Building construction site risk assessment system based on AI
Data acquisition through multispectral cameras and environmental sensors, combined with deep learning and spatiotemporal convolutional networks, the AI risk assessment system has solved the problems of poor identification performance and insufficient real-time monitoring in complex environments, and achieved high accuracy and real-time risk assessment , ensure the safety of the construction site.
Patent Information
- Application Number
- CN202410968847.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-18
- Publication Date
- 2024-11-29
AI Technical Summary
The existing AI risk assessment system has poor identification performance in complex construction environments, especially in low light, snowy reflection and high dynamic range scenarios, and lacks real-time monitoring and abnormal detection capabilities, so it is impossible to detect safety hazards in a timely manner. .
Using multi-spectral cameras and environmental sensors to collect data in real time, improve image quality through multi-level filtering, contrast enhancement and light reflection compensation technologies, combine deep learning models and spatiotemporal convolutional networks to identify workers' behavior and equipment, and use dynamic weighted scoring models to calculate risk scores , and dynamically adjust notification priorities through the decision module.
It realizes high-accuracy identification and real-time risk assessment in complex environments, timely discovers that no safety equipment is worn or high-risk operations are operated, ensures safety on the construction site, and improves the stability and identification accuracy of the system under various weather conditions.
Smart Images

Figure CN119047820A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of AI recognition technology, and in particular to an AI-based construction site risk assessment system. Background Art
[0002] With the rapid development of artificial intelligence (AI) technology, AI has been widely used in risk assessment of construction sites. Existing technologies use AI to monitor and manage construction sites in real time, and improve the efficiency of safety management through image recognition, behavior detection, and risk assessment. However, traditional AI recognition and assessment systems still have some problems and shortcomings in complex construction environments.
[0003] Existing AI risk assessment systems have poor recognition performance in low-light environments, snowy reflective environments, and high dynamic range scenarios, making it difficult to obtain clear and accurate image data under special lighting conditions. In addition, the accuracy of multi-task recognition in complex environments is not high, and traditional AI systems cannot fully and accurately identify worker behavior and equipment wearing when dealing with multiple tasks. At the same time, traditional AI systems lack real-time monitoring and anomaly detection methods for worker behavior, and are unable to promptly detect and prevent safety hazards. In addition, existing static risk assessment methods lack dynamism, cannot reflect the actual risk situation at the construction site in real time, and lack timely and effective early warning and decision support.
[0004] To this end, those skilled in the art provide an AI-based construction site risk assessment system to solve the problems raised in the above background technology. Summary of the invention
[0005] In view of the shortcomings of the prior art, the present invention provides an AI-based construction site risk assessment system, which solves the problems of poor composite recognition ability and limitations in the prior art.
[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: an AI-based construction site risk assessment system, including a data acquisition module, a data processing module, an object recognition and detection module, a risk assessment module, a decision module, and a data management module;
[0007] The data acquisition module is used to collect images and environmental data of the construction site in real time through a multi-spectral camera and an environmental sensor;
[0008] The data processing module is used to improve the clarity and quality of the image and perform data fusion through multi-stage filtering, contrast enhancement and light reflection compensation technology;
[0009] The object recognition and detection module is used to recognize worker behavior and equipment and detect abnormal behavior;
[0010] The risk assessment module is used to bring in the environment, behavior and device status factor models to calculate the risk score in real time;
[0011] The decision module is used to construct an assessment threshold and dynamically adjust the notification priority according to the risk level when the threshold is exceeded;
[0012] The data management module is used to store data in layers and set access control.
[0013] Preferably, the data acquisition module includes a camera unit and a sensor unit; the camera unit includes an infrared camera, a panoramic camera, and an HDR camera; and the sensor unit includes a light sensor and a weather sensor.
[0014] Preferably, the data processing module includes an image preprocessing unit and a data fusion unit; the image preprocessing unit includes a multi-stage filter chain, a contrast enhancement unit, and a light reflection compensation unit; the data fusion unit fuses multi-spectrum through a Laplace pyramid fusion algorithm and fuses multi-sensor data through Kalman filtering.
[0015] Preferably, the object recognition and detection module includes a deep learning model unit and a behavior recognition unit; the deep learning model unit is used to recognize the worker's behavior and equipment wearing status; the behavior recognition unit is used to recognize and analyze the worker's behavior, detect abnormal behavior, and provide real-time warning.
[0016] Preferably, the deep learning unit processing steps are:
[0017] a. Feature extraction: Convolutional neural network is used for calculation, and the calculation formula is:
[0018]
[0019] Among them, f i,j,k is the pixel value in the feature map, W m,n,k is the convolution kernel, I i+m,j+n is the pixel value of the input image, b k is the bias, σ is the activation function;
[0020] b. Multi-task learning model: set up shared feature extraction layers and task-specific optimization layers;
[0021] c. Transfer learning: Use large-scale data sets for pre-training, and fine-tune the model based on data collected from the construction site after training to adapt the model to the characteristics of a specific environment. The calculation formula is:
[0022]
[0023] Among them, L(θ) is the loss function, θ is the model parameter, N is the number of samples, and yi is the true label, is the predicted value.
[0024] Preferably, the behavior recognition unit processes the following steps:
[0025] a. Spatiotemporal behavior analysis: Spatiotemporal convolutional network is used to extract spatiotemporal behavior features. The calculation formula is:
[0026]
[0027] Among them, f i,j,k,t is the pixel value in the spatiotemporal feature map, W m,n,p,k is the spatiotemporal convolution kernel, I i+m,j+n,t+p is the pixel value of the input spatiotemporal image, b k is bias;
[0028] b. Behavior classification: Based on the extracted spatiotemporal features, the workers’ behaviors are classified. The calculation formula is:
[0029]
[0030] in, is the output of behavior classification, W i is the weight of the classification layer, T st It is a spatiotemporal feature;
[0031] c. Anomaly detection: Long short-term memory network is used to identify abnormal behavior.
[0032] Preferably, the risk assessment module includes a real-time risk scoring unit;
[0033] The real-time risk scoring unit uses a dynamic weighted scoring model to calculate the score, and the calculation steps are:
[0034] a. Determination of risk factors: Setting risk factors for construction sites;
[0035] b. Dynamic weight calculation: Assign a dynamic weight w to each risk factor i (t), the weight is dynamically adjusted with time t, and the calculation formula is:
[0036]
[0037] Among them, σ i is the standard deviation of the ith risk factor, which is used to measure the volatility of the risk factor, α i (t) is an adjustment factor based on current environmental factors and is defined as:
[0038] α i (t) = 1 + β i f(e(t))
[0039] Among them, β i is the sensitivity parameter of the adjustment factor, and f(e(t)) is the influence function of environmental factors on the weight;
[0040] c. Risk score calculation: Calculate the comprehensive risk score R(t) based on the dynamic weight and the current value of the risk factor:
[0041]
[0042] Among them, to ensure that the risk score R(t) is within a reasonable range, regularization processing is required, which is defined as:
[0043]
[0044] Among them, R min and R max are the minimum and maximum risk scores in the historical data, respectively.
[0045] Preferably, the decision module includes an evaluation threshold setting unit and a notification priority dynamic unit;
[0046] The assessment threshold setting unit is used to set and dynamically adjust the assessment threshold of the risk score so as to trigger corresponding warning and decision-making operations at different risk levels;
[0047] The notification priority dynamic unit is used to dynamically adjust the priority of the notification according to the current risk level.
[0048] Preferably, the data acquisition module is electrically connected to the data processing module, the data processing module is network-connected to the object recognition and detection module, the risk assessment module is network-connected to the object recognition and detection module, the decision-making module is network-connected to the object recognition and detection module, and the data management module is electrically connected to the data acquisition module.
[0049] The AI-based construction site risk assessment method is used for the AI-based construction site risk assessment system, including the following steps:
[0050] S1. Real-time collection of infrared images, visible light images and HDR images of the construction site through multispectral cameras. These cameras cover key areas of the construction site. At the same time, environmental sensors are used to collect light intensity, temperature, humidity and weather conditions at the site.
[0051] S2. Preprocess the collected images, perform denoising through a multi-stage filter chain to reduce image noise, use contrast-limited adaptive histogram equalization to enhance image contrast and improve image quality, build a light reflection model and perform dynamic adjustment to reduce the impact of light reflection on image quality, and finally use the Laplace pyramid fusion algorithm to fuse multispectral images to generate high-quality composite images, and fuse multi-sensor data through the Kalman filter;
[0052] S3, using convolutional neural network algorithm to extract features from the processed image, including low-level features of edge and texture and high-level features of shape and object, and using multi-task learning model to identify different tasks based on the shared feature extraction layer;
[0053] S4. Use the spatiotemporal convolutional network algorithm to extract the spatiotemporal behavior characteristics of workers, combine the information of spatial and temporal dimensions to perform behavior analysis, and classify behaviors based on spatiotemporal characteristics to determine the wearing of safety equipment and high-risk operations. Finally, use the long short-term memory network to detect abnormal behaviors of workers, capture long-term dependencies, and identify abnormal behavior patterns.
[0054] S5. Combine real-time monitoring data and current environmental factors, use a dynamic weighted scoring model to dynamically adjust weights based on the current value, historical mean and standard deviation of each risk factor to calculate a comprehensive risk score;
[0055] S6. Set and dynamically adjust the assessment threshold of the risk score to distinguish different risk levels. Determine the current risk level by comparing the risk score with the set threshold. Dynamically adjust the priority of the notification and complete the notification based on the current risk level.
[0056] The present invention provides an AI-based construction site risk assessment system, which has the following beneficial effects:
[0057] 1. Through the spatiotemporal convolutional network and the long short-term memory network, the system of the present invention can conduct a detailed analysis of the spatiotemporal behavior characteristics of the workers, combine the information of the spatial and temporal dimensions, perform behavior classification and anomaly detection, and promptly discover the behavior of not wearing safety equipment or performing high-risk operations, so as to ensure safe operation at the construction site.
[0058] 2. The present invention combines real-time monitoring data and current environmental factors, adopts a dynamic weighted scoring model, and dynamically adjusts the weights to calculate the comprehensive risk score based on the current value, historical mean and standard deviation of each risk factor, thereby ensuring the real-time and accuracy of risk assessment, and timely distinguishing different risk levels by setting and dynamically adjusting the risk score threshold.
[0059] 3. The present invention adopts light reflection model construction and dynamic adjustment technology, which significantly reduces the reflection problem in snowy days and strong light conditions, improves the image availability and recognition accuracy, and ensures the stable operation of the system under various weather conditions.
[0060] 4. The present invention uses convolutional neural networks and multi-task learning models. The system can extract low-level and high-level features (such as edges, textures, shapes, and objects) from processed images, and accurately identify workers' behaviors and equipment wearing conditions, thereby improving the accuracy of multi-task recognition in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] Figure 1 It is a system framework structure diagram of the present invention;
[0062] Figure 2 It is a data acquisition module framework diagram of the present invention;
[0063] Figure 3 It is a data processing module framework diagram of the present invention;
[0064] Figure 4 This is a framework diagram of an object recognition and detection module of the present invention;
[0065] Figure 5 It is a schematic diagram of the risk assessment module framework of the present invention;
[0066] Figure 6 It is a schematic diagram of the decision module framework of the present invention;
[0067] Figure 7 It is a schematic diagram of the method steps of the present invention. DETAILED DESCRIPTION
[0068] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings 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 are within the scope of protection of the present invention.
[0069] Embodiment 1:
[0070] Please see attached Figure 1 , an embodiment of the present invention provides an AI-based construction site risk assessment system, including a data acquisition module, a data processing module, an object recognition and detection module, a risk assessment module, a decision module, and a data management module;
[0071] The data acquisition module is electrically connected to the data processing module, the data processing module is network-connected to the object recognition and detection module, the risk assessment module is network-connected to the object recognition and detection module, the decision-making module is network-connected to the object recognition and detection module, and the data management module is electrically connected to the data acquisition module.
[0072] Please see attached Figure 2 The data acquisition module is used to collect images and environmental data of the construction site in real time through multi-spectral cameras and environmental sensors; it includes a camera unit and a sensor unit; the camera unit includes an infrared camera, a panoramic camera, and an HDR camera; the sensor unit includes a light sensor and a weather sensor.
[0073] Specifically, the camera unit consists of an infrared camera, a panoramic camera, and an HDR camera, which can provide high-quality image data under various lighting conditions. The infrared camera is used to capture images in low-light or night conditions, the panoramic camera provides a wide-angle view to cover a larger area of the construction site, and the HDR camera captures details in a high dynamic range scene. The sensor unit includes a light sensor and a weather sensor, which are used to monitor the light intensity and weather conditions (such as temperature, humidity, precipitation, etc.) on site, respectively. Through these devices, the system can comprehensively and accurately obtain the real-time status of the construction site, providing a basis for subsequent data processing and risk assessment.
[0074] Please see attached Figure 3 The data processing module is used to improve the clarity and quality of the image and perform data fusion through multi-stage filtering, contrast enhancement and light reflection compensation technology; it includes an image preprocessing unit and a data fusion unit; the image preprocessing unit includes a multi-stage filter chain, a contrast enhancement unit, and a light reflection compensation unit; the data fusion unit fuses multi-spectra through a Laplace pyramid fusion algorithm and fuses multi-sensor data through a Kalman filter.
[0075] Specifically, firstly, Gaussian filtering, mean filtering and median filtering are sequentially performed on the collected image data to reduce different types of noise and improve the clarity of the image.
[0076] Contrast enhancement: The CLAHE method is used to enhance the contrast of the image, and the image details are improved through block processing and contrast limitation.
[0077] Light reflection compensation: Build a light reflection model and dynamically adjust the light reflection area according to real-time lighting data to reduce the impact of highlight reflection.
[0078] Image fusion: Use the Laplace pyramid fusion algorithm to perform layered fusion on the preprocessed multispectral images to generate a high-quality composite image. The specific steps are as follows:
[0079] Generate Laplacian pyramid: Generate a corresponding Laplacian pyramid for each input image (such as infrared image and visible light image). The Laplacian pyramid image of each layer is obtained by subtracting the Gaussian pyramid image of the adjacent layer.
[0080] Fusing Laplacian Pyramids: Fusing the Laplacian pyramids of multispectral images according to predefined fusion rules (such as weighted averaging, selecting high-frequency information, etc.). At each level, an appropriate fusion strategy is selected to preserve the details and advantages of each image.
[0081] Reconstruct the fused image: Reconstruct the fused Laplacian pyramid to generate a comprehensive high-quality image. The reconstruction process restores the fused image of the original resolution by expanding and adding layer by layer.
[0082] Sensor data fusion: Through Kalman filtering, state prediction, covariance prediction, Kalman gain calculation, state update and covariance update are performed on multi-sensor data to improve data accuracy and consistency.
[0083] Through the above data processing steps, the image preprocessing unit effectively reduces noise, improves image clarity and contrast, and reduces the impact of light reflection on image quality. The data fusion unit uses the Laplace pyramid fusion algorithm and Kalman filtering technology to efficiently fuse multispectral images and multi-sensor data to generate high-quality integrated images and consistent sensor data. These processing and fusion steps provide a high-quality and reliable data foundation for subsequent risk assessment, improving the accuracy and reliability of the entire system.
[0084] Please see attached Figure 4 The object recognition and detection module is used to identify worker behavior and equipment and detect abnormal behavior; it includes a deep learning model unit and a behavior recognition unit; the deep learning model unit is used to identify worker behavior and equipment wearing status; the behavior recognition unit is used to identify and analyze worker behavior, detect abnormal behavior, and provide real-time warning.
[0085] Specifically, the deep learning unit processing steps are:
[0086] a. Feature extraction: Convolutional neural network is used for calculation, and the calculation formula is:
[0087]
[0088] Among them, f i,j,k is the pixel value in the feature map, W m,n,k is the convolution kernel, I i+m,j+n is the pixel value of the input image, b k is the bias, σ is the activation function;
[0089] b. Multi-task learning model: set up shared feature extraction layers and task-specific optimization layers;
[0090] c. Transfer learning: Use large-scale data sets for pre-training, and fine-tune the model based on data collected from the construction site after training to adapt the model to the characteristics of a specific environment. The calculation formula is:
[0091]
[0092] Among them, L(θ) is the loss function, θ is the model parameter, N is the number of samples, and y i is the true label, is the predicted value.
[0093] Specifically, the behavior recognition unit processes the following steps:
[0094] a. Spatiotemporal behavior analysis: Spatiotemporal convolutional network is used to extract spatiotemporal behavior features. The calculation formula is:
[0095]
[0096] Among them, f i,j,k,t is the pixel value in the spatiotemporal feature map, W m,n,p,k is the spatiotemporal convolution kernel, I i+m,j+n,t+p is the pixel value of the input spatiotemporal image, b k is bias;
[0097] b. Behavior classification: Based on the extracted spatiotemporal features, the workers’ behaviors are classified. The calculation formula is:
[0098]
[0099] in, is the output of behavior classification, W i is the weight of the classification layer, T st It is a spatiotemporal feature;
[0100] c. Anomaly detection: Long short-term memory network is used to identify abnormal behavior.
[0101] Through the processing steps of the above object recognition and detection modules, the system can efficiently and accurately identify worker behavior and safety equipment wearing conditions, and detect abnormal behavior. The deep learning model unit improves the recognition ability and adaptability of the model through convolutional neural networks, multi-task learning, and transfer learning. The behavior recognition unit realizes comprehensive analysis and abnormality detection of worker behavior through spatiotemporal convolutional networks and long short-term memory networks, provides real-time warnings, and enhances the safety management level of the construction site.
[0102] Please see attached Figure 5,The risk assessment module is used to bring in the environment, behavior and equipment state factor models to calculate the risk score in real time; the risk assessment module includes a real time risk scoring unit;
[0103] Specifically, the real-time risk scoring unit uses a dynamic weighted scoring model to calculate the score. The calculation steps are as follows:
[0104] a. Determination of risk factors: Setting risk factors for construction sites;
[0105] b. Dynamic weight calculation: Assign a dynamic weight w to each risk factor i (t), the weight is dynamically adjusted with time t, and the calculation formula is:
[0106]
[0107] Among them, σ i is the standard deviation of the ith risk factor, which is used to measure the volatility of the risk factor, α i (t) is an adjustment factor based on current environmental factors and is defined as:
[0108] α i (t) = 1 + β i f(e(t))
[0109] Among them, β i is the sensitivity parameter of the adjustment factor, and f(e(t)) is the influence function of environmental factors on the weight;
[0110] c. Risk score calculation: Calculate the comprehensive risk score R(t) based on the dynamic weight and the current value of the risk factor:
[0111]
[0112] Among them, to ensure that the risk score R(t) is within a reasonable range, regularization processing is required, which is defined as:
[0113]
[0114] Among them, R min and R max are the minimum and maximum risk scores in the historical data, respectively.
[0115] Specifically, through the processing steps of the above risk assessment module, the system can dynamically and accurately calculate the comprehensive risk score of the construction site. Risk factors cover multiple aspects such as environment, behavior and equipment status to ensure the comprehensiveness of risk assessment. Dynamic weight calculation adjusts weights according to real-time environmental factors, so that the scoring model can adapt to site changes and improve the accuracy of risk assessment. Regularization processing ensures that the risk score is within a reasonable range, which facilitates subsequent risk level classification and decision-making.
[0116] Please see attached Figure 6 The decision module is used to construct an evaluation threshold. When the threshold is exceeded, the notification priority is dynamically adjusted according to the risk level. The decision module includes an evaluation threshold setting unit and a notification priority dynamic unit. The evaluation threshold setting unit is used to set and dynamically adjust the evaluation threshold of the risk score so as to trigger corresponding warnings and decision operations under different risk levels. The notification priority dynamic unit is used to dynamically adjust the notification priority according to the current risk level.
[0117] Specifically, the running steps are:
[0118] Initial threshold setting: Based on historical data and expert advice, multiple risk score thresholds are set to distinguish different risk levels.
[0119] Dynamically adjust thresholds: Based on real-time monitoring data and current environmental factors, dynamically adjust the assessment thresholds of risk scores to adapt them to actual on-site conditions.
[0120] Risk level classification: Compare the real-time risk score with the set threshold to determine the current risk level and trigger corresponding warnings and decision-making operations.
[0121] Initial notification priority setting: Set the initial notification priority for each risk level to ensure that different levels of risks have different emergency handling requirements.
[0122] Dynamically adjust notification priority: Dynamically adjust notification priority based on current risk level and environmental changes, and update the urgency of notifications in real time.
[0123] Select notification method: According to the adjusted notification priority, select the appropriate notification method (such as email, SMS, phone call, etc.) to ensure that high-priority notifications are delivered quickly through multiple methods.
[0124] Please see attached Figure 1 ,The data management module is used to store data in layers and set access control.
[0125] Specifically, the running steps are:
[0126] Local storage: Short-term storage of real-time collected data to ensure that data can be quickly accessed and processed. Use high-speed storage devices to store recent data and support fast read and write operations.
[0127] Cloud storage: long-term storage and backup of historical data and large-scale data. Use cloud storage services to upload large amounts of data to the cloud to ensure data security and persistence.
[0128] Data tiering: Data is stored in tiers based on its importance and access frequency. Real-time data and frequently accessed data are stored locally, while high-volume, infrequently accessed data are stored in the cloud.
[0129] Role definition: Define different user roles based on the responsibilities and permissions of system users to ensure the security of data access.
[0130] Permission allocation: Role-based access control (RBAC) assigns corresponding data access permissions to each role to prevent unauthorized access.
[0131] Access log: A log that records all data access operations, including access time, access user, access data type, and other information to improve system security.
[0132] Through the processing steps of the above data management module, the system can effectively store data in layers, improving data storage efficiency and access speed. The layered strategy of short-term storage and long-term storage optimizes the utilization of storage resources, ensuring the rapid processing of real-time data and the safe preservation of historical data. Role-based access control (RBAC) and detailed access log records enhance data security and privacy protection and prevent unauthorized access. The audit function of the access log can also promptly detect and handle abnormal access behaviors, further improving the security and reliability of the system.
[0133] Embodiment 2:
[0134] Please see attached Figure 7 , an embodiment of the present invention provides an AI-based construction site risk assessment method, comprising the following steps:
[0135] S1. Real-time collection of infrared images, visible light images and HDR images of the construction site through multispectral cameras. These cameras cover key areas of the construction site. At the same time, environmental sensors are used to collect light intensity, temperature, humidity and weather conditions at the site.
[0136] S2. Preprocess the collected images, perform denoising through a multi-stage filter chain to reduce image noise, use contrast-limited adaptive histogram equalization to enhance image contrast and improve image quality, build a light reflection model and perform dynamic adjustment to reduce the impact of light reflection on image quality, and finally use the Laplace pyramid fusion algorithm to fuse multispectral images to generate high-quality composite images, and fuse multi-sensor data through the Kalman filter;
[0137] S3, using convolutional neural network algorithm to extract features from the processed image, including low-level features of edge and texture and high-level features of shape and object, and using multi-task learning model to identify different tasks based on the shared feature extraction layer;
[0138] S4. Use the spatiotemporal convolutional network algorithm to extract the spatiotemporal behavior characteristics of workers, combine the information of spatial and temporal dimensions to perform behavior analysis, and classify behaviors based on spatiotemporal characteristics to determine the wearing of safety equipment and high-risk operations. Finally, use the long short-term memory network to detect abnormal behaviors of workers, capture long-term dependencies, and identify abnormal behavior patterns.
[0139] S5. Combine real-time monitoring data and current environmental factors, use a dynamic weighted scoring model to dynamically adjust weights based on the current value, historical mean and standard deviation of each risk factor to calculate a comprehensive risk score;
[0140] S6. Set and dynamically adjust the assessment threshold of the risk score to distinguish different risk levels. Determine the current risk level by comparing the risk score with the set threshold. Dynamically adjust the priority of the notification and complete the notification based on the current risk level.
[0141] Specifically, through the above steps, the AI-based construction site risk assessment method realizes efficient and comprehensive construction site monitoring and risk management. Real-time data collection of multispectral cameras and environmental sensors ensures comprehensive coverage and accurate environmental information. Image preprocessing and data fusion technology improves image quality, reduces the impact of noise and light reflection, and enhances data consistency and accuracy through multi-sensor data fusion.
[0142] Convolutional neural networks and multi-task learning models ensure accurate identification of worker behavior and equipment wearing. Spatiotemporal convolutional networks and long short-term memory networks achieve detailed spatiotemporal behavior analysis and real-time abnormal behavior detection to ensure safe on-site operations. The dynamic weighted scoring model combines real-time monitoring data and environmental factors to accurately calculate the comprehensive risk score, improving the real-time and accuracy of risk assessment.
[0143] By setting and dynamically adjusting the risk score threshold, it ensures that the corresponding warning and decision-making operations are triggered at different risk levels, and dynamically adjusts the notification priority to ensure that high-priority notifications can be delivered in a timely manner. Overall, this method significantly improves the risk monitoring and management capabilities of the construction site, effectively reduces safety hazards, and ensures the safe and smooth progress of the construction process.
[0144] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. AI-based construction site risk assessment system, characterized by: It includes data acquisition module, data processing module, object recognition and detection module, risk assessment module, decision-making module and data management module; The data acquisition module is used to collect images and environmental data of the construction site in real time through a multi-spectral camera and an environmental sensor; The data processing module is used to improve the clarity and quality of the image and perform data fusion through multi-stage filtering, contrast enhancement and light reflection compensation technology; The object recognition and detection module is used to recognize worker behavior and equipment and detect abnormal behavior; The risk assessment module is used to bring in the environment, behavior and device status factor models to calculate the risk score in real time; The decision module is used to construct an assessment threshold and dynamically adjust the notification priority according to the risk level when the threshold is exceeded; The data management module is used to store data in layers and set access control.
2. The AI-based construction site risk assessment system according to claim 1 is characterized in that: The data acquisition module includes a camera unit and a sensor unit; the camera unit includes an infrared camera, a panoramic camera, and an HDR camera; the sensor unit includes a light sensor and a weather sensor.
3. The AI-based construction site risk assessment system according to claim 1 is characterized in that: The data processing module includes an image preprocessing unit and a data fusion unit; the image preprocessing unit includes a multi-stage filter chain, a contrast enhancement unit, and a light reflection compensation unit; the data fusion unit fuses multi-spectrum data through a Laplace pyramid fusion algorithm and fuses multi-sensor data through a Kalman filter.
4. The AI-based construction site risk assessment system according to claim 1 is characterized in that: The object recognition and detection module includes a deep learning model unit and a behavior recognition unit; the deep learning model unit is used to recognize the worker's behavior and equipment wearing status; the behavior recognition unit is used to recognize and analyze the worker's behavior, detect abnormal behavior, and issue real-time warnings.
5. The AI-based construction site risk assessment system according to claim 4 is characterized in that: The deep learning unit processing steps are: a. Feature extraction: Convolutional neural network is used for calculation, and the calculation formula is: Among them, f i,j,k is the pixel value in the feature map, W m,n,k is the convolution kernel, I i+m,j+n is the pixel value of the input image, b k is the bias, σ is the activation function; b. Multi-task learning model: set up shared feature extraction layers and task-specific optimization layers; c. Transfer learning: Use large-scale data sets for pre-training, and fine-tune the model based on data collected from the construction site after training to adapt the model to the characteristics of a specific environment. The calculation formula is: Among them, L(θ) is the loss function, θ is the model parameter, N is the number of samples, and y i is the true label, is the predicted value.
6. The AI-based construction site risk assessment system according to claim 4 is characterized in that: The behavior recognition unit processing steps are: a. Spatiotemporal behavior analysis: Spatiotemporal convolutional network is used to extract spatiotemporal behavior features. The calculation formula is: Among them, f i,j,k,t is the pixel value in the spatiotemporal feature map, W m,n,p,k is the spatiotemporal convolution kernel, I i+m,j+n,t+p is the pixel value of the input spatiotemporal image, b k is bias; b. Behavior classification: Based on the extracted spatiotemporal features, the workers’ behaviors are classified. The calculation formula is: in, is the output of behavior classification, W i is the weight of the classification layer, T st It is a spatiotemporal feature; c. Anomaly detection: Long short-term memory network is used to identify abnormal behavior.
7. The AI-based construction site risk assessment system according to claim 1 is characterized in that: The risk assessment module includes a real-time risk scoring unit; The real-time risk scoring unit uses a dynamic weighted scoring model to calculate the score, and the calculation steps are: a. Determination of risk factors: Setting risk factors for construction sites; b. Dynamic weight calculation: Assign a dynamic weight w to each risk factor i (t), the weight is dynamically adjusted with time t, and the calculation formula is: Among them, σ i is the standard deviation of the ith risk factor, which is used to measure the volatility of the risk factor, α i (t) is an adjustment factor based on current environmental factors and is defined as: a i (t)=1+β i ·f(e(t)) Among them, β i is the sensitivity parameter of the adjustment factor, and f(e(t)) is the influence function of environmental factors on the weight; c. Risk score calculation: Calculate the comprehensive risk score R(t) based on the dynamic weight and the current value of the risk factor: Among them, to ensure that the risk score R(t) is within a reasonable range, regularization processing is required, which is defined as: Among them, R min and R max are the minimum and maximum risk scores in the historical data, respectively.
8. The AI-based construction site risk assessment system according to claim 1, characterized in that: The decision module includes an evaluation threshold setting unit and a notification priority dynamic unit; The assessment threshold setting unit is used to set and dynamically adjust the assessment threshold of the risk score so as to trigger corresponding warning and decision-making operations at different risk levels; The notification priority dynamic unit is used to dynamically adjust the priority of the notification according to the current risk level.
9. The AI-based construction site risk assessment system according to claim 1, characterized in that: The data acquisition module is electrically connected to the data processing module, the data processing module is network-connected to the object recognition and detection module, the risk assessment module is network-connected to the object recognition and detection module, the decision module is network-connected to the object recognition and detection module, and the data management module is electrically connected to the data acquisition module.
10. An AI-based construction site risk assessment method, used in an AI-based construction site risk assessment system according to any one of claims 1 to 9, characterized in that: The following steps are involved: S1. Real-time collection of infrared images, visible light images and HDR images of the construction site through multispectral cameras. These cameras cover key areas of the construction site. At the same time, environmental sensors are used to collect light intensity, temperature, humidity and weather conditions at the site. S2. Preprocess the collected images, perform denoising through a multi-stage filter chain to reduce image noise, use contrast-limited adaptive histogram equalization to enhance image contrast and improve image quality, build a light reflection model and perform dynamic adjustment to reduce the impact of light reflection on image quality, and finally use the Laplace pyramid fusion algorithm to fuse multispectral images to generate high-quality composite images, and fuse multi-sensor data through the Kalman filter; S3, using convolutional neural network algorithm to extract features from the processed image, including low-level features of edge and texture and high-level features of shape and object, and using multi-task learning model to identify different tasks based on the shared feature extraction layer; S4. Use the spatiotemporal convolutional network algorithm to extract the spatiotemporal behavior characteristics of workers, combine the information of spatial and temporal dimensions to perform behavior analysis, and classify behaviors based on spatiotemporal characteristics to determine the wearing of safety equipment and high-risk operations. Finally, use the long short-term memory network to detect abnormal behaviors of workers, capture long-term dependencies, and identify abnormal behavior patterns. S5. Combine real-time monitoring data and current environmental factors, use a dynamic weighted scoring model to dynamically adjust weights based on the current value, historical mean and standard deviation of each risk factor to calculate a comprehensive risk score; S6. Set and dynamically adjust the assessment threshold of the risk score to distinguish different risk levels. Determine the current risk level by comparing the risk score with the set threshold. Dynamically adjust the priority of the notification and complete the notification based on the current risk level.
Citation Information
Cited By
An artificial intelligence-based anti-oil theft warning system and method for oil tank trucks
CN119741761A
Student laboratory chemical operation control method and system
CN119763181A
A method and system for controlling chemical operations in student laboratories
CN119763181B
Electric power engineering construction safety monitoring system based on deep learning
CN119831180A
Method and system for automatically monitoring and early warning site safety risk
CN119832499A
