Construction safety monitoring method and system based on CV large model

Abstract

The invention discloses a construction safety monitoring method based on a CV large model, and relates to the technical field of construction monitoring, and the method comprises the steps: obtaining construction video data, and employing the CV large model to recognize the behavior characteristics of construction personnel; carrying out vector comparison on the behavior characteristics and a standard behavior characteristic library, and identifying high-risk construction monitoring personnel; establishing a multi-behavior joint risk analysis model to determine a peak monitoring period; carrying out clustering analysis on the peak monitoring period to obtain a maximum cluster, and judging the difference significance of the maximum cluster; if the difference is significant, establishing a spatial analysis model to identify an aggregation effect, evaluating a risk behavior self-correction hysteresis degree, positioning a high-risk area, and realizing collaborative safety monitoring; the construction safety monitoring system based on the CV large model comprises a feature recognition module, a behavior comparison module, a peak value extraction module, a clustering analysis module and a hysteresis evaluation module, and is used for realizing the method. The construction safety monitoring level can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of construction monitoring, and in particular to a construction safety monitoring method and system based on a CV large model. Background Art

[0002] With the development of computer vision (CV) technology, its application in construction safety monitoring has gradually attracted attention. In the field of construction safety monitoring, the existing technology has many shortcomings. Compared with the construction safety monitoring method based on CV large model, there is a gap in risk warning, resource utilization, risk analysis and collaborative management.

[0003] Existing technologies have limited capabilities in identifying construction workers’ behavioral characteristics. They cannot automatically extract construction workers’ attention behavior characteristics, spatial location characteristics, and safety equipment characteristics. If construction safety monitoring technology based on the CV large model is adopted, it can conduct all-round, real-time monitoring and analysis of construction workers’ behaviors, improve the timeliness and accuracy of risk behavior identification, and effectively reduce the probability of accidents.

[0004] The existing technology is disconnected from the safety management in the time and space dimensions, and cannot achieve coordinated monitoring. In the time dimension, the existing technology has not established an effective multi-behavior joint risk analysis model, and cannot comprehensively process the various behavior data of construction personnel to determine the peak monitoring period where risky behaviors are concentrated. By calculating the behavior inaccuracy rate and building a multi-behavior joint risk analysis model, the peak monitoring period is determined, and the accuracy of safety management in the time dimension is improved; In the spatial dimension, the existing technology does not have the ability to identify the spatial analysis model and the agglomeration effect of personnel positions, and cannot identify the agglomeration effect of the construction monitoring personnel positions, making it difficult to determine the high-risk areas. If the spatial analysis model established by it is adopted, the personnel position information can be collected and analyzed in real time, and the agglomeration effect and high-risk areas can be identified, which will facilitate the safety management personnel to reasonably allocate resources and improve the safety management efficiency in the spatial dimension. Summary of the invention

[0005] The object of the present invention is to provide a construction safety monitoring method and system based on a CV large model to solve at least one of the above-mentioned prior art problems.

[0006] In a first aspect, the present invention provides a construction safety monitoring method based on a CV large model, comprising the following steps: Step 1: Obtain video data of deep foundation pit excavation and support operations, and identify the behavioral characteristics of construction workers through the CV large model; Step 2: Perform vector comparison analysis on the behavior characteristics of construction personnel and the pre-built standard behavior characteristic library, identify risky behaviors and mark construction personnel with the highest behavior risk level as construction monitoring personnel; Step 3: Obtain the behavioral characteristics of self-correction of construction monitoring personnel during the monitoring period, calculate the behavioral inaccuracy rates of different construction monitoring personnel, and establish a multi-behavior joint risk analysis model to extract the peak monitoring period; Step 4: Perform cluster analysis on the peak monitoring periods of all construction monitoring personnel to obtain the largest cluster, and determine whether the difference of the largest cluster is significant; Step 5. If the difference is significant, it is used to establish a spatial analysis model to identify the clustering effect of the locations of construction monitoring personnel and to evaluate the degree of self-correction hysteresis of the risk behavior of construction monitoring personnel.

[0007] In a second aspect, the present invention provides a construction safety monitoring system based on a CV large model, comprising the following modules: Feature recognition module: used to obtain video data of deep foundation pit excavation and support operations, and identify the behavioral characteristics of construction personnel through the CV large model; Behavior comparison module: used to perform vector comparison analysis on the behavior characteristics of construction workers and the pre-built standard behavior characteristic library, identify risky behaviors and mark construction workers with the highest behavior risk level as construction monitoring personnel; Peak extraction module: used to obtain the behavioral characteristics of self-correction of construction monitoring personnel within the monitoring period, calculate the behavioral inaccuracy rates of different construction monitoring personnel, and establish a multi-behavior joint risk analysis model to extract the peak monitoring period; Cluster analysis module: used to perform cluster analysis on the peak monitoring period of all construction monitoring personnel, obtain the largest cluster, and determine whether the difference of the largest cluster is significant; Hysteresis assessment module: If the difference is significant, it is used to establish a spatial analysis model to identify the clustering effect of the locations of construction monitoring personnel and to assess the degree of self-correction hysteresis of the risk behavior of construction monitoring personnel.

[0008] Beneficial effects of the present invention: 1. Utilize the image recognition capability of the CV large model to extract the attention behavior characteristics, spatial location characteristics, and safety equipment characteristics of construction workers, so as to detect the behavior characteristics of construction workers in advance and prevent construction safety accidents; perform vector comparison between the behavior characteristics of construction workers and the standard behavior feature library, and identify risky behaviors by calculating the Manhattan distance. Divide the behavior risk levels and mark the construction monitoring personnel to focus on high-risk individuals and reduce the possibility of accidents. At the same time, allocate safety monitoring resources in a targeted manner according to the risk level.

[0009] 2. By calculating the behavior inaccuracy rate, a multi-behavior joint risk analysis model is established to determine the peak monitoring period. The time period with the most concentrated risk behaviors during the construction process is located to provide a basis for timely intervention by safety management personnel. By analyzing the KL divergence within the peak monitoring period, the concentration of risk behaviors is evaluated, and management strategies are adjusted in a timely manner to increase control over high-risk periods and areas.

[0010] 3. Perform cluster analysis on the peak monitoring period of all construction monitoring personnel to obtain the largest cluster. Integrate construction personnel with similar risk characteristics, grasp the risk behavior pattern as a whole, and identify the key groups and time periods where risk behaviors occur. Determine whether the difference of the largest cluster is significant, evaluate the reliability and effectiveness of the clustering results, and provide a scientific basis for subsequent safety decisions. If the difference is significant, a special management strategy can be formulated for this group.

[0011] 4. Used to establish a spatial analysis model to identify the clustering effect of the location of construction monitoring personnel, and further determine high-risk areas from a spatial dimension. By calculating the regional hysteresis index, the degree of self-correction hysteresis of the risk behavior of construction monitoring personnel is evaluated. Based on the evaluation results, decisions are made to increase the frequency of supervision, adjust the construction process, or strengthen warning measures, so as to accurately invest limited resources in high-risk areas, improve the overall safety management efficiency, and ensure construction safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0013] Figure 1 is a flow chart of the construction monitoring behavior analysis of the present invention; Figure 2 is a flowchart of construction safety time and space analysis provided by Embodiment 2 of the present invention; Figure 3 This is a module diagram of a construction safety monitoring system based on a CV large model provided in Example 3 of the present invention. DETAILED DESCRIPTION

[0014] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0015] Embodiment 1 Deep foundation pit operations involve soil excavation and support structure installation, which are prone to accidents such as collapse, object impact, and mechanical injury. At the same time, the distance between personnel and the edge of the foundation pit, the radius of mechanical operation, etc. can be directly quantified through visual positioning. Monitoring and analyzing the construction scenes of deep foundation pit excavation and support operations is conducive to focusing on the core pain points of construction safety. like Figure 1 As shown, a construction safety monitoring method based on a CV large model provided by an embodiment of the present invention includes the following steps: Step 1: Obtain video data of deep foundation pit excavation and support operations, and identify the behavioral characteristics of construction workers through the CV large model; In some embodiments, multiple video acquisition devices are installed around the foundation pit and on the top of the tower crane to obtain construction video data in the deep foundation pit excavation and support operation scene; Identify construction workers in construction video data through the CV large model and extract their behavioral characteristics; The behavioral characteristics include: attention behavior characteristics, spatial location characteristics, and safety equipment characteristics; It should be further explained that the attention behavior characteristics reflect the construction workers' alertness to the risks in the working environment; the CV large model is used to analyze the construction workers' line of sight direction and calculate the angle between the line of sight direction and the foundation pit normal line, so as to timely detect the distraction state of the construction workers, prevent accidents such as soil collapse and mechanical collision caused by delayed reaction, and ensure the safety of construction operations; The spatial position feature quantifies the relative position relationship between construction personnel and dangerous areas such as the edge of the foundation pit and the mechanical operation area through the CV large model; the distance to the edge of the foundation pit is calculated based on the monocular vision ranging formula, and the trajectory of the robot arm is reconstructed through multi-camera vision to identify illegal entry into the construction warning area or mechanical blind spot, triggering real-time alarms or emergency stops of equipment, reducing the probability of accidents such as falling and squeezing, and optimizing the safety management and control of the construction area; Safety equipment features use the CV large model to verify the wearing standard of protective equipment such as helmets and seat belts; use the Faster R-CNN algorithm to detect the fit of helmets and analyze the spatial relationship between the seat belt buckle and the human skeleton; wearing standardized seat belts and helmets can directly reduce the risk of injuries such as head impact and high-altitude falls, and at the same time, as a visual indicator of safety management compliance, promote the implementation of on-site safety regulations; It should be noted that the role of extracting the behavioral characteristics of construction workers is: Function 1: Realize risk warning. Through the analysis of attention behavior characteristics, the alertness of construction personnel can be monitored in real time to prevent delayed response due to distraction. The spatial location characteristics can quantify the distance to the dangerous area and trigger real-time alarm or emergency stop of equipment. Function 2: Provide a basis for data modeling and extract the behavioral characteristics of construction workers to provide basic data for subsequent vector comparison and analysis.

[0016] Step 2: Perform vector comparison analysis on the behavior characteristics of construction personnel and the pre-built standard behavior characteristic library, identify risky behaviors and mark construction personnel with the highest behavior risk level as construction monitoring personnel; In some embodiments, standard behavior characteristics in deep foundation pit excavation and support construction are obtained to build a standard behavior characteristic library; For example, standard behavior characteristics include: the angle between the construction worker's line of sight and the foundation pit normal is less than 30 degrees, the distance between the construction worker and the edge of the foundation pit is greater than 1.5 meters, the operation radius between the construction worker and the operating equipment is greater than 3 meters, the fit of the helmet is greater than 0.85, and the safety belt buckle is located within ±10cm of the waist midline; Extract the standard vector based on the standard feature behavior feature library, obtain the characteristic behavior of construction workers in real time through the CV large model, and construct the behavior vector; It should be noted that the normative vectors are constructed based on the minimum standards of the standard behavioral feature library; Exemplarily, the normative vector and the behavior vector are constructed as follows: the minimum value of the normative sight line angle: when the sight line angle is within [25°, 30°], the corresponding element of the vector is set to 0, otherwise it is set to 1; Minimum value of the distance from the foundation pit edge: if the edge distance is between [1.5m, 2.0m], the corresponding element of the vector is set to 0, otherwise it is set to 1; Minimum value of mechanical operation radius: If the operation radius is within [3.0m, 3.5m], the corresponding element of the vector is set to 0, otherwise it is set to 1; Minimum value of helmet fit: If the helmet fit range is in [0.85, 0.90], the corresponding element of the vector is set to 0, otherwise it is set to 1; The minimum value of the seat belt buckle position: if the seat belt buckle is within ±10cm of the waist centerline, the corresponding element of the vector is set to 0, otherwise it is set to 1; According to the minimum requirements of the standard behavior feature library, a 5-dimensional standard vector is constructed: y=[0,0,0,0,0]; The elements of the 5-dimensional normative vector correspond to the minimum value of the distance to the foundation pit edge, the minimum value of the mechanical operation radius, the minimum value of the fit of the helmet, and the minimum value of the safety belt buckle position; Based on the norm vector and the behavior vector, the Manhattan distance between the norm vector and the behavior vector is calculated; If the Manhattan distance of the construction worker exceeds the preset standard distance range, the construction worker's behavior characteristics are marked as risky behavior; Obtain the time when the Manhattan distance corresponding to the construction personnel exceeds the preset standard distance range, as well as the total monitoring time; Calculate the deviation ratio of the time when the Manhattan distance exceeds the preset standard distance range to the total monitoring time to obtain the time deviation ratio; Based on the different time deviation ratios of construction workers, the construction workers' behavior characteristics are divided into multiple behavior risk levels; Based on different risk levels, the system sends warning notifications of different risk levels to construction personnel; If the construction worker's behavioral characteristics are at the highest behavioral risk level, the construction worker will be marked as a construction monitoring personnel; It should be noted that the construction behavior characteristics contain 5-dimensional vectors, and Manhattan distance can effectively measure the similarity of vector directions in high-dimensional space and reduce the impact of dimensional differences on behavior recognition and judgment; The role of construction monitoring personnel is to: Function 1: Focus on high-risk individuals in construction safety and reduce the possibility of accidents. Marking construction monitoring personnel can focus on high-risk individuals and achieve safe construction risk prevention and control. Taking deep foundation pit operations as an example, monitoring personnel's illegal entry into dangerous areas and failure to wear safety equipment correctly can easily cause collapse, fall and other accidents. Focusing on marking and monitoring can timely detect and stop dangerous behaviors and reduce the possibility of accidents.

[0017] Function 2: Optimize the allocation of safety monitoring resources. At the foundation pit construction site, safety management resources are limited. Marking construction monitoring personnel helps to reasonably allocate these safety monitoring resources. By limiting the management focus to construction monitoring personnel, safety officers can conduct more targeted supervision and guidance, thereby improving safety management efficiency; Function three: Provide data support for subsequent risk analysis. By obtaining the behavioral characteristics of self-correction of construction monitoring personnel during the monitoring period, calculating the behavioral inaccuracy rate, and establishing a multi-behavior joint risk analysis model to extract the peak monitoring period, and then performing cluster analysis on the peak monitoring period of all construction monitoring personnel, etc., it can deeply explore the potential risks in the construction process.

[0018] The technical solution of this embodiment is: obtain the video data of deep foundation pit excavation and support operation, identify the behavioral characteristics of construction personnel through the CV large model; conduct vector comparison analysis between the behavioral characteristics of construction personnel and the pre-built standard behavioral characteristic library, identify risky behaviors and mark the construction personnel with the highest behavioral risk level as construction monitoring personnel; divide the behavioral risk level and mark the construction monitoring personnel to focus on high-risk individuals and reduce the possibility of accidents. At the same time, safety monitoring resources are allocated in a targeted manner according to the risk level.

[0019] Embodiment 2 like Figure 2As shown, a construction safety monitoring method based on a CV large model also includes the following steps: Step 3: Obtain the behavioral characteristics of self-correction of construction monitoring personnel during the monitoring period, calculate the behavioral inaccuracy rates of different construction monitoring personnel, and establish a multi-behavior joint risk analysis model to extract the peak monitoring period; During the monitoring period, the CV large model is used to obtain the behavioral characteristics of the construction monitoring personnel's self-correction of risky behaviors; It should be noted that the self-correction of risky behaviors by construction monitoring personnel means that after receiving the early warning notice, the construction monitoring personnel self-correct their own risky behaviors so that their behavior characteristics are no longer risky behaviors; If the monitoring personnel do not self-correct the risky behavior within the monitoring period, or fail to complete the self-correction of the risky behavior within the monitoring period; Obtain the number of times that construction monitoring personnel fail to complete self-correction of risky behaviors and fail to correct risky behaviors during the monitoring period; Based on the number of times that the construction monitoring personnel failed to complete the self-correction of risky behaviors, the frequency of the failure of the construction monitoring personnel to complete the self-correction of risky behaviors during the monitoring period was calculated to obtain the behavior inaccuracy rate; By formula: Construct a multi-behavior joint risk analysis model; in Obtain the KL divergence of different behavioral characteristics of construction monitoring personnel during the monitoring period, where P(x1, x2, x3) is the actual joint distribution of behavioral inaccuracy rate, and x1, x2, and x3 represent attention behavior characteristics, spatial position characteristics, and safety equipment characteristics, respectively; Q(x1), Q(x2), and Q(x3) are the canonical joint distributions of different characteristic behaviors, which are characterized by the behavioral inaccuracy rate threshold of the risk behavior of construction monitoring personnel; It should be noted that the KL divergence (Kullback-Leibler Divergence) is used to quantify the difference between the actual behavior distribution and the standard behavior distribution of construction monitoring personnel, reflecting the concentration of risk behaviors during the monitoring period. Q (x1), Q (x2), and Q (x3) are set by technicians in this field based on experience; Calculate the KL divergence of each monitoring period and construct a period divergence sequence; Perform mutation analysis on the KL divergence of the periodic divergence sequence to identify the peak period of the KL divergence; The mutation analysis calculates the standard deviation of the periodic divergence sequence and obtains the monitoring period with KL divergence higher than 3 times the standard deviation as the peak period; Get all the peak periods in the periodic divergence sequence. If the peak periods are adjacent in the time dimension, merge them. Obtain the peak period and merge it, and then find the peak period with the largest time dimension as the peak monitoring period. It should be noted that the purpose of determining the peak monitoring period is: Function 1. Locate high-risk periods: By establishing a multi-behavior joint risk analysis model to calculate the KL divergence, and performing mutation analysis on the periodic divergence sequence to determine the peak monitoring period, it is possible to accurately find the time period with the most concentrated risk behaviors during the construction process. For example, in deep foundation pit construction, different construction stages and changes in the working environment may cause fluctuations in risk behaviors. The peak monitoring period can accurately lock in specific periods with high risks, such as concentrated operations of large machinery and cross-construction of multiple types of work, to help safety managers pay attention and intervene in time to prevent accidents; Function 2: Provide effective data for cluster analysis: extract the cycle duration, peak intensity and other characteristics of the peak monitoring cycle, and then classify the construction monitoring personnel with similar risk characteristics into one category through the clustering algorithm, which is helpful to discover the risk behavior patterns of different groups of construction personnel, deeply understand the distribution law of risk behavior, and provide a basis for formulating targeted safety management measures; Function 3: Evaluate the concentration of risky behaviors: The higher the KL divergence value, the greater the difference between actual behavior and normative behavior, and the more concentrated the risky behavior. By determining the peak monitoring period and analyzing the KL divergence therein, security managers can intuitively understand the severity of the risk, adjust management strategies in a timely manner, and increase control over high-risk periods and areas.

[0020] Step 4: Perform cluster analysis on the peak monitoring periods of all construction monitoring personnel to obtain the largest cluster, and determine whether the difference of the largest cluster is significant; Obtain the peak monitoring period of all construction monitoring personnel and extract the periodic characteristics of the peak monitoring period; Perform Min-Max normalization on the periodic features and construct a periodic feature group; The cycle characteristics include: cycle duration, i.e., the duration of the peak monitoring cycle; peak intensity, i.e., the peak value of the behavior mismatch rate within the peak monitoring cycle; Based on the period feature groups of all construction monitoring personnel, the construction monitoring personnel with similar peak monitoring periods are clustered by clustering algorithm to obtain multiple clusters. By formula: Calculate the silhouette coefficient s(i) of each sample in the largest cluster, where i represents the sample number of the largest cluster, that is, the number of the construction monitoring personnel; a(i) represents the average distance from sample i to all other samples in the largest cluster, which is used to characterize the compactness of the sample within the cluster; b(i) represents the minimum average distance from sample i to all other clusters, which is used to characterize the inter-cluster separation of samples. All other clusters refer to the largest cluster excluding the largest cluster from different clusters. It should be noted that a(i) is calculated by calculating the average value of the Euclidean distance between the periodic characteristics of construction monitoring personnel i and other construction monitoring personnel in the largest cluster; b(i) is calculated by calculating the average Manhattan distance between the cycle characteristics of construction monitor i and other construction monitors in all other clusters; Calculate the mean of the silhouette coefficients of all samples in the largest cluster, calculate the proximity between the mean and the preset upper limit of the silhouette coefficient, and obtain the silhouette proximity value; Among them, the upper limit of the silhouette coefficient is 1; Compare the preset proximity range values ​​of the contour proximity values. If the contour proximity values ​​are within the preset proximity range values, it is considered that the difference of the largest cluster is significant. If the contour proximity value is not within the preset proximity range, it is considered that the difference between the largest clusters is not significant; It should be noted that calculating the largest cluster can integrate construction monitoring personnel with similar risk characteristics; by clustering the peak monitoring cycles of all construction monitoring personnel, similar cycle characteristics are classified into one category to form multiple clusters, among which the largest cluster contains the largest number of personnel groups with similar risk characteristics; it is helpful to grasp the risk behavior patterns of construction personnel as a whole and identify the key groups and time periods where risk behaviors occur in a concentrated manner; For example, in deep foundation pit construction, it may be found that construction workers in a certain cluster generally have similar risk behaviors such as inattention and illegal operations at a specific construction stage, such as when earth excavation is nearing the end, which provides a clear focus direction for safety management; Judging whether the difference of the largest cluster is significant can evaluate the reliability and effectiveness of the clustering results and provide a basis for subsequent safety management decisions. If the difference is significant, it means that the risk characteristics of the personnel in the largest cluster are obviously consistent and unique, and the clustering result is valid. Safety management can formulate special management strategies for this group, such as strengthening the training of the personnel in the cluster and increasing the frequency of supervision during specific periods of time.

[0021] Step 5: If the difference is significant, it is used to establish a spatial analysis model to identify the clustering effect of the construction monitoring personnel's location and evaluate the degree of self-correction hysteresis of the risk behavior of the construction monitoring personnel; Obtain the spatial coordinates of the construction monitoring personnel in the deep foundation pit construction area, and calculate the distance between any two construction monitoring personnel in the deep foundation pit based on the spatial coordinates; The Gaussian kernel function is used to determine the distance attenuation weights between any two construction monitoring personnel and establish a spatial weight matrix. Based on the spatial weight matrix, a spatial analysis model is constructed through the global Moran index to identify the clustering effect of the construction monitoring personnel's location; It should be noted that the value range of the global Moran index is [-1,1]. If the global Moran index is greater than 0, it indicates a positive spatial correlation, that is, there is a clustering effect in the location of construction monitoring personnel; The role of judging whether there is a clustering effect in the location of construction monitoring personnel is: Function 1: Locate high-risk areas: After determining the peak monitoring period of risk concentration in the time dimension, the clustering effect further determines the construction safety monitoring risk from the spatial dimension. When a clustering effect is found in the location of construction monitoring personnel, it indicates that risky behaviors are more likely to occur in a concentrated manner in a specific spatial area of ​​the construction site.

[0022] For example, in deep foundation pit construction, it is found that construction monitoring personnel gather in certain areas. These areas may have increased risk behaviors, such as illegal operations and insufficient safety distances, due to factors such as narrow space and frequent cross-construction operations. By identifying this spatial aggregation effect, safety managers can locate high-risk areas and take targeted measures in a timely manner, such as setting up warning signs and strengthening on-site supervision, to effectively prevent accidents. Function 2: Achieve collaborative safety management: Considering the risk concentration in the time dimension and the agglomeration effect in the space dimension, the safety management of the construction site can be more collaborative and efficient. In the agglomeration area, the behaviors of construction workers affect each other. Through the analysis of the agglomeration effect, safety managers can organize construction workers in the agglomeration area to carry out collaborative safety management activities. Conduct collective safety training to improve the overall safety awareness of construction workers, enhance the safety management level of the construction site, and ensure the smooth progress of construction; If the global Moran index is less than 0, it indicates negative spatial correlation, that is, there is a discrete effect in the location of construction monitoring personnel; The global Moran index is equal to 0, indicating that the locations of construction monitoring personnel are randomly distributed; If there is a clustering effect in the location of the construction monitoring personnel, the deep foundation pit construction area where the construction monitoring personnel are located is merged according to the distance attenuation weight between any two construction monitoring personnel to obtain the construction monitoring area; It should be noted that by setting a distance attenuation weight threshold, if the distance attenuation weights of any two construction monitoring personnel are higher than the distance attenuation weight threshold, the deep foundation pit construction areas where the construction monitoring personnel are located will be merged; wherein the distance attenuation weight threshold is set by professional and technical personnel in this field based on experience; Obtain the difference between the self-correction time and the monitoring time of the dangerous behavior of the construction monitoring personnel in the construction monitoring area after the system sends warning notifications of different risk levels to the construction personnel, and obtain the behavior correction delay; Obtain the average behavior correction delay of construction monitoring personnel in the construction monitoring area and the average behavior correction delay of all construction monitoring areas; The regional hysteresis index is obtained by calculating the ratio of the average behavior correction delay of construction monitoring personnel in the construction monitoring area to the average behavior correction delay of all construction monitoring areas; Based on the regional hysteresis index as a measure to evaluate the degree of self-correction hysteresis of risk behaviors of construction monitoring personnel; It should be noted that by calculating the regional hysteresis index, we can intuitively understand how fast personnel in different construction monitoring areas correct risky behaviors. For example, in deep foundation pit construction, if the regional hysteresis index is high, it means that personnel in the area are slow to correct themselves and the safety hazards last for a long time. Based on this, the safety management department can decide to increase the frequency of supervision in the area, adjust the construction process or strengthen warning measures, accurately invest limited resources in high-risk areas, improve overall safety management efficiency, and reduce the probability of accidents.

[0023] The technical solution of this embodiment is: obtaining the behavioral characteristics of self-correction of construction monitoring personnel within the monitoring period, calculating the behavioral inaccuracy rates of different construction monitoring personnel, and establishing a multi-behavior joint risk analysis model to extract the peak monitoring period; performing cluster analysis on the peak monitoring periods of all construction monitoring personnel to obtain the largest cluster, and judging whether the difference of the largest cluster is significant; if the difference is significant, using it to establish a spatial analysis model to identify the clustering effect of the locations of construction monitoring personnel, and assessing the degree of self-correction hysteresis of the risk behaviors of construction monitoring personnel; based on the assessment results, making decisions to increase the frequency of supervision, adjust the construction process, or strengthen warning measures, accurately invest limited resources in high-risk areas, improve overall safety management efficiency, and ensure construction safety.

[0024] Embodiment 3 like Figure 3 As shown, a construction safety monitoring system based on a CV large model includes the following modules: Feature recognition module: used to obtain video data of deep foundation pit excavation and support operations, and identify the behavioral characteristics of construction personnel through the CV large model; Identify construction workers in construction video data through the CV large model and extract their behavioral characteristics; The behavioral characteristics include: attention behavior characteristics, spatial location characteristics, and safety equipment characteristics; Behavior comparison module: used to perform vector comparison analysis on the behavior characteristics of construction workers and the pre-built standard behavior characteristic library, identify risky behaviors and mark construction workers with the highest behavior risk level as construction monitoring personnel; Obtain standard behavior characteristics in deep foundation pit excavation and support construction, and build a standard behavior characteristic library; Extract the standard vector based on the standard feature behavior feature library, obtain the characteristic behavior of construction workers in real time through the CV large model, and construct the behavior vector; Based on the norm vector and the behavior vector, the Manhattan distance between the norm vector and the behavior vector is calculated; If the Manhattan distance of the construction worker exceeds the preset standard distance range, the construction worker's behavior characteristics are marked as risky behavior; Obtain the time when the Manhattan distance corresponding to the construction personnel exceeds the preset standard distance range, as well as the total monitoring time; Calculate the deviation ratio of the time when the Manhattan distance exceeds the preset standard distance range to the total monitoring time to obtain the time deviation ratio; Based on the different time deviation ratios of construction workers, the construction workers' behavior characteristics are divided into multiple behavior risk levels; Based on different risk levels, the system sends warning notifications of different risk levels to construction personnel; If the construction worker's behavioral characteristics are at the highest behavioral risk level, the construction worker will be marked as a construction monitoring personnel; Peak extraction module: used to obtain the behavioral characteristics of self-correction of construction monitoring personnel within the monitoring period, calculate the behavioral inaccuracy rates of different construction monitoring personnel, and establish a multi-behavior joint risk analysis model to extract the peak monitoring period; During the monitoring period, the CV large model is used to obtain the behavioral characteristics of the construction monitoring personnel's self-correction of risky behaviors; If the monitoring personnel do not self-correct the risky behavior within the monitoring period, or fail to complete the self-correction of the risky behavior within the monitoring period; Obtain the number of times that construction monitoring personnel fail to complete self-correction of risky behaviors and fail to correct risky behaviors during the monitoring period; Based on the number of times that the construction monitoring personnel failed to complete the self-correction of risky behaviors, the frequency of the failure of the construction monitoring personnel to complete the self-correction of risky behaviors during the monitoring period was calculated to obtain the behavior inaccuracy rate; By formula: Construct a multi-behavior joint risk analysis model; in Obtain the KL divergence of different behavioral characteristics of construction monitoring personnel during the monitoring period, where P(x1, x2, x3) is the actual joint distribution of behavioral inaccuracy rate, and x1, x2, and x3 represent attention behavior characteristics, spatial position characteristics, and safety equipment characteristics, respectively; Q(x1), Q(x2), and Q(x3) are the canonical joint distributions of different characteristic behaviors, which are characterized by the behavioral inaccuracy threshold of the risk behavior of construction monitoring personnel; Calculate the KL divergence of each monitoring period and construct a period divergence sequence; Perform mutation analysis on the KL divergence of the periodic divergence sequence to identify the peak period of the KL divergence; The mutation analysis calculates the standard deviation of the periodic divergence sequence and obtains the monitoring period with KL divergence higher than 3 times the standard deviation as the peak period; Get all the peak periods in the periodic divergence sequence. If the peak periods are adjacent in the time dimension, merge them. Obtain the peak period and merge it, and then find the peak period with the largest time dimension as the peak monitoring period. Cluster analysis module: used to perform cluster analysis on the peak monitoring period of all construction monitoring personnel, obtain the largest cluster, and determine whether the difference of the largest cluster is significant; Obtain the peak monitoring period of all construction monitoring personnel and extract the periodic characteristics of the peak monitoring period; Perform Min-Max normalization on the periodic features and construct a periodic feature group; The cycle characteristics include: cycle duration, i.e., the duration of the peak monitoring cycle; peak intensity, i.e., the peak value of the behavior mismatch rate within the peak monitoring cycle; Based on the period feature groups of all construction monitoring personnel, the construction monitoring personnel with similar peak monitoring periods are clustered by clustering algorithm to obtain multiple clusters. By formula: Calculate the silhouette coefficient s(i) of each sample in the largest cluster, where i represents the sample number of the largest cluster, that is, the number of the construction monitoring personnel; a(i) represents the average distance from sample i to all other samples in the largest cluster, which is used to characterize the compactness of the sample within the cluster; b(i) represents the minimum average distance from sample i to all other clusters, which is used to characterize the inter-cluster separation of samples. All other clusters refer to the largest cluster excluding the largest cluster from different clusters. It should be noted that a(i) is calculated by calculating the average value of the Euclidean distance between the periodic characteristics of construction monitoring personnel i and other construction monitoring personnel in the largest cluster; b(i) is calculated by calculating the average Manhattan distance between the cycle characteristics of construction monitor i and other construction monitors in all other clusters; Calculate the mean of the silhouette coefficients of all samples in the largest cluster, calculate the proximity between the mean and the preset upper limit of the silhouette coefficient, and obtain the silhouette proximity value; Compare the preset proximity range values ​​of the contour proximity values. If the contour proximity values ​​are within the preset proximity range values, it is considered that the difference of the largest cluster is significant. If the contour proximity value is not within the preset proximity range, it is considered that the difference between the largest clusters is not significant; Hysteresis assessment module: If the difference is significant, it is used to establish a spatial analysis model to identify the clustering effect of the construction monitoring personnel's location and assess the degree of self-correction hysteresis of the risk behavior of the construction monitoring personnel; Obtain the spatial coordinates of the construction monitoring personnel in the deep foundation pit construction area, and calculate the distance between any two construction monitoring personnel in the deep foundation pit based on the spatial coordinates; The Gaussian kernel function is used to determine the distance attenuation weights between any two construction monitoring personnel and establish a spatial weight matrix. Based on the spatial weight matrix, a spatial analysis model is constructed through the global Moran index to identify the clustering effect of the construction monitoring personnel's location; If the global Moran index is less than 0, it indicates negative spatial correlation, that is, there is a discrete effect in the location of construction monitoring personnel; The global Moran index is equal to 0, indicating that the locations of construction monitoring personnel are randomly distributed; If there is a clustering effect in the location of the construction monitoring personnel, the deep foundation pit construction area where the construction monitoring personnel are located is merged according to the distance attenuation weight between any two construction monitoring personnel to obtain the construction monitoring area; Obtain the difference between the self-correction time and the monitoring time of the dangerous behavior of the construction monitoring personnel in the construction monitoring area after the system sends warning notifications of different risk levels to the construction personnel, and obtain the behavior correction delay; Obtain the average behavior correction delay of construction monitoring personnel in the construction monitoring area and the average behavior correction delay of all construction monitoring areas; The regional hysteresis index is obtained by calculating the ratio of the average behavior correction delay of construction monitoring personnel in the construction monitoring area to the average behavior correction delay of all construction monitoring areas; The regional hysteresis index is used to assess the degree of self-correction hysteresis of risky behaviors of construction monitoring personnel.

[0025] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A construction safety monitoring method based on a CV large model, characterized in that: The following steps are involved: Obtain the behavioral characteristics of self-correction of construction monitoring personnel during the monitoring period, and establish a multi-behavior joint risk analysis model to extract the peak monitoring period; Perform cluster analysis on the peak monitoring periods of all construction monitoring personnel to obtain the largest cluster, and determine whether the differences in the largest cluster are significant; If the difference is significant, it is used to establish a spatial analysis model to identify the clustering effect of the locations of construction monitoring personnel and to evaluate the degree of self-correction hysteresis of the risk behavior of construction monitoring personnel.

2. A construction safety monitoring method based on a CV large model according to claim 1, characterized in that: The method for obtaining the construction monitoring personnel is as follows: Obtain video data of deep foundation pit excavation and support operations, and identify the behavioral characteristics of construction workers through the CV large model; The behavioral characteristics of construction workers are compared and analyzed with the pre-built standard behavioral characteristics library to identify risky behaviors and mark construction workers with the highest behavioral risk level as construction monitoring personnel.

3. A construction safety monitoring method based on a CV large model according to claim 2, characterized in that: The method for performing vector comparison analysis is: Obtain standard behavior characteristics in deep foundation pit excavation and support construction, and build a standard behavior characteristic library; Extract the normative vector based on the normative feature behavior feature library, obtain the characteristic behavior of construction workers in real time through the CV large model, construct the behavior vector, and calculate the Manhattan distance between the normative vector and the behavior vector; If the Manhattan distance of the construction worker exceeds the preset standard distance range, the construction worker's behavior characteristics will be marked as risky behavior.

4. A construction safety monitoring method based on a CV large model according to claim 1, characterized in that: The peak monitoring period is obtained as follows: During the monitoring period, the CV large model is used to obtain the behavioral characteristics of the construction monitoring personnel's self-correction of risky behaviors; Calculate the frequency of construction monitoring personnel failing to complete self-correction of risk behaviors during the monitoring period to obtain the behavior inaccuracy rate; Based on the behavior inaccuracy rate, a multi-behavior joint risk analysis model is constructed to obtain the KL divergence of different behavior characteristics of construction monitoring personnel in the monitoring period and construct a period divergence sequence; Perform mutation analysis on the KL divergence of the periodic divergence sequence to obtain the peak period; After merging the peak periods, the peak period with the largest time dimension is selected as the peak monitoring period.

5. A construction safety monitoring method based on a CV large model according to claim 4, characterized in that: The peak period is obtained as follows: The standard deviation of the periodic divergence sequence is calculated, and the monitoring period in which the KL divergence is higher than 3 times the standard deviation is obtained as the peak period.

6. A construction safety monitoring method based on a CV large model according to claim 1, characterized in that: The way to determine whether the difference of the largest cluster is significant is: Obtain the peak monitoring period of all construction monitoring personnel, extract the periodic features of the peak monitoring period, and construct a periodic feature group; Based on the period feature groups of all construction monitoring personnel, the construction monitoring personnel with similar peak monitoring periods are clustered by clustering algorithm to obtain multiple clusters. By calculating the silhouette coefficient of multiple clusters, the silhouette proximity value is obtained; If the contour proximity value is within the preset proximity range, it is considered that the difference between the largest clusters is significant.

7. A construction safety monitoring method based on a CV large model according to claim 6, characterized in that: The method for obtaining the contour proximity value is as follows: Calculate the mean of the silhouette coefficients of all samples in the largest cluster, and calculate the proximity between the mean and the preset upper limit of the silhouette coefficient to obtain the silhouette proximity value.

8. The construction safety monitoring method based on the CV large model according to claim 1 is characterized in that: The evaluation method of the degree of self-correction hysteresis of the risk behavior of the monitored personnel is: A spatial analysis model was constructed through the global Moran index to determine the agglomeration effect of the construction monitoring personnel's location; If there is a clustering effect, the distances of the construction monitoring personnel are combined and analyzed to determine the construction monitoring area; The regional hysteresis index is obtained by calculating the ratio of the average behavior correction delay of construction monitoring personnel in the construction monitoring area to the average behavior correction delay of all construction monitoring areas; The regional hysteresis index is used to assess the degree of self-correction hysteresis of risky behaviors of construction monitoring personnel.

9. The construction safety monitoring method based on the CV large model according to claim 1 is characterized in that: The method for identifying the clustering effect of the construction monitoring personnel's location is: Obtain the spatial coordinates of the construction monitoring personnel in the deep foundation pit construction area and calculate the distance between any two construction monitoring personnel in the deep foundation pit; The Gaussian kernel function is used to determine the distance attenuation weights between any two construction monitoring personnel, and a spatial weight matrix is ​​established. The spatial analysis model is constructed through the global Moran index to identify the clustering effect of the construction monitoring personnel's locations.

10. A construction safety monitoring system based on a CV large model, used to implement a construction safety monitoring method based on a CV large model as described in any one of claims 1 to 9, characterized in that: Includes the following modules: Peak extraction module: used to obtain the behavioral characteristics of self-correction of construction monitoring personnel within the monitoring period, and establish a multi-behavior joint risk analysis model to extract the peak monitoring period; Cluster analysis module: used to perform cluster analysis on the peak monitoring period of all construction monitoring personnel, obtain the largest cluster, and determine whether the difference of the largest cluster is significant; Hysteresis assessment module: If the difference is significant, it is used to establish a spatial analysis model to identify the clustering effect of the locations of construction monitoring personnel and to assess the degree of self-correction hysteresis of the risk behavior of construction monitoring personnel.

Images