A construction site personnel risk identification method and system based on image processing

Abstract

The application provides a construction site personnel risk identification method and system based on image processing, and relates to the technical field of risk identification, comprising: first, collecting real-time, working condition and personnel image data of the construction site, dividing the area into multiple risk areas, combining site materials, equipment operation, fire-fighting supplies and other working condition data to calculate a risk assessment value according to a preset proportion and score to determine a risk assessment area. Then, the personnel image data is preprocessed to obtain personnel behavior data, the behavior prediction model is used to obtain behavior prediction probability data, and a danger judgment value is calculated according to the risk assessment area, and if the value exceeds the threshold, an early warning is triggered, and an optimization feedback mechanism is also provided to record the early warning event to optimize the early warning mechanism. The application calculates the danger judgment value of the behavior risk data and the behavior probability data in multiple risk assessment areas, and performs early warning according to the danger judgment value, thereby enhancing the timeliness and accuracy of risk early warning, and further ensuring the safety of construction workers.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of risk identification, in particular to a construction site personnel risk identification method and system based on image processing. BACKGROUND

[0002] The construction site is a complex and dangerous working environment, with a large number of personnel, complex operation processes, and a large number of equipment and materials, which can easily cause various safety accidents. Traditional safety management methods mainly rely on manual patrol and experience judgment, which has the problems of low efficiency, insufficient accuracy, and difficulty in realizing real-time monitoring. With the rapid development of computer technology and image processing technology, applying them to construction site personnel risk identification has become an innovative and effective solution. By processing and analyzing personnel images in real time on the construction site, abnormal behavior and potential dangers of personnel can be discovered in a timely manner, providing scientific basis and decision support for safety management. High-definition cameras are deployed in key areas of the construction site to ensure that they can cover areas where personnel are active frequently, such as construction areas, passageways, and material storage areas. The installation position and angle of the cameras need to be carefully designed to ensure that clear and complete personnel image data is obtained. At the same time, considering the complexity of the construction site environment, such as light changes, dust, and obstructions, cameras with wide dynamic range, low light performance, and certain dust and water resistance are selected. Personnel image data collected in the form of video streams or image sequences is transmitted to the backend processing system, providing a source of basic data for subsequent image processing and analysis.

[0003] Through the comprehensive application of a series of technical means such as personnel image data collection, preprocessing, feature extraction, behavior recognition, and risk assessment, real-time and accurate monitoring and risk assessment of personnel behavior on the construction site can be achieved. This not only helps to improve the efficiency and accuracy of safety management on the construction site, timely discovers and prevents the occurrence of safety accidents, but also provides new ideas and methods for intelligent safety management in the construction industry. With the continuous development and innovation of image processing technology and computer vision algorithms, it is believed that their application in the field of construction site personnel risk identification will become more mature and perfect, making greater contributions to building a safer and more efficient construction site environment.

[0004] In the prior art, the actual construction site has a complex environment that is difficult to accurately quantify the risk level of personnel in different risk areas, and construction workers are easily distracted by their work, which can cause danger in subsequent behavior, lack of early warning mechanisms for early warning, and thus avoid accidents. SUMMARY

[0005] The application provides a construction site personnel risk identification method and system based on image processing, to solve the defects that it is difficult to accurately quantify the danger degree of personnel in different risk areas and cannot predict personnel dangerous behaviors in advance in the prior art.

[0006] In one aspect, the application provides a construction site personnel risk identification method based on image processing, comprising:

[0007] S1: Collect real-time data, working condition data and personnel image data in the construction site, divide the construction site into multiple risk areas according to a preset safety standard, calculate the working condition data according to a preset proportion, obtain a risk assessment value, score the multiple risk areas according to the risk assessment value, and obtain corresponding risk assessment areas. The working condition data includes site material data, equipment operation data and fire-fighting supply data. The step of obtaining the preset proportion includes:

[0008] According to the analysis of the durability and safety of materials, the reliability and failure rate of equipment, and the effectiveness and coverage of fire-fighting facilities, the key factors affecting the safety of the construction site are obtained.

[0009] The key factors are divided into a target layer and a criterion layer, and the weight vector of the key factors is obtained by calculating the target layer and the criterion layer through the eigenvalue method.

[0010] It is judged whether the sum of the weight vector is equal to one. If yes, the preset proportion is obtained, otherwise the target layer and the criterion layer are recalculated.

[0011] S2: Preprocess the personnel image data to obtain personnel behavior data, establish a behavior prediction model, input the personnel behavior data into the prediction model, and obtain behavior prediction probability data.

[0012] S3: Calculate a danger judgment value according to the behavior prediction probability data combined with the corresponding risk assessment area, judge whether the danger judgment value exceeds a preset threshold, if yes, consider that the personnel are in a dangerous state, trigger an early warning mechanism, otherwise consider that the personnel are in a safe state and no alarm is needed.

[0013] S4: Establish an optimization feedback mechanism, record the event triggering the early warning mechanism to obtain a warning event, and optimize the early warning mechanism according to the warning event.

[0014] According to the construction site personnel risk identification method based on image processing provided by the application, in step S1, the step of obtaining the risk assessment value includes:

[0015] S11: According to the industry standard, the actual situation of the construction site and the safety management requirement, use the proportion setting algorithm to determine the preset proportion of the site material data, the equipment operation state and the fire-fighting supply data.

[0016] S12: calculating the risk assessment value in the risk assessment formula according to the preset proportion.

[0017] According to the building site personnel risk identification method based on image processing provided by the application, in step S12, the risk assessment formula is expressed as:

[0018]

[0019] In the formula, is the weight of the site material data, is the weight of the equipment operation state, is the weight of the fire-fighting supply data, is the site material data, is the equipment operation state, is the fire-fighting supply data, is the weight of each classification in the site material data, is the weight of each construction equipment operation parameter in the equipment operation state, is the weight of each classification in the fire-fighting supply data.

[0020] According to the building site personnel risk identification method based on image processing provided by the application, in step S2, the step of obtaining the personnel behavior data includes:

[0021] According to the clarity standard, the noise of the personnel image data is removed to obtain effective personnel activity images.

[0022] The posture feature, the action trajectory feature and the behavior feature in the effective personnel activity data are extracted.

[0023] The posture feature, the action trajectory feature and the behavior feature are subjected to personnel behavior identification and behavior classification to obtain personnel behavior sequences and behavior statistical data.

[0024] The personnel behavior sequences and the behavior statistical data are subjected to data cleaning, and the feature data is extracted from the personnel behavior sequences, and the feature data and the behavior statistical data are aggregated by using the summation method to obtain the personnel behavior data.

[0025] According to the building site personnel risk identification method based on image processing provided by the application, in step S2, the step of obtaining the behavior prediction probability data includes:

[0026] The behavior prediction model is constructed based on the recurrent neural network, and the behavior prediction model is trained by using the behavior statistical data.

[0027] The behavior prediction model is optimized using a grid search method, and whether the behavior prediction model performance reaches an excellent value is evaluated by accuracy, precision and recall. If yes, the optimization of the behavior prediction model is stopped, and the behavior prediction probability data is obtained by inputting the personnel behavior data into the behavior prediction model. Otherwise, the optimization of the behavior prediction model is continued.

[0028] According to the building site personnel risk identification method based on image processing provided by the application, in step S3, the plurality of risk areas include low risk areas, medium risk areas and high risk areas. The step of obtaining the danger judgment value includes:

[0029] S31: According to the safety regulations of the construction industry and the probability of different behaviors occurring by using historical data, the behavior prediction probability data is classified into behavior risk data and behavior probability data.

[0030] S32: Quantize the behavior risk data and behavior probability data to obtain the risk coefficient.

[0031] S32: Calculate the danger judgment value of the behavior risk data and the behavior probability data in the plurality of risk assessment areas.

[0032] According to the building site personnel risk identification method based on image processing provided by the application, in step S33, the formula of the danger judgment value is:

[0033]

[0034] In the formula, is the number of types of behavior risk data, is the probability of the occurrence of the first behavior, is the risk coefficient corresponding to the first behavior, is the risk assessment value of the risk area where the first behavior occurs.

[0035] According to the building site personnel risk identification method based on image processing provided by the application, in step S4, the step of optimizing the early warning mechanism includes:

[0036] S41: Clean the real-time data to obtain personnel activity data, and use the personnel activity data to update the behavior prediction model according to the feedback frequency, and re-extract various features in the effective personnel activity data to obtain target features.

[0037] S42: Adjust the target features using the back propagation algorithm.

[0038] S43: Record the events triggering the early warning mechanism to obtain early warning events, and deeply analyze the causes of the early warning events to obtain analysis results.​​​

[0039] S44: According to the analysis result, the warning rules and the preset threshold value in the warning mechanism are adjusted, and the priority and the notification mode of the warning mechanism are adjusted according to the importance and the urgency.

[0040] According to the building site personnel risk identification method based on image processing provided by the present application, in step S42, the step of adjusting the target feature comprises:

[0041] S421: The target feature is combined into an input vector and input into a behavior detection model to output a target behavior.

[0042] S422: A cross-entropy loss function is selected for a classification task, and the gradients of the weights and the bias are calculated according to the chain rule.

[0043] S423: The weights and the bias are updated using a gradient descent algorithm according to the gradients, so as to adjust the target feature.

[0044] On the other hand, the present application also provides a building site personnel risk identification system based on image processing, which adopts any one of the building site personnel risk identification methods based on image processing described above, and the risk identification system comprises:

[0045] The data acquisition and regional risk assessment module acquires real-time data, working condition data and personnel image data in the building site, divides the building site into multiple risk regions according to a preset safety standard, calculates the working condition data according to a preset proportion, obtains a risk assessment value, and scores the multiple risk regions according to the risk assessment value to obtain corresponding risk assessment regions.

[0046] The personnel image data processing and behavior prediction module pre-processes the personnel image data to obtain personnel behavior data, establishes a behavior prediction model, inputs the personnel behavior data into the prediction model, and obtains behavior prediction probability data.

[0047] The danger judgment and warning triggering module calculates a danger judgment value according to the behavior prediction probability data and the corresponding risk assessment region, judges whether the danger judgment value exceeds a preset threshold value, and if yes, considers that the personnel is in a dangerous state and triggers a warning mechanism, otherwise considers that the personnel is in a safe state and does not need to alarm.

[0048] The optimization feedback mechanism module establishes an optimization feedback mechanism, records the events triggering the warning mechanism to obtain warning events, and optimizes the warning mechanism according to the warning events.

[0049] The application provides a construction site personnel risk identification method and system based on image processing, risk assessment regions are obtained by dividing risk regions according to risk assessment values, the problem that personnel safety risks in a construction site are difficult to effectively identify and accurately control is solved, the beneficial effect that high-risk regions can be accurately positioned in advance is achieved, the construction party can strengthen safety protection measures, optimize resource allocation, and improve the overall safety guarantee level of the construction site.

[0050] The application provides a construction site personnel risk identification method and system based on image processing, behavior prediction probability data is obtained by inputting personnel behavior data into a behavior prediction model, the problem that personnel dangerous behaviors are difficult to predict in advance and accurate personnel risk warning basis is lacking is solved, the beneficial effect that the timeliness and accuracy of risk warning are enhanced and safety management measures are optimized is achieved.

[0051] The application provides a construction site personnel risk identification method and system based on image processing, dangerous judgment values of behavior risk data and behavior probability data in multiple risk assessment regions are calculated, and warning is performed according to the dangerous judgment values, the problem that the dangerous degree of personnel in different risk regions in the past construction site is difficult to accurately quantify and the warning mechanism is often broad and cannot timely issue an alarm for the dangerous condition of specific personnel in a specific region is solved, the safety of construction workers is further ensured when the construction workers work. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0053] Figure 1 It is a flowchart of a construction site personnel risk identification method based on image processing provided by the embodiment of the application. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical scheme and advantages of the application more clear, the technical scheme in the application will be clearly and completely described below in combination with the drawings in the application. Obviously, the described embodiments are some embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor belong to the protection scope of the application.

[0055] The following will be described in combination with Figure 1This invention describes a method and system for risk identification of construction site personnel based on image processing.

[0056] like Figure 1 As shown in the figure, the present invention provides a method and system for risk identification of construction site personnel based on image processing. The method mainly includes the following steps:

[0057] S1: Collect real-time data, operational data, and personnel image data within the construction site. Divide the construction site into multiple risk zones according to preset safety standards. Calculate the operational data based on preset weights to obtain risk assessment values. Then, score the multiple risk zones based on the risk assessment values ​​to obtain corresponding risk assessment areas. Operational data includes site material data, equipment operation data, and fire-fighting equipment data. The steps for obtaining preset weights include:

[0058] The key factors affecting on-site safety at construction sites are identified by analyzing the durability and safety of materials, the reliability and failure rate of equipment, and the effectiveness and coverage of fire protection facilities.

[0059] The key factors are divided into a target layer and a criterion layer. The weight vector of the key factors is calculated using the eigenvalue method for the target layer and the criterion layer. The formula is expressed as follows:

[0060]

[0061] In the formula, It is the first in the weight vector One factor, The key factor is the One factor, Indicates the number of key factors. It is the index variable for summation.

[0062] Determine if the sum of the weight vectors is equal to one. If it is, obtain the preset weight; otherwise, recalculate the weight vectors.

[0063] In step S1, the steps for obtaining the risk assessment value include:

[0064] S11: Based on industry standards, actual site conditions, and safety management requirements, a weighting algorithm is used to determine the preset weights for site material data, equipment operating status, and fire-fighting equipment data. Site material data is defined as follows: ,include( ),in It is the material size, It is the shape of the material, It is the material density, It refers to material hardness, defining the equipment operating status as... ,include( ), wherein is a performance parameter, is an operating mode, is an operating stability, is a device status indication, defining a fire-fighting equipment data classification as , comprising ( ), wherein is basic information data, is performance parameter data, is usage and operation data, is authentication and compliance data.

[0065] S12: Calculate the risk assessment value in the risk assessment formula according to the preset proportion.

[0066] The risk assessment formula is expressed as:

[0067]

[0068] In the formula, is the weight of site material data, is the weight of equipment operating state, is the weight of fire-fighting equipment data, is site material data, is equipment operating state, is fire-fighting equipment data, is the weight of each classification in the site material data, is the weight of each construction equipment operating parameter in the equipment operating state, is the weight of each classification in the fire-fighting equipment data.

[0069] S2: Preprocess the personnel image data to obtain personnel behavior data, establish a behavior prediction model, input the personnel behavior data into the prediction model, and obtain behavior prediction probability data.

[0070] In step S2, the step of obtaining personnel behavior data includes:

[0071] According to the clarity standard, the personnel image data is subjected to noise removal to obtain effective personnel activity images. By accurately identifying the noise components in the image and using appropriate filtering techniques to eliminate them, clear and accurate effective personnel activity images are successfully obtained, laying a solid foundation for further analysis.

[0072] Extract the posture features, motion trajectory features and behavior features in the effective personnel activity data. Among them, the posture features cover the relative position and angle information of each part of the body, the motion trajectory features record the motion path and speed change of the personnel within a certain time range in detail, and the behavior features reflect the behavior mode and purpose tendency of the personnel.

[0073] By analyzing posture features, movement trajectory features, and behavioral features, personnel behavior identification and classification are performed to obtain personnel behavior sequences and statistical data. This data can intuitively reflect key information such as personnel behavior patterns, activity regularities, and the frequency of various behaviors on construction sites. It is of paramount value for in-depth analysis of personnel behavior risks and the development of targeted safety management strategies.

[0074] Data cleaning is performed on the personnel behavior sequences and behavioral statistics, and feature data is extracted from the personnel behavior sequences. The feature data and behavioral statistics are then aggregated using a summation method to obtain the personnel behavior data.

[0075] The steps to obtain behavior prediction probability data include:

[0076] A behavior prediction model is constructed based on a recurrent neural network, and behavioral statistics are used to train the behavior prediction model.

[0077] The behavior prediction model is optimized using a grid search method. The model's performance is then evaluated based on accuracy, precision, and recall. If the model achieves excellent performance, optimization stops, and the human behavior data is input into the model to obtain the predicted behavior probability data. Otherwise, the optimization process continues.

[0078] S3: Calculate the danger judgment value based on the behavioral prediction probability data and the corresponding risk assessment area. Determine whether the danger judgment value exceeds the preset threshold. If it does, the personnel are considered to be in danger and the early warning mechanism is triggered. Otherwise, the personnel are considered to be in a safe state and no alarm is required.

[0079] In step S3, the multiple risk areas include low-risk areas, medium-risk areas, and high-risk areas. The steps for obtaining the hazard assessment value include:

[0080] S31: Based on safety regulations in the construction industry and by using historical data to statistically analyze the probability of different behaviors occurring, behavioral prediction probability data is classified into behavioral risk data and behavioral probability data.

[0081] S32: Quantify behavioral risk data and behavioral probability data to obtain a risk coefficient. The formula is expressed as:

[0082]

[0083] In the formula, It is a marker variable ( =1 represents behavioral risk data. =2 represents behavioral probability data). It is a count of relevant indicators. It is the probability of the behavior occurring. is a hazard degree coefficient, is a frequency quantization value, is a quantity conversion coefficient.

[0084] For the behavior risk data, corresponding weight coefficients are set according to different grades of safety hazards, and then the risk coefficients that can intuitively reflect the risk severity are obtained by weighted calculation combined with the occurrence probability. For the behavior probability data, the risk coefficients with clear numerical significance are also generated by quantitatively converting the occurrence frequency in different construction stages and different regional environments. Through this quantitative method, the abstract behavior data can be converted into measurable and comparable risk coefficient indicators, providing accurate data basis for subsequent risk judgment.

[0085] S33: Calculate the risk judgment value of the behavior risk data and the behavior probability data in the plurality of risk assessment regions. The risk judgment value can accurately reflect the overall risk degree of personnel behavior in a specific risk assessment region, and provides a key judgment basis for subsequent triggering of the early warning mechanism, so as to timely and effectively prevent and respond to possible dangerous situations.

[0086] The formula of the risk judgment value is:

[0087]

[0088] In the formula, is the number of types of behavior risk data, is the probability of the occurrence of the i-th behavior, is the risk coefficient corresponding to the i-th behavior, is the risk assessment value of the risk region in which the i-th behavior occurs. S4: Establish an optimized feedback mechanism, record the event triggering the early warning mechanism, obtain the early warning event, and optimize the early warning mechanism according to the early warning event. In step S4, the steps of optimizing the early warning mechanism include:

[0089] S41: Clean the real-time data to obtain personnel activity data, and use the personnel activity data to update the behavior prediction model according to the feedback frequency, and re-extract various features in the effective personnel activity data to obtain target features.

[0090] S42: Adjust the target features using the back propagation algorithm.

[0091]

[0092]

[0093] ​​​S43: Record the event triggering the early warning mechanism to obtain a warning event, and analyze the cause of the warning event in depth to obtain an analysis result.

[0094] S44: According to the analysis result, adjust the early warning rules and preset thresholds set in the early warning mechanism, and adjust the priority and notification mode of the early warning mechanism according to the importance and urgency.

[0095] In step S42, the step of adjusting the target features includes:

[0096] S421: From the numerous features obtained after preprocessing and feature extraction, carefully select the target features that are representative and key. Then, strictly follow the specific combination rules and order to cleverly combine these target features into a complete input vector. The construction of this input vector needs to fully consider the relevance between each feature and their importance in the overall behavior representation. Then, accurately input this carefully constructed input vector into the behavior detection model that is pre-trained and specially designed for construction site personnel behavior detection. After the complex calculation and analysis process inside the model, the target behavior corresponding to the input vector is finally output, which is the specific behavior category or action pattern of the construction site personnel determined by the model based on the input features.

[0097] S422: During the training and optimization process of the behavior detection model, for the classification task, after careful consideration and comparison, the cross-entropy loss function is selected as the key indicator to measure the difference between the model's prediction results and the true labels. This is because the cross-entropy loss function can effectively reflect the deviation between the model's predicted probability distribution and the true label's probability distribution when dealing with classification problems, thus providing a clear direction for model optimization. Then, according to the chain rule, which is widely used and extremely important in deep learning, the loss function is differentiated with respect to the weights and biases in the model. Through a meticulous and rigorous calculation process, the gradients of the weights and biases are obtained. These gradient values accurately represent the sensitivity of the loss function to the changes in weights and biases under the current model parameters, and they will serve as the key basis for updating the model parameters in the future, guiding the model to gradually optimize in the direction of reducing the loss function value.

[0098] S423: After obtaining the gradient of the weight and bias, the weight and bias in the model are updated according to the gradient descent algorithm, which is a classic and effective optimization algorithm. The core idea of the gradient descent algorithm is to gradually adjust the model parameters in the opposite direction of the gradient, so that the loss function value can be continuously reduced. Specifically, according to the pre-set learning rate, the weight and bias are respectively adjusted in the opposite direction of their respective gradients. The selection of the learning rate is crucial, which controls the step size of each parameter update. If the learning rate is too large, it may cause the model to skip the optimal solution during training, or even fail to converge. If the learning rate is too small, it will slow down the model training speed, requiring a large amount of time and computing resources. Through this weight and bias updating method based on the gradient descent algorithm, the effect of the target feature in the model can be effectively adjusted, so that the model can more accurately detect and classify the behavior of the construction site personnel, continuously improve the performance and accuracy of the model, and better adapt to the complex and variable construction site environment and personnel behavior patterns.

[0099] The construction site personnel risk identification method based on image processing provided by the embodiment can solve the problems that personnel safety risks in the construction site are difficult to effectively identify and accurately control, and can achieve the beneficial effects of accurately positioning high-risk areas in advance, so that the construction party can targetedly strengthen safety protection measures, optimize resource allocation, and improve the overall safety guarantee level of the construction site. By establishing a behavior prediction model and inputting personnel behavior data to obtain behavior prediction probability data, the problems that personnel dangerous behaviors are difficult to predict in advance and there is a lack of accurate personnel risk warning basis are solved, and the beneficial effects of enhancing the timeliness and accuracy of risk warning and optimizing safety management measures are achieved. By calculating the dangerous judgment value of the behavior risk data and the behavior probability data in the plurality of risk evaluation regions, and warning according to the dangerous judgment value, the problems that the dangerous degree of personnel in different risk regions in the construction site is difficult to accurately quantify, and the warning mechanism is often broad and cannot timely issue an alarm for the dangerous situation of specific personnel in a specific region are solved. When the construction workers work, the warning is given in advance, which can further protect the safety of the construction workers.

[0100] Based on the same overall inventive concept, the present application also protects a construction site personnel risk identification system based on image processing, which can adopt the above-mentioned construction site personnel risk identification method based on image processing. The risk identification system comprises:

[0101] The data acquisition and regional risk assessment module acquires real-time data, working condition data and personnel image data in the construction site, divides the construction site into multiple risk regions according to a preset safety standard, calculates the working condition data according to a preset proportion, obtains a risk assessment value, scores the multiple risk regions according to the risk assessment value, and obtains corresponding risk assessment regions;

[0102] The personnel image data processing and behavior prediction module pre-processes the personnel image data to obtain personnel behavior data, establishes a behavior prediction model, inputs the personnel behavior data into the prediction model, and obtains behavior prediction probability data.

[0103] The danger judgment and early warning triggering module calculates a danger judgment value according to the behavior prediction probability data and the corresponding risk assessment region, judges whether the danger judgment value exceeds a preset threshold, and if yes, considers that the personnel are in a dangerous state, triggers the early warning mechanism, and if not, considers that the personnel are in a safe state and no alarm is needed.

[0104] The optimization feedback mechanism module establishes an optimization feedback mechanism, records the event triggering the early warning mechanism to obtain an early warning event, and optimizes the early warning mechanism according to the early warning event.

[0105] The device embodiments described above are only illustrative, wherein the units illustrated as separate components can or can not be physically separated, and the components illustrated as units can or can not be physical units, i.e., can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. Those skilled in the art can understand and implement it without creative labor.

[0106] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and necessary universal hardware platforms, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of software products, which can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and include a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0107] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the same; and although the present application has been described in detail with reference to the foregoing embodiments, it should be appreciated by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features thereof can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An image processing-based construction site personnel risk identification method, characterized by, The method comprises the following steps: S1: collecting real-time data, working condition data and personnel image data in the construction site, dividing the construction site into multiple risk areas according to a preset safety standard, calculating a risk assessment value according to a preset proportion of the working condition data, and scoring the multiple risk areas according to the risk assessment value to obtain corresponding risk assessment areas; The working condition data comprises site material data, equipment operation data and fire-fighting supply data; The step of obtaining the preset proportion comprises: obtaining key factors affecting the safety of the construction site according to the durability and safety of the materials, the reliability and failure rate of the equipment, and the effectiveness and coverage of the fire-fighting facilities; dividing the key factors into a target layer and a criterion layer, and calculating the target layer and the criterion layer by the eigenvalue method to obtain a weight vector of the key factors; determining whether the sum of the weight vector is equal to one, if yes, obtaining the preset proportion, and if no, recalculating the weight vector; S2: preprocessing the personnel image data to obtain personnel behavior data, and establishing a behavior prediction model, inputting the personnel behavior data into the prediction model to obtain behavior prediction probability data; S3: calculating a danger judgment value according to the behavior prediction probability data combined with the corresponding risk assessment area, determining whether the danger judgment value exceeds a preset threshold, if yes, considering that the personnel are in a dangerous state, triggering an early warning mechanism, and if no, considering that the personnel are in a safe state, and no alarm is needed; The multiple risk areas comprise a low-risk area, a medium-risk area and a high-risk area; the step of obtaining the danger judgment value comprises: S31: classifying the behavior prediction probability data according to the safety regulations of the construction industry and the probability of different behaviors occurring according to historical data, into behavior risk data and behavior probability data; S32: quantifying the behavior risk data and the behavior probability data to obtain a risk coefficient; S33: calculating the danger judgment value of the behavior risk data and the behavior probability data in the multiple risk assessment areas; S4: establishing an optimization feedback mechanism, recording events triggering the early warning mechanism to obtain early warning events, and optimizing the early warning mechanism according to the early warning events.

2. The construction site personnel risk identification method based on image processing according to claim 1, characterized in that, In step S1, the step of obtaining the risk assessment value comprises: S11: determining the preset proportions of the site material data, the equipment operation state and the fire-fighting supply data by using a proportion setting algorithm according to industry standards, actual site conditions and safety management requirements; S12: calculating the risk assessment value in a risk assessment formula according to the preset proportions.

3. The construction site personnel risk identification method based on image processing according to claim 2, characterized in that, In step S12, the risk assessment formula is expressed as: ； wherein, is the weight of the site material data, is the weight of the equipment operation state, is the weight of the fire-fighting equipment data, is the site material data, is the equipment operation state, is the fire-fighting equipment data, is the weight of each classification in the site material data, is the weight of each construction equipment operation parameter in the equipment operation state, is the weight of each classification in the fire-fighting equipment data.

4. The construction site personnel risk identification method based on image processing according to claim 1, characterized in that, In step S2, the step of obtaining the personnel behavior data comprises: removing noise from the personnel image data according to a clarity standard to obtain effective personnel activity images; extracting posture features, motion trajectory features and behavior features from the effective personnel activity images; performing personnel behavior recognition and behavior classification on the posture features, the motion trajectory features and the behavior features to obtain personnel behavior sequences and behavior statistical data; The personnel behavior sequence and the behavior statistical data are data cleaned, and feature data is extracted from the personnel behavior sequence, and the feature data and the behavior statistical data are aggregated using a summation method to obtain personnel behavior data.

5. The construction site personnel risk identification method based on image processing according to claim 1, characterized in that, In step S2, the step of obtaining the behavior prediction probability data comprises: An behavior prediction model is constructed based on a recurrent neural network, and the behavior prediction model is trained using the behavior statistical data; The behavior prediction model is optimized using a grid search method, and whether the behavior prediction model performance reaches an excellent value is evaluated through accuracy, precision and recall rate, if yes, the optimization of the behavior prediction model is stopped, and the personnel behavior data is input into the behavior prediction model to obtain behavior prediction probability data; otherwise, the optimization of the behavior prediction model is continued.

6. The construction site personnel risk identification method based on image processing according to claim 1, characterized in that, In step S33, the formula of the danger judgment value is: ； In the formula, It is the number of types of behavioral risk data. It is the first The probability of this behavior occurring. It is the first The risk coefficient corresponding to this behavior It is the first Risk assessment value of the risk area where this behavior occurs.

7. The construction site personnel risk identification method based on image processing according to claim 1, characterized in that, In step S4, the step of optimizing the early warning mechanism comprises: S41: cleaning the real-time data to obtain personnel activity data, and updating the behavior prediction model using the personnel activity data according to the feedback frequency to re-extract various features in the effective personnel activity image to obtain target features; S42: adjusting the target features using a back propagation algorithm; S43: recording an event triggering the early warning mechanism to obtain a warning event, and deeply analyzing the cause of the warning event to obtain an analysis result; S44: according to the analysis result, adjusting the early warning rules and the preset threshold value in the early warning mechanism, and adjusting the priority and notification mode of the early warning mechanism according to importance and urgency.

8. The construction site personnel risk identification method based on image processing according to claim 7, characterized in that, In step S42, the step of adjusting the target features comprises: S421: combining the target features into an input vector, inputting into the behavior detection model, and outputting to obtain a target behavior; S422: selecting a cross-entropy loss function for a classification task, and calculating a gradient of weight and bias according to a chain rule; S423: updating the weight and bias using a gradient descent algorithm according to the gradient, so as to adjust the target features.

9. A construction site personnel risk identification system based on image processing, which adopts a construction site personnel risk identification method based on image processing according to any one of claims 1 to 8, characterized in that, The risk identification system comprises: A data acquisition and regional risk assessment module: collecting real-time data, working condition data and personnel image data in the construction site, dividing the construction site into multiple risk regions according to a preset safety standard, calculating the risk assessment value according to a preset proportion, and scoring the multiple risk regions according to the risk assessment value to obtain corresponding risk assessment regions; A personnel image data processing and behavior prediction module: preprocessing the personnel image data to obtain personnel behavior data, establishing a behavior prediction model, and inputting the personnel behavior data into the prediction model to obtain behavior prediction probability data; A danger judgment and early warning triggering module: calculating a danger judgment value according to the behavior prediction probability data and the corresponding risk assessment region, judging whether the danger judgment value exceeds a preset threshold value, if yes, considering that the personnel is in a dangerous state, triggering an early warning mechanism, otherwise, considering that the personnel is in a safe state, and no alarm is needed; The optimization feedback mechanism module: an optimization feedback mechanism is established, an early warning mechanism event is triggered, an early warning event is obtained, and the early warning mechanism is optimized according to the early warning event.

Images