp and an "emergency" priority mark to it, and sends it to the safety early warning interface of the central control system through the internal security network. After receiving it, the interface immediately displays the early warning instruction on the chief safety officer's monitoring terminal in the main control room according to the "emergency" mark.

[0168] This step transforms unstructured risk assessment results into standardized, machine-processable structured instructions, ensuring accuracy, compliant format, and clear priorities throughout the entire process of instruction generation, verification, transmission, and distribution. Ultimately, the instructions are precisely delivered to the appropriate handling positions, achieving an automated closed loop from risk identification to efficient and reliable delivery of early warning information. This significantly improves the response speed and accuracy of nuclear power plant safety management.

[0169] Figure 2 This application provides a schematic diagram of the structure of an AI-based multi-target video behavior real-time analysis system, such as... Figure 2 As shown, the system includes:

[0170] The receiving module 21 is used to receive real-time video stream data from a radiation-resistant camera and real-time radiation field data from a radiation dose sensor, wherein the real-time video stream data includes multi-person movement information and the real-time radiation field data includes dynamic radiation intensity distribution information.

[0171] The fusion module 22 is used to perform spatiotemporal synchronous fusion of the real-time video stream data and the real-time radiation field data to form a multimodal monitoring data stream;

[0172] The identification module 23 is used to dynamically identify the matching status between personnel's work behavior and preset radiation safety specifications based on the multimodal monitoring data stream, so as to obtain a matching status identifier that reflects the degree of compliance of individual personnel's work behavior.

[0173] The judgment module 24 is used to judge the radiation safety risk behaviors existing in multi-person collaborative operations in real time based on the matching status identifier and generate a judgment result;

[0174] The generation module 25 is used to automatically generate an early warning instruction for the radiation safety risk behavior based on the judgment result, and send the early warning instruction to the nuclear power plant central control system.

[0175] Figure 2 The aforementioned AI-based multi-target video behavior real-time analysis system can perform... Figure 1 The implementation principle and technical effects of the AI-based real-time video behavior analysis method described in the illustrated embodiment will not be repeated here. The specific methods by which each module and unit of the AI-based real-time video behavior analysis system in the above embodiments perform operations have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0176] In one possible design, Figure 2 The AI-based real-time video behavior analysis system of the embodiment shown can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0177] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0178] The processing component 32 is used for the above Figure 1 The embodiment describes an AI-based real-time analysis method for multi-target video behavior.

[0179] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0180] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0181] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0182] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0183] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0184] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0185] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown is an AI-based real-time video behavior analysis method.

[0186] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0187] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0188] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0189] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A real-time multi-target video behavior analysis method based on AI, characterized in that, include: The system receives real-time video stream data from a radiation-resistant camera and real-time radiation field data from a radiation dose sensor, wherein the real-time video stream data includes multi-person movement information and the real-time radiation field data includes dynamic radiation intensity distribution information. The real-time video stream data and the real-time radiation field data are spatiotemporally fused to form a multimodal monitoring data stream; Based on the multimodal monitoring data stream, the matching status between personnel's work behavior and the preset radiation safety specifications is dynamically identified to obtain a matching status identifier that reflects the degree of compliance of individual personnel's work behavior. Based on the matching status identifier, the real-time judgment of radiation safety risk behaviors in multi-person collaborative operations and the generation of judgment results include: analyzing the matching status identifier of each person to identify combinations of people simultaneously located in high-radiation areas within a preset time window; determining the spatial distance between each person in the combination and detecting the work actions being performed by each person to obtain the work action type and spatial distribution relationship of the combination; judging whether the work action type and spatial distribution relationship of the combination of people have an interaction mode that could escalate radiation exposure; when it is determined that the interaction mode that could escalate radiation exposure exists, confirming the radiation safety risk behavior by combining the real-time radiation intensity information of the radiation area where the combination of people is located; and generating a judgment result containing the risk type, the identifier of the person involved, and the risk level based on the confirmed radiation safety risk behavior. Based on the judgment result, an early warning instruction for the radiation safety risk behavior is automatically generated and sent to the nuclear power plant's central control system.

2. The method according to claim 1, characterized in that, The real-time video stream data and the real-time radiation field data are spatiotemporally fused to form a multimodal monitoring data stream, including: Extract video data segments containing personnel outlines and directions of movement from the real-time video stream data, and simultaneously extract radiation intensity readings corresponding to the timestamps of the video data segments from the radiation field data. Establish a location mapping relationship between the personnel location information in each video data segment and the spatial region corresponding to the radiation intensity reading at the same time; Based on the location mapping relationship, video data segments at the same time are combined with radiation intensity readings to form personnel radiation-related data units; Based on multiple personnel radiation-related data units generated within a continuous time frame, a multimodal monitoring data stream is constructed, arranged in chronological order and containing both visual motion information and radiation environment information.

3. The method according to claim 1, characterized in that, Based on the multimodal monitoring data stream, the matching status between personnel's work behavior and preset radiation safety specifications is dynamically identified to obtain a matching status identifier reflecting the degree of compliance of individual personnel's work behavior, including: Behavioral pattern segments of each person are identified from the multimodal monitoring data stream. These behavioral pattern segments include the person's movement trajectory, the type of work action performed by the person, and radiation intensity information of the person's location. Each person's behavioral pattern fragment is compared item by item with the corresponding clauses in the preset radiation safety regulations, which include the upper limit of allowed stay time, the limit of movement range, and the cumulative radiation exposure dose threshold for different work areas. Generate behavioral compliance information for each person based on the comparison results; Based on the compliance information of each person's behavior within a continuous time period, a matching status identifier reflecting the degree of compliance of a single person's work behavior is generated.

4. The method according to claim 1, characterized in that, Determining whether the types of work actions and spatial distribution relationships of the personnel group exhibit any interactive patterns that could escalate radiation exposure includes: Define a set of interaction modes related to radiation safety, including tool passing mode, close-range collaboration mode, and area cross-operation mode; The relative movement trends between individuals in the personnel group are detected, as well as the positional changes between individuals and equipment in the personnel group. The detected job action types and their spatial distribution are matched with the interaction patterns. When the matching results show that the relationship between the type of work action and the spatial distribution matches any of the interaction modes, the cumulative radiation exposure dose of the personnel combination under the interaction mode is calculated within a preset time period. The growth rate of the cumulative radiation exposure dose is compared with a preset dose growth rate threshold; If the comparison results show that the growth rate of the cumulative radiation exposure dose exceeds the dose growth rate threshold, then an interaction mode that triggers an escalation of radiation exposure is confirmed.

5. The method according to claim 1, characterized in that, Based on the judgment result, an early warning instruction is automatically generated for the radiation safety risk behavior, and the early warning instruction is sent to the nuclear power plant's central control system, including: Based on the risk type and risk level contained in the judgment result, select the corresponding early warning instruction template from the preset instruction template library; Fill the personnel identification and risk location information from the judgment results into the corresponding fields of the selected early warning instruction template to form a structured early warning instruction; Add timestamps and priority markers to structured warning commands to obtain structured warning commands with timestamps and priority markers; The structured early warning commands with timestamps and priority markers are transmitted to the security early warning interface of the nuclear power plant's central control system through the internal security communication network of the nuclear power plant. After receiving the structured early warning command with timestamp and priority marker, the safety early warning interface of the nuclear power plant's central control system distributes the structured early warning command to the corresponding safety monitoring terminal according to the priority marker of the structured early warning command.

6. The method according to claim 5, characterized in that, The personnel identification and risk location information from the judgment results are entered into the corresponding fields of the selected early warning instruction template to form a structured early warning instruction, including: Identify the predefined field categories in the selected early warning instruction template, wherein the field categories include personnel identification field, geographical location field, risk description field, and time information field; Extract the corresponding data content from the judgment result, wherein the corresponding data content includes the personnel number involved, the risk occurrence area number, the risk description field filled with the risk type and risk level text description, and the time when the risk behavior was identified; According to the format defined in the early warning instruction template, fill in the personnel number involved in the personnel identification field, the risk occurrence area number in the geographical location field, the risk description field with a text description of the risk type and risk level, and the time information field with the time when the risk behavior was identified. Logical validation is performed on the personnel identification field, geographical location field, risk description field, and time information field after the data content is entered, so that the correspondence between personnel identification and geographical location conforms to the preset personnel access rules for the nuclear power plant area. The personnel identification field, geographical location field, risk description field, and time information field that have passed logical verification are combined and encapsulated to form a structured early warning instruction that conforms to the communication protocol of the nuclear power plant's central control system.

7. A real-time multi-target video behavior analysis system based on AI, applied to any one of the real-time multi-target video behavior analysis methods based on AI according to claims 1-6, characterized in that, include: The receiving module is used to receive real-time video stream data from a radiation-resistant camera and real-time radiation field data from a radiation dose sensor, wherein the real-time video stream data includes multi-person movement information and the real-time radiation field data includes dynamic radiation intensity distribution information. The fusion module is used to perform spatiotemporal synchronous fusion of the real-time video stream data and the real-time radiation field data to form a multimodal monitoring data stream; The identification module is used to dynamically identify the matching status between personnel's work behavior and preset radiation safety specifications based on the multimodal monitoring data stream, so as to obtain a matching status identifier that reflects the degree of compliance of individual personnel's work behavior. The judgment module is used to judge the radiation safety risk behaviors existing in multi-person collaborative operations in real time based on the matching status identifier and generate the judgment result; The generation module is used to automatically generate early warning instructions for the radiation safety risk behaviors based on the judgment results, and send the early warning instructions to the nuclear power plant's central control system.

8. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the AI-based real-time video behavior analysis method as described in any one of claims 1 to 6.

9. A computer storage medium, characterized in that, The device contains a computer program that, when executed by a computer, implements an AI-based real-time video behavior analysis method as described in any one of claims 1 to 6.

Images