An AI image recognition-based limited space operation risk monitoring and alarm method

Abstract

The present application relates to the technical field of safety monitoring and image recognition, in particular to a limited space operation risk monitoring and alarm method based on AI image recognition, comprising the following steps: S1: acquiring multi-angle video image streams of a limited space operation area in real time; S2: extracting a dangerous source feature set and a personnel behavior feature set in the image; S3: generating a static risk level of the current environment according to the dangerous source feature set; S4: calculating a posture stability parameter and a space occupation parameter of the current posture of the operator; S5: forming a dynamic behavior risk index; S6: performing weighted superposition and generating corresponding graded early warning signals. The present application realizes multi-dimensional risk identification and graded early warning of the limited space operation environment by fusing posture stability, space occupation degree and illegal behavior recognition results, and improves the accuracy of risk perception and the intelligent level of response.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring and image recognition technology, and in particular to a method for monitoring and alarming risks in confined space operations based on AI image recognition. Background Technology

[0002] Confined space operations are widespread in typical enclosed or semi-enclosed locations such as municipal pipelines, underground tunnels, storage tanks, and containers. These locations have complex internal structures, limited space, and are often accompanied by hazards such as toxic gas leaks, oxygen deficiency, high temperatures, or falls, making them high-risk areas for work-related accidents. To ensure operational safety, most existing monitoring systems use fixed video surveillance or portable gas detectors for single-point alarms, but these systems struggle to achieve comprehensive risk identification based on personnel behavior and environmental conditions, and lack a real-time judgment mechanism based on the relationship between dynamic personnel behavior and the distribution of hazard sources. Therefore, how to integrate image recognition technology with personnel behavior perception to build a more real-time and intelligent confined space risk identification mechanism has become a key direction for improving safety management.

[0003] However, existing technologies for image-based risk identification generally suffer from the following problems: First, they fail to quantify the stability and proximity to danger of workers based on their three-dimensional posture and spatial orientation information; second, the identification of hazards typically lacks systematic quantitative assessment, making it impossible to generate structured static risk levels; and third, the identification of personnel violations only records events without integrating them with posture or environmental conditions, resulting in delayed risk warnings and a high false alarm rate. Therefore, there is an urgent need for an AI-based image recognition-based method for risk monitoring and alarming in confined spaces to address these issues. Summary of the Invention

[0004] To achieve the above objectives, this invention provides a method for risk monitoring and alarm of confined space operations based on AI image recognition.

[0005] A method for risk monitoring and alarm in confined space operations based on AI image recognition includes the following steps: S1: Real-time acquisition of multi-angle video image streams of the confined space operation area, and preprocessing them to obtain a standardized image sequence; S2: Perform parallel analysis on standardized image sequences using an AI image recognition engine to extract sets of hazard source features and sets of personnel behavior features from the images; S3: Based on the set of hazard source characteristics, match the preset hazard source database to generate the static risk level of the current environment; S4: Based on the set of personnel behavior characteristics, calculate the posture stability parameters and space occupancy parameters of the current posture of the operator; S5: Based on the attitude stability parameters and space occupancy parameters, conduct attitude space risk coupling analysis to generate attitude space risk coefficients; and combine the results of identifying other explicit violations by personnel to jointly form a dynamic behavior risk index; S6: The static risk level and the dynamic behavioral risk index are weighted and superimposed to obtain a comprehensive risk coefficient. When the comprehensive risk coefficient exceeds the preset threshold, a corresponding graded early warning signal is generated and sent to the monitoring terminal.

[0006] Optionally, S1 specifically includes: S11: Set up no less than two video acquisition devices in the confined space operation area, facing the operation entrance channel and the core operation surface respectively, and assign a unique device identifier to each video acquisition device; synchronously acquire the real-time video stream output by each video acquisition device at a uniform sampling frame rate to form a multi-angle video image stream distinguished by device identifier; S12: Perform timestamp alignment processing on the multi-angle video image streams formed in S11, map the frame data of each video stream to a unified time axis, and perform frame sequence completion processing on missing frames to obtain a time-synchronized multi-angle frame sequence. S13: Perform image normalization preprocessing on the time-synchronized multi-angle frame sequence, including resolution unification, pixel intensity normalization and image denoising, and output a normalized frame sequence that meets the preset input specifications. S14: Perform viewpoint consistency processing on the standardized frame sequence. Based on the fixed installation pose parameters of each video acquisition device, convert the standardized frame sequence under different viewpoints to the image representation under a unified coordinate system, and arrange them in a unified time axis order to form a standardized image sequence.

[0007] Optionally, S2 specifically includes: S21: Receive the standardized image sequence output by S1, divide the standardized image sequence into a frame-by-frame input queue according to a unified timeline, and bind each frame image to the corresponding timestamp and acquisition viewpoint identifier to form a structured input frame set for parallel analysis. S22: Input the structured input frame set into the AI ​​image recognition engine, perform synchronous inference processing on multi-view frames at the same timestamp, and obtain the candidate detection results of hazard sources and candidate detection results of personnel corresponding to each viewpoint; S23: Perform category confirmation and location parameter extraction on the candidate detection results of the hazard source, output the hazard source category label, the coordinates of the hazard source target box and the hazard source edge contour parameters, and assemble the hazard source category label and the hazard source location parameters into a hazard source feature set; S24: Perform human key point detection and human posture parameter extraction on the candidate detection results of personnel, output the three-dimensional coordinates of human key points, body orientation and personnel contour parameters of the workers, and assemble the three-dimensional coordinates of human key points, body orientation and personnel contour parameters into a set of personnel behavior features. S25: Align the hazard source feature set formed in S23 with the personnel behavior feature set formed in S24 by timestamp to obtain the hazard source feature set-personnel behavior feature set pairing result for each timestamp.

[0008] Optionally, S3 specifically includes: S31: Receive the set of hazard source features output by S2, specifically including hazard source category labels, hazard source target box coordinates and hazard source edge contours, and organize each hazard source record into a structured hazard source entry; S32: Call the preset hazard source database, use the hazard source category label of each structured hazard source entry as the search key to perform a matching query, and obtain the set of basic risk parameters of the hazard source that corresponds one-to-one with the hazard source category label, including the basic severity coefficient and the category weight coefficient; S33: Calculate the hazard occupancy ratio based on the hazard edge contour parameters of each structured hazard source entry, and obtain the hazard exposure coefficient accordingly; S34: Based on the baseline severity coefficient and category weight coefficient obtained in S32, and combined with the exposure coefficient obtained in S33, calculate the static risk contribution value of each hazard. ; S35: Weighted summation of the static risk contribution values ​​of all hazard sources at the same time stamp to obtain the static risk score of the current environment. And map the static risk score to the static risk level.

[0009] Optionally, S4 specifically includes: S41: Based on the three-dimensional coordinates of key points of the human body, select key points related to human body support to form a support plane, and calculate the projection position of the human body's center of gravity on the support plane to obtain the center of gravity projection parameters. S42: Based on the three-dimensional coordinates of the human body's center of gravity obtained in S41 and the normal vector of the supporting plane, calculate the stability angle of the human body's posture relative to the supporting plane. As one of the attitude stability parameters, its calculation formula is: ,in, For stability angle; The normal vector of the supporting plane; It is the attitude vector pointing from the body's center of gravity to the direction the body is facing; S43: Combine the centroid projection parameters obtained in S41 with the stability angle obtained in S42 to form the attitude stability parameters.

[0010] Optionally, S4 further includes: S44: Based on the personnel contour parameters, calculate the Euclidean distance from each sampling point on the boundary of the worker's contour to the nearest hazard source edge or obstacle edge in the hazard source feature set, and take the minimum value as the space occupancy parameter. The calculation formula is as follows: ,in, This refers to the space usage parameter; The first on the personnel outline boundary Coordinates of each sampling point; The first one on the edge of the hazard source or obstacle The coordinates of the boundary points.

[0011] Optionally, S5 specifically includes: S51: Receives the attitude stability parameters and space occupancy parameters output by S4, and organizes the stability angle and centroid projection in the attitude stability parameters into attitude risk input items, and organizes the space occupancy parameters into space risk input items, forming attitude-space parameter pairs under the same timestamp; S52: Determine the degree of attitude instability based on the stability angle and the center of gravity projection, and form an attitude risk state marker; specifically, when the stability angle exceeds the preset stability angle threshold, the corresponding timestamp is marked as an attitude instability state; when the center of gravity projection exceeds the boundary of the support polygon formed by the support-related key points, the corresponding timestamp is marked as a center of gravity boundary crossing state; S53: Determine the degree of space restriction based on space occupancy parameters and form a space risk status marker. Specifically, when the space occupancy parameter is less than the preset safe distance threshold, the corresponding timestamp is marked as a space-restricted state. S54: Jointly determine the attitude risk state and the spatial risk state, and generate the attitude spatial risk coefficient based on the combination relationship of the two types of risk states at the same timestamp. S55: Receive the identification results of other explicit violations by personnel, classify them, and generate a list of violation triggers, wherein the list of violation triggers contains a trigger identifier corresponding to each type of violation; S56: The attitude space risk coefficient and the violation triggering flag are comprehensively evaluated to generate a dynamic behavior risk index.

[0012] Optionally, S54 specifically includes: S541: Receive the attitude risk state marker and spatial risk state marker under the same timestamp, and encode them into a binary risk state vector. ,in, This indicates an attitude risk status flag, with a value of 1 indicating that the attitude risk status has been triggered and a value of 0 indicating that it has not been triggered. This indicates a space risk status flag; a value of 1 indicates that the space-restricted status has been triggered, and a value of 0 indicates that it has not been triggered. S542: Based on the preset risk combination mapping rules, the binary risk state vector is combined and judged to generate the basic level value of the attitude space risk coefficient; specifically, when When corresponding to a high-risk level, when or When the corresponding medium risk level is reached, This corresponds to a low-risk level. S543: Map the base level value to a numerical attitude space risk coefficient. The mapping formula is as follows: ,in, The attitude space risk coefficient; and These are the preset low-risk and high-risk coefficient boundary values, and they satisfy... ; This represents the normalized value corresponding to the base level value obtained from S542, where low risk corresponds to 0, medium risk corresponds to 0.5, and high risk corresponds to 1. S544: Bind the attitude space risk coefficient to the corresponding timestamp to form a time-seriesd attitude space risk coefficient.

[0013] Optionally, S56 specifically includes: S561: Receive the attitude space risk coefficient output in step S54. And the list of violations triggered by S55, and align the two at the same timestamp to form a risk input pair for comprehensive assessment; S562: Based on the preset violation type weight table, assign corresponding violation weight values ​​to each trigger identifier in the violation trigger list, and summarize all violation weight values ​​under the same timestamp to obtain the violation risk level. ; S563: Incorporate attitude space risk factor Risk level of violations A fusion calculation is performed to generate a dynamic behavioral risk index, the calculation expression of which is as follows: ,in, It is a dynamic behavioral risk index; S564: Bind the dynamic behavior risk index to the corresponding timestamp to form a time-series dynamic behavior risk index output.

[0014] Optionally, S6 specifically includes: S61: Receive the static risk level generated by S3 and the dynamic behavioral risk index generated by S56, and convert the static risk level into the corresponding static risk coefficient; S62: Weight the static risk coefficient and the dynamic behavioral risk index to form a comprehensive risk coefficient at the current timestamp. ; S63: Incorporate risk coefficient The warning level is determined by comparing the data with the preset risk warning level thresholds within a given range, according to the following rules: when At that time, a Level 1 warning signal is generated; when At that time, a level-two warning signal is generated; when At that time, a three-level early warning signal is generated; among them, The preset comprehensive risk threshold parameter satisfies ; S64: The warning level and comprehensive risk coefficient under the corresponding timestamp are encapsulated together into a structured warning signal and sent to the monitoring terminal in real time through the monitoring data channel.

[0015] The beneficial effects of this invention are: This invention constructs an integrated AI image recognition process that integrates the three-dimensional posture information of workers, the spatial distribution of hazards, and the results of violation identification. For the first time, it jointly models posture stability parameters and space occupancy parameters to form a posture space risk coefficient, which is then integrated with the violation risk quantity to generate a dynamic behavior risk index. This significantly improves the real-time performance and accuracy of dynamic risk identification in confined space operation scenarios.

[0016] This invention transforms static risk levels into quantitative static risk coefficients, and integrates them with dynamic behavioral risk indices according to weights to generate a comprehensive risk coefficient. By comparing the coefficients with multi-level thresholds, it outputs graded early warning signals and realizes real-time monitoring and early warning push for different risk levels, effectively enhancing the closed-loop response capability and intelligence level of the confined space operation safety management system. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention 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 only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of a confined space operation risk monitoring and alarm method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the process for forming a dynamic behavioral risk index according to an embodiment of the present invention. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0020] like Figures 1-2 As shown, a method for risk monitoring and alarming in confined space operations based on AI image recognition includes the following steps: S1: Real-time acquisition of multi-angle video image streams of the confined space operation area, and preprocessing them to obtain a standardized image sequence; S1 specifically includes: S11: Set up no less than two video acquisition devices in the confined space operation area, facing the operation entrance channel and the core operation surface respectively, and assign a unique device identifier to each video acquisition device; synchronously acquire the real-time video stream output by each video acquisition device at a uniform sampling frame rate to form a multi-angle video image stream distinguished by device identifier; S12: Perform timestamp alignment processing on the multi-angle video image streams formed in S11, map the frame data of each video stream to a unified time axis, and perform frame sequence completion processing on missing frames to obtain a time-synchronized multi-angle frame sequence. S13: Perform image normalization preprocessing on the time-synchronized multi-angle frame sequence, including resolution unification, pixel intensity normalization and image denoising, and output a normalized frame sequence that meets the preset input specifications. S14: Perform viewpoint consistency processing on the standardized frame sequence. Based on the fixed installation pose parameters of each video acquisition device, the standardized frame sequence from different viewpoints is converted into an image representation in a unified coordinate system and arranged in a unified timeline order to form a standardized image sequence. The above steps ensure that the image data input to the subsequent AI image recognition engine is consistent in time dimension and imaging specifications by performing device-identified acquisition, timestamp alignment and unified standardization preprocessing on multi-angle video image streams, and converting images from different viewpoints into an image representation in a unified coordinate system. This improves the stability and comparability of the extraction of hazard source feature sets and personnel behavior feature sets.

[0021] S2: Perform parallel analysis on standardized image sequences using an AI image recognition engine to extract sets of hazard source features and sets of personnel behavior features from the images; S2 specifically includes: S21: Receive the standardized image sequence output by S1, divide the standardized image sequence into a frame-by-frame input queue according to a unified timeline, and bind each frame image to the corresponding timestamp and acquisition viewpoint identifier to form a structured input frame set for parallel analysis. S22: Input the structured input frame set into the AI ​​image recognition engine, perform synchronous inference processing on multi-view frames at the same timestamp, and obtain the candidate detection results of hazard sources and candidate detection results of personnel corresponding to each viewpoint; S23: Perform category confirmation and location parameter extraction on the candidate detection results of the hazard source, output the hazard source category label, the coordinates of the hazard source target box and the hazard source edge contour parameters, and assemble the hazard source category label and the hazard source location parameters into a hazard source feature set; S24: Perform human key point detection and human posture parameter extraction on the candidate detection results of personnel, output the three-dimensional coordinates of human key points, body orientation and personnel contour parameters of the workers, and assemble the three-dimensional coordinates of human key points, body orientation and personnel contour parameters into a set of personnel behavior features. S25: Align the hazard source feature set formed in S23 with the personnel behavior feature set formed in S24 by timestamp to obtain the hazard source feature set-personnel behavior feature set pairing result for each timestamp; the above steps perform synchronous reasoning on multi-view standardized images at the same timestamp and extract hazard source category and spatial positioning parameters, human body key point three-dimensional coordinates and posture contour parameters respectively, and then align and associate features by timestamp, so that the subsequent static risk level generation and posture space risk calculation have consistent temporal basis and complete feature input, thereby improving the stability and traceability of the risk assessment link.

[0022] S3: Based on the set of hazard source characteristics, match the preset hazard source database to generate the static risk level of the current environment; S3 specifically includes: S31: Receive the set of hazard source features output by S2, specifically including hazard source category labels, hazard source target box coordinates and hazard source edge contours, and organize each hazard source record into a structured hazard source entry; S32: Call the preset hazard source database, use the hazard source category label of each structured hazard source entry as the search key to perform a matching query, and obtain the set of basic risk parameters of the hazard source that corresponds one-to-one with the hazard source category label, including the basic severity coefficient and the category weight coefficient; S33: Calculate the hazard occupancy ratio based on the hazard edge contour parameters of each structured hazard entry, and obtain the hazard exposure coefficient accordingly; the calculation formula is as follows: ,in, For the first Exposure coefficient of each hazard source; For the first The pixel area of ​​the hazard source contour region is determined by the edge contour parameters of the hazard source. The pixel area of ​​a preset reference region in the standardized image sequence is defined, and the preset reference region is consistent with the input specifications of the standardized image sequence. S34: Based on the baseline severity coefficient and category weight coefficient obtained in S32, and combined with the exposure coefficient obtained in S33, calculate the static risk contribution value of each hazard. The calculation formula is as follows: ,in, For the first The static risk contribution value of each hazard source; In order to be with the first The category weight coefficient corresponding to each hazard source category label; In order to be with the first The base severity coefficient corresponding to each hazard category label; For the first Exposure coefficient of each hazard source; S35: Weighted summation of the static risk contribution values ​​of all hazard sources at the same time stamp to obtain the static risk score of the current environment. And map the static risk score to a static risk level; where the calculation expression for the static risk score is: ,in, Static risk scoring; The number of hazard source entries under the same timestamp; For the first The static risk contribution value of each hazard source.

[0023] The mapping of static risk scores to static risk levels is as follows: when At that time, the static risk level of the current environment is determined to be Level I; when At that time, the static risk level of the current environment is determined to be Level II; when At that time, the static risk level of the current environment was determined to be Level III; when At that time, the static risk level of the current environment was determined to be Level IV; among which, The preset static risk scoring level threshold is met. .

[0024] S4: Based on the set of personnel behavior features, calculate the posture stability parameters and space occupancy parameters of the current posture of the operator; the posture stability parameters include the stability angle and the center of gravity projection, and the space occupancy parameters are the real-time dynamic distance from the outline of the operator to the edge of the nearest obstacle or hazard source; S4 specifically includes: S41: Based on the three-dimensional coordinates of key human body points, select key points related to human body support to form a support plane, and calculate the projection position of the human body's center of gravity on the support plane to obtain the center of gravity projection parameters; wherein, the three-dimensional coordinates of the human body's center of gravity are obtained by weighted summation of the three-dimensional coordinates of the human body's key points according to preset weights, and the calculation formula is as follows: ,in, Three-dimensional coordinates of the human body's center of gravity; The number of key human body points used in the calculation; For the first Three-dimensional coordinates of key points on an individual's body; In order to be with the first The weight coefficients corresponding to key points of an individual's body, and satisfying the following conditions: ; S42: Based on the three-dimensional coordinates of the human body's center of gravity obtained in S41 and the normal vector of the supporting plane, calculate the stability angle of the human body's posture relative to the supporting plane. As one of the attitude stability parameters, its calculation formula is: ,in, For stability angle; The normal vector of the supporting plane; It is the attitude vector pointing from the body's center of gravity to the direction the body is facing; S43: The centroid projection parameters obtained in S41 and the stability angle obtained in S42 are combined to form the attitude stability parameters, which are used to characterize the overall stability of the operator's current attitude.

[0025] S4 also includes: S44: Based on the personnel contour parameters, calculate the Euclidean distance from each sampling point on the boundary of the worker's contour to the nearest hazard source edge or obstacle edge in the hazard source feature set, and take the minimum value as the space occupancy parameter. The calculation formula is as follows: ,in, This refers to the space usage parameter; The first on the personnel outline boundary Coordinates of each sampling point; The first one on the edge of the hazard source or obstacle The above steps quantify the projection of the human body's center of gravity, stability angle, and minimum distance between the human outline and the edge of the hazard source based on the three-dimensional coordinates of the human body's key points and the body orientation. This allows the stability of the worker's posture and the degree of spatial constraint to be expressed in the form of calculable parameters, providing a clear, continuous, and physically meaningful input basis for subsequent posture-space risk coupling analysis.

[0026] S5: Based on the attitude stability parameters and space occupancy parameters, conduct attitude space risk coupling analysis to generate attitude space risk coefficients; and combine the results of identifying other explicit violations by personnel to jointly form a dynamic behavior risk index; S5 specifically includes: S51: Receives the attitude stability parameters and space occupancy parameters output by S4, and organizes the stability angle and centroid projection in the attitude stability parameters into attitude risk input items, and organizes the space occupancy parameters into space risk input items, forming attitude-space parameter pairs under the same timestamp; S52: Determine the degree of attitude instability based on the stability angle and the center of gravity projection, and form an attitude risk state marker; specifically, when the stability angle exceeds the preset stability angle threshold, the corresponding timestamp is marked as an attitude instability state; when the center of gravity projection exceeds the boundary of the support polygon formed by the support-related key points, the corresponding timestamp is marked as a center of gravity boundary crossing state; S53: Determine the degree of space restriction based on space occupancy parameters and form a space risk status marker. Specifically, when the space occupancy parameter is less than the preset safe distance threshold, the corresponding timestamp is marked as a space-restricted state. S54: Jointly determine the attitude risk state and the spatial risk state, and generate the attitude spatial risk coefficient based on the combination relationship of the two types of risk states at the same timestamp, which is used to characterize the coupling risk level of attitude instability and spatial constraint. S55: Receive the identification results of other explicit violations by personnel, classify them, and generate a list of violation triggers. The list of violation triggers contains the trigger identifier corresponding to each type of violation. S56: The attitude space risk coefficient and the violation triggering flag are comprehensively evaluated to generate a dynamic behavior risk index. The above steps jointly determine the attitude risk state and the space risk state to generate the attitude space risk coefficient, and then comprehensively evaluate the risk coefficient and the explicit violation triggering flag to form a dynamic behavior risk index. This makes the dynamic risk assessment process simple and clear in structure, and at the same time provides a stable technical foundation for further limiting the coupling rules and fusion strategies.

[0027] S54 specifically includes: S541: Receive the attitude risk state marker and spatial risk state marker under the same timestamp, and encode them into a binary risk state vector. ,in, This indicates an attitude risk status flag, with a value of 1 indicating that the attitude risk status has been triggered and a value of 0 indicating that it has not been triggered. This indicates a space risk status flag; a value of 1 indicates that the space-restricted status has been triggered, and a value of 0 indicates that it has not been triggered. S542: Based on the preset risk combination mapping rules, the binary risk state vector is combined and judged to generate the basic level value of the attitude space risk coefficient; specifically, when When corresponding to a high-risk level, when or When the corresponding medium risk level is reached, This corresponds to a low-risk level. S543: Map the base level value to a numerical attitude space risk coefficient. The mapping formula is as follows: ,in, The attitude space risk coefficient; and These are the preset low-risk and high-risk coefficient boundary values, and they satisfy... ; This represents the normalized value corresponding to the base level value obtained from S542, where low risk corresponds to 0, medium risk corresponds to 0.5, and high risk corresponds to 1. S544: Bind the attitude space risk coefficient to the corresponding timestamp to form a time-series attitude space risk coefficient; the above steps combine and encode the attitude risk state and the spatial risk state at the same timestamp, and generate a numerical attitude space risk coefficient according to a clear combination mapping rule, so that the coupling risk of attitude instability and spatial constraint is quantified on a unified scale, thereby providing a stable, comparable risk input with clear judgment basis for the subsequent calculation of dynamic behavior risk index.

[0028] S56 specifically includes: S561: Receive the attitude space risk coefficient output in step S54. And the list of violations triggered by S55, and align the two at the same timestamp to form a risk input pair for comprehensive assessment; S562: Based on the preset violation type weight table, assign corresponding violation weight values ​​to each trigger identifier in the violation trigger list, and summarize all violation weight values ​​under the same timestamp to obtain the violation risk level. Its calculation expression is: ,in, Risk level for violations; The number of violation types; For the first Weighting coefficients corresponding to different types of violations; For the first The trigger flag for this type of violation is set to 1 when triggered and 0 when not triggered. S563: Incorporate attitude space risk factor Risk level of violations A fusion calculation is performed to generate a dynamic behavioral risk index, the calculation expression of which is as follows: ,in, It is a dynamic behavioral risk index; S564: Bind the dynamic behavior risk index with the corresponding timestamp to form a time-series dynamic behavior risk index output; the above steps use the posture space risk coefficient as the basic risk quantity and transform the violation trigger identifier into a calculable violation risk quantity, and then generate the dynamic behavior risk index according to a unified fusion rule, so that personnel posture instability, spatial constraints and explicit violations are included in the same quantitative framework for comprehensive evaluation, thereby improving the calculability, consistency and risk representation completeness of the dynamic behavior risk index generation process.

[0029] S6: The static risk level and the dynamic behavioral risk index are weighted and superimposed to obtain the comprehensive risk coefficient. When the comprehensive risk coefficient exceeds the preset threshold, the corresponding graded early warning signal is generated and sent to the monitoring terminal. S6 specifically includes: S61: Receive the static risk level generated by S3 and the dynamic behavioral risk index generated by S56, and convert the static risk level into the corresponding static risk coefficient. The static risk coefficient uses a fixed numerical form to represent the risk intensity of each level. The higher the level, the larger the corresponding coefficient value. S62: Weight the static risk coefficient and the dynamic behavioral risk index to form a comprehensive risk coefficient at the current timestamp. Its calculation expression is: ,in, To assess the overall risk factor, The static risk coefficient is obtained by mapping from the static risk level. It is a dynamic behavioral risk index. These are the static and dynamic risk weight coefficients, respectively, satisfying... and ; S63: Incorporate risk coefficient The warning level is determined by comparing the data with the preset risk warning level thresholds within a given range, according to the following rules: when At that time, a Level 1 warning signal is generated; when At that time, a level-two warning signal is generated; when At that time, a three-level early warning signal is generated; among them, The preset comprehensive risk threshold parameter satisfies ; S64: The warning level and comprehensive risk coefficient under the corresponding timestamp are jointly encapsulated into a structured warning signal and sent to the monitoring terminal in real time through the monitoring data channel, so that the supervisory personnel can make response decisions or trigger subsequent safety control actions. The above steps map the static risk level to a quantifiable coefficient, and perform weighted superposition with the dynamic behavioral risk index to generate a comprehensive risk coefficient. Then, combined with the set threshold level, a corresponding warning signal is generated and sent to the monitoring terminal. This realizes the joint judgment and real-time response to static and dynamic risks in the confined space operation environment, and improves the accuracy of overall risk monitoring and the timeliness of warning and handling.

[0030] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0031] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for risk monitoring and alarm in confined space operations based on AI image recognition, characterized in that, Includes the following steps: S1: Real-time acquisition of multi-angle video image streams of the confined space operation area, and preprocessing them to obtain a standardized image sequence; S2: Perform parallel analysis on standardized image sequences using an AI image recognition engine to extract sets of hazard source features and sets of personnel behavior features from the images; S3: Based on the set of hazard source characteristics, match the preset hazard source database to generate the static risk level of the current environment; S4: Based on the set of personnel behavior characteristics, calculate the posture stability parameters and space occupancy parameters of the current posture of the operator; S5: Based on the attitude stability parameters and space occupancy parameters, perform attitude space risk coupling analysis to generate attitude space risk coefficient; Combined with the results of identifying other explicit violations by personnel, a dynamic behavioral risk index is formed. S6: The static risk level and the dynamic behavioral risk index are weighted and superimposed to obtain a comprehensive risk coefficient. When the comprehensive risk coefficient exceeds the preset threshold, a corresponding graded early warning signal is generated and sent to the monitoring terminal.

2. The method for risk monitoring and alarm of confined space operations based on AI image recognition according to claim 1, characterized in that, S1 specifically includes: S11: Set up no less than two video acquisition devices in the confined space operation area, facing the operation entrance channel and the core operation surface respectively, and assign a unique device identifier to each video acquisition device; synchronously acquire the real-time video stream output by each video acquisition device at a uniform sampling frame rate to form a multi-angle video image stream distinguished by device identifier; S12: Perform timestamp alignment processing on the multi-angle video image streams formed in S11, map the frame data of each video stream to a unified time axis, and perform frame sequence completion processing on missing frames to obtain a time-synchronized multi-angle frame sequence. S13: Perform image normalization preprocessing on the time-synchronized multi-angle frame sequence, including resolution unification, pixel intensity normalization and image denoising, and output a normalized frame sequence that meets the preset input specifications. S14: Perform viewpoint consistency processing on the standardized frame sequence. Based on the fixed installation pose parameters of each video acquisition device, convert the standardized frame sequence under different viewpoints to the image representation under a unified coordinate system, and arrange them in a unified time axis order to form a standardized image sequence.

3. The method for risk monitoring and alarm of confined space operations based on AI image recognition according to claim 1, characterized in that, S2 specifically includes: S21: Receive the standardized image sequence output by S1, divide the standardized image sequence into a frame-by-frame input queue according to a unified timeline, and bind each frame image to the corresponding timestamp and acquisition viewpoint identifier to form a structured input frame set for parallel analysis. S22: Input the structured input frame set into the AI ​​image recognition engine, perform synchronous inference processing on multi-view frames at the same timestamp, and obtain the candidate detection results of hazard sources and candidate detection results of personnel corresponding to each viewpoint; S23: Perform category confirmation and location parameter extraction on the candidate detection results of the hazard source, output the hazard source category label, the coordinates of the hazard source target box and the hazard source edge contour parameters, and assemble the hazard source category label and the hazard source location parameters into a hazard source feature set; S24: Perform human key point detection and human posture parameter extraction on the candidate detection results of personnel, output the three-dimensional coordinates of human key points, body orientation and personnel contour parameters of the workers, and assemble the three-dimensional coordinates of human key points, body orientation and personnel contour parameters into a set of personnel behavior features. S25: Align the hazard source feature set formed in S23 with the personnel behavior feature set formed in S24 by timestamp to obtain the hazard source feature set-personnel behavior feature set pairing result for each timestamp.

4. The method for risk monitoring and alarm of confined space operations based on AI image recognition according to claim 1, characterized in that, S3 specifically includes: S31: Receive the set of hazard source features output by S2, specifically including hazard source category labels, hazard source target box coordinates and hazard source edge contours, and organize each hazard source record into a structured hazard source entry; S32: Call the preset hazard source database, use the hazard source category label of each structured hazard source entry as the search key to perform a matching query, and obtain the set of basic risk parameters of the hazard source that corresponds one-to-one with the hazard source category label, including the basic severity coefficient and the category weight coefficient; S33: Calculate the hazard occupancy ratio based on the hazard edge contour parameters of each structured hazard source entry, and obtain the hazard exposure coefficient accordingly; S34: Based on the baseline severity coefficient and category weight coefficient obtained in S32, and combined with the exposure coefficient obtained in S33, calculate the static risk contribution value of each hazard. ; S35: Weighted summation of the static risk contribution values ​​of all hazard sources at the same time stamp to obtain the static risk score of the current environment. And map the static risk score to the static risk level.

5. A method for monitoring and alarming risks in confined space operations based on AI image recognition according to claim 3, characterized in that, S4 specifically includes: S41: Based on the three-dimensional coordinates of key points of the human body, select key points related to human body support to form a support plane, and calculate the projection position of the human body's center of gravity on the support plane to obtain the center of gravity projection parameters. S42: Based on the three-dimensional coordinates of the human body's center of gravity obtained in S41 and the normal vector of the supporting plane, calculate the stability angle of the human body's posture relative to the supporting plane. As one of the attitude stability parameters, its calculation formula is: ,in, For stability angle; The normal vector of the supporting plane; It is the attitude vector pointing from the body's center of gravity to the direction the body is facing; S43: Combine the centroid projection parameters obtained in S41 with the stability angle obtained in S42 to form the attitude stability parameters.

6. A method for risk monitoring and alarm of confined space operations based on AI image recognition according to claim 3, characterized in that, S4 further includes: S44: Based on the personnel contour parameters, calculate the Euclidean distance from each sampling point on the boundary of the worker's contour to the nearest hazard source edge or obstacle edge in the hazard source feature set, and take the minimum value as the space occupancy parameter. The calculation formula is as follows: ,in, This refers to the space usage parameter; The first on the personnel outline boundary Coordinates of each sampling point; The first one on the edge of the hazard source or obstacle The coordinates of the boundary points.

7. A method for risk monitoring and alarm of confined space operations based on AI image recognition according to claim 1, characterized in that, S5 specifically includes: S51: Receives the attitude stability parameters and space occupancy parameters output by S4, and organizes the stability angle and centroid projection in the attitude stability parameters into attitude risk input items, and organizes the space occupancy parameters into space risk input items, forming attitude-space parameter pairs under the same timestamp; S52: Determine the degree of attitude instability based on the stability angle and the center of gravity projection, and form an attitude risk state marker; specifically, when the stability angle exceeds the preset stability angle threshold, the corresponding timestamp is marked as an attitude instability state; when the center of gravity projection exceeds the boundary of the support polygon formed by the support-related key points, the corresponding timestamp is marked as a center of gravity boundary crossing state. S53: Determine the degree of space restriction based on space occupancy parameters and form a space risk status marker. Specifically, when the space occupancy parameter is less than the preset safe distance threshold, the corresponding timestamp is marked as a space-restricted state. S54: Jointly determine the attitude risk state and the spatial risk state, and generate the attitude spatial risk coefficient based on the combination relationship of the two types of risk states at the same timestamp. S55: Receive the identification results of other explicit violations by personnel, classify them, and generate a list of violation triggers, wherein the list of violation triggers contains a trigger identifier corresponding to each type of violation; S56: The attitude space risk coefficient and the violation triggering flag are comprehensively evaluated to generate a dynamic behavior risk index.

8. A method for risk monitoring and alarm of confined space operations based on AI image recognition according to claim 1, characterized in that, Specifically, S54 includes: S541: Receive the attitude risk state marker and spatial risk state marker under the same timestamp, and encode them into a binary risk state vector. ,in, This indicates an attitude risk status flag, with a value of 1 indicating that the attitude risk status has been triggered and a value of 0 indicating that it has not been triggered. This indicates a space risk status flag; a value of 1 indicates that the space-restricted status has been triggered, and a value of 0 indicates that it has not been triggered. S542: Based on the preset risk combination mapping rules, the binary risk state vector is combined and judged to generate the basic level value of the attitude space risk coefficient; specifically, when When corresponding to a high-risk level, when or When the corresponding medium risk level is reached, This corresponds to a low-risk level. S543: Map the base level value to a numerical attitude space risk coefficient. The mapping formula is as follows: ,in, The attitude space risk coefficient; and These are the preset low-risk and high-risk coefficient boundary values, and they satisfy... ; This represents the normalized value corresponding to the base level value obtained from S542, where low risk corresponds to 0, medium risk corresponds to 0.5, and high risk corresponds to 1. S544: Bind the attitude space risk coefficient to the corresponding timestamp to form a time-seriesd attitude space risk coefficient.

9. A method for risk monitoring and alarm of confined space operations based on AI image recognition according to claim 8, characterized in that, Specifically, S56 includes: S561: Receive the attitude space risk coefficient output in step S54. And the list of violations triggered by S55, and align the two at the same timestamp to form a risk input pair for comprehensive assessment; S562: Based on the preset violation type weight table, assign corresponding violation weight values ​​to each trigger identifier in the violation trigger list, and summarize all violation weight values ​​under the same timestamp to obtain the violation risk level. ; S563: Incorporate attitude space risk factor Risk level of violations A fusion calculation is performed to generate a dynamic behavioral risk index, the calculation expression of which is as follows: ,in, It is a dynamic behavioral risk index; S564: Bind the dynamic behavior risk index to the corresponding timestamp to form a time-series dynamic behavior risk index output.

10. A method for risk monitoring and alarm of confined space operations based on AI image recognition according to claim 9, characterized in that, S6 specifically includes: S61: Receive the static risk level generated by S3 and the dynamic behavioral risk index generated by S56, and convert the static risk level into the corresponding static risk coefficient; S62: Weight the static risk coefficient and the dynamic behavioral risk index to form a comprehensive risk coefficient at the current timestamp. ; S63: Incorporate risk coefficient The warning level is determined by comparing the data with the preset risk warning level thresholds within a given range, according to the following rules: when At that time, a Level 1 warning signal is generated; when At that time, a level-two warning signal is generated; when At that time, a three-level early warning signal is generated; among them, The preset comprehensive risk threshold parameter satisfies ; S64: The warning level and comprehensive risk coefficient under the corresponding timestamp are encapsulated together into a structured warning signal and sent to the monitoring terminal in real time through the monitoring data channel.

Images