Substation operation risk identification method based on multi-view video and high-precision positioning

Abstract

The invention relates to a substation operation risk identification method based on a multi-view video and high-precision positioning, and the method comprises the steps: obtaining monitoring video streaming images of a plurality of monitoring video point locations, and extracting the image features of an operator; rejecting monitoring video stream images with unclear or incomplete image features of operators; the coordinates of the operating personnel are determined through a Beidou high-precision positioning system or a UWB high-precision positioning system or a system combining the Beidou high-precision positioning system and the UWB high-precision positioning system, and the spatial position of the operating personnel is determined by combining the distances between the operating personnel and a plurality of monitors; determining a current operation process and link according to the spatial position of the operator; and judging the operation risk behavior type of the operating personnel in the current operation process and link according to the image features of the operating personnel and the spatial position of the operating personnel. According to the method, the problems that in the prior art, due to equipment shielding and illumination angle problems in a transformer substation, personnel operation risk behaviors cannot be accurately recognized through monitoring, and misjudgment is caused due to the fact that the spatial distance cannot be accurately quantified are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of substation operation monitoring, and relates to a substation operation risk identification method based on multi-view video and high-precision positioning. Background Art

[0002] As an important link in power transmission and distribution, substations are responsible for converting high-voltage electricity into low-voltage electricity and ensuring reliable power supply. However, the operating environment of substations is complex, and the operation process involves many high-risk factors, especially personnel operation or action errors will lead to serious safety accidents. Therefore, with the continuous growth of power demand and the expansion of the power grid, higher requirements are placed on the monitoring and identification of substation personnel's operating behavior risks.

[0003] However, the current video surveillance has a single viewing angle, which can easily cause the monitoring to fail to accurately identify the work of personnel due to equipment obstruction and lighting angle problems in the substation. The use of multiple video surveillance cannot be correlated with each other, and there is a lack of means to establish full-domain three-dimensional spatial monitoring, which leads to the inability to accurately quantify spatial distances and causes misjudgment. Summary of the invention

[0004] The present invention provides a method for identifying substation operation risks based on multi-view video and high-precision positioning, aiming to make full use of the construction achievements of the existing substation video monitoring system and high-precision positioning service system, and carry out joint prevention and control of substation operation risks with deep coordination of "multi-view video + high-precision positioning" in complex spaces, so as to solve the problem that the existing technology is prone to misjudgment caused by the inability to accurately identify personnel operation behaviors and the inability to accurately quantify spatial distances due to equipment obstruction and lighting angle problems in the substation.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] The present application provides a method for identifying substation operation risks based on multi-view video and high-precision positioning, comprising the following steps:

[0007] S1. Obtain surveillance video stream images of multiple surveillance video points, and extract image features of operators through the surveillance video stream images;

[0008] S2. Eliminate the monitoring video stream images with unclear or incomplete image features of the operator;

[0009] S3. Determine the coordinates of the operator through the Beidou system or UWB system, and determine the spatial position of the operator based on the distance between the operator and multiple monitoring devices;

[0010] S4. Determine the current operation process and link according to the spatial position of the operator;

[0011] S5. Determine whether the operator has any risky behavior in the current operation process and link based on the operator's image features and the operator's spatial position;

[0012] S6. When operators engage in risky operational behaviors in the current operational procedures and links, the smart safety helmets worn by the operators or the voice broadcasting system in the substation will automatically broadcast the current risky operational behaviors.

[0013] Furthermore, the acquisition of monitoring video stream images of multiple monitoring video points and the extraction of operator image features through the monitoring video stream images include the following steps:

[0014] S11. By installing cameras at different locations in the substation, multi-angle monitoring video streams can be obtained.

[0015] S12. Video stream processing: Use video processing software or framework to process the surveillance video stream and extract single-frame images from it. You can set the time interval to extract images to ensure that the operator's behavior is captured.

[0016] S13, image preprocessing: preprocessing the extracted image, including scaling, denoising, brightness adjustment, etc., to improve the effect of subsequent feature extraction. Commonly used preprocessing methods include histogram equalization, filtering, etc.

[0017] S14. Image feature extraction: Use machine learning models to extract operator image features.

[0018] Furthermore, in step S2, the step of removing the monitoring video stream images whose image features of the operators are unclear or incomplete comprises the following steps:

[0019] S21. Use computer vision technology to identify the face, key body parts and posture of the operator's image features to determine whether the operator's features in the image are complete;

[0020] S22, evaluating whether the image is clear using image quality assessment indicators, where the image quality assessment indicators include clarity, contrast, and brightness;

[0021] S23. Based on the quality assessment results, images that meet the standards are screened out from the video stream, and images that do not meet the clarity standards or have incomplete staff features are removed.

[0022] Further, in step S3, the coordinates of the operator are determined by the Beidou system or the UWB system, and the spatial position of the operator is determined in combination with the distance between the operator and multiple monitoring devices, including the following steps:

[0023] S31. System preparation: Equip the operators with Beidou positioning modules or UWB positioning modules, and arrange the required base stations or signal transmitters in the operation area to ensure signal coverage and stability;

[0024] S32, coordinate positioning: obtain the coordinates of the operator in space through the Beidou system or UWB system;

[0025] S33, distance calculation: after determining the coordinates of the operator, calculate the distance between the operator and multiple surveillance video points;

[0026] S34, spatial position determination: determining the current spatial position of the operator according to the coordinates of the operator and the distances between the operator and the multiple monitoring video points;

[0027] S35. Real-time monitoring and data updating: The spatial position of the operators is transmitted to the monitoring system in real time for monitoring and analysis, and the operators are tracked in real time by continuously acquiring and updating the position data.

[0028] Furthermore, the current operation process and link is determined according to the spatial position of the operator, specifically: the operation areas of different operation processes and links are preset, and the current operation process and link is determined according to the operation area where the spatial position of the operator is located.

[0029] Furthermore, the risky operational behaviors include unqualified personnel, wrong operating objects, entering the wrong interval, crossing the safety line, being close to live objects, wrongly hanging or missing ground wires, wrongly closing or missing ground switches, climbing without wearing safety belts, climbing with unmanned ladders, staying under the boom, large appliances and equipment being too close, and not wearing safety helmets.

[0030] Furthermore, judging whether the operator has any operation risk behavior in the current operation process and link according to the operator image features and the operator spatial position includes the following steps:

[0031] S51. Data collection: Collect large-scale image features of operators in historical periods and corresponding spatial location data samples of operators;

[0032] S52, data annotation: annotating the types of operation risk behaviors existing in the image features of the operators;

[0033] S53, using the image features and spatial positions of operators as explanatory variables and the corresponding operation risk behavior types as response variables, training a random forest model;

[0034] S54. Input the current operator image features and the operator spatial position into the random forest model to determine the operation risk behavior type of the current operation process and link.

[0035] Furthermore, the random forest model includes the following construction steps:

[0036] S531, using data fusion technology to process the image features of the operator and the spatial position data of the operator to obtain the operator behavior feature vector;

[0037] S532, taking the operator behavior feature vector as input and the operation risk behavior type as target output, training a random forest model;

[0038] S533. Use confusion matrix to evaluate the accuracy of random forest model in predicting the type of operational risk behavior.

[0039] Furthermore, the data fusion technology specifically includes:

[0040] Data encoding is performed based on the image features of the operator to obtain the behavior code;

[0041] The spatial position weight is calculated according to the spatial position of the operator, specifically: a coordinate system is established with the symmetry center of the specific working area of ​​the operator as the origin, and the spatial position weight is calculated according to the distance between the spatial position of the operator and the symmetry center;

[0042] The behavior code and spatial position weight are combined to form the operator behavior feature vector.

[0043] Furthermore, in step S6, the current operation risk behavior type is automatically broadcasted, and when the operator does not stop the operation risk behavior within the specified time, an alarm message is reported to the management personnel.

[0044] Beneficial effects of the present invention:

[0045] By acquiring the monitoring video stream images of multiple monitoring video points, the image features of the operators are extracted; the monitoring video stream images with unclear or incomplete image features of the operators are eliminated; the coordinates of the operators are determined by the Beidou system or the UWB system or a system that has both, and the spatial position of the operators is determined in combination with the distance between the operators and multiple monitorings; the current operation process and link are determined according to the spatial position of the operators; the type of operation risk behavior of the operators in the current operation process and link is determined according to the image features of the operators and the spatial position of the operators. The present invention solves the problem that the existing technology is prone to misjudgment caused by the inability of monitoring to accurately identify the operation risk behavior of the operators due to equipment obstruction and lighting angle problems in the substation, and the inability to accurately quantify the spatial distance. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0047] Figure 1The present invention is a flowchart of a method for identifying substation operation risks based on multi-view video and high-precision positioning.

[0048] Figure 2 The present invention is a flowchart of extracting operator image features by monitoring video stream images in one embodiment of the present invention.

[0049] Figure 3 This is a flow chart for judging whether an operator has any operational risk behavior in the current operating process and link based on the operator's image features and the operator's spatial position in one embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation mode, structure, characteristics and effects of the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments.

[0051] See also Figure 1 The present application provides a method for identifying substation operation risks based on multi-view video and high-precision positioning, comprising the following steps:

[0052] S1. Obtain surveillance video stream images of multiple surveillance video points, and extract image features of operators through the surveillance video stream images;

[0053] In this embodiment, by arranging multiple monitoring points to obtain video streams, the monitoring effect of the operation site can be effectively improved. First, after installing the camera at different key positions, the monitoring video stream is obtained in real time, so that the operation area can be fully covered to ensure that the various activities of the operators are captured. Then it is necessary to extract key frame images from the video stream and pre-process these images, including noise reduction, adjusting brightness and size, etc., to improve the accuracy of subsequent analysis. On the basis of image processing, computer vision technology is used to extract the characteristics of the operators. Traditional methods such as HOG (Histogram of Oriented Gradients) and SIFT (Scale Invariant Feature Transform) can be selected, or deep learning methods such as convolutional neural networks (CNN) can be used to identify and classify operators in images. In addition, the extracted image features can also be used to analyze the behavior of the operators, and machine learning algorithms are used to determine the specific activities they perform, which is crucial to improving safety management. Finally, by combining the extracted features with the results of behavioral analysis, real-time monitoring and feedback of the behavior of the operators can be achieved, effectively preventing safety hazards and ensuring the safety and efficiency of the working environment. Such a system not only improves the management level of the operation site, but also provides data support and decision-making basis for safe production.

[0054] Furthermore, the acquisition of monitoring video stream images of multiple monitoring video points and the extraction of operator image features through the monitoring video stream images include the following steps:

[0055] Obtaining surveillance video stream images of multiple surveillance video points and extracting image features of operators can usually be achieved through the following steps:

[0056] S11. By installing cameras at different locations in the substation, multi-angle monitoring video streams can be obtained.

[0057] It should be noted that monitoring should ensure that the distribution of cameras can cover key operating areas to facilitate comprehensive monitoring of the dynamics of operators.

[0058] S12. Video stream processing: Use video processing software or framework (such as OpenCV, FFmpeg, etc.) to process the surveillance video stream and extract single-frame images from it. You can set the time interval to extract images to ensure that the operator's behavior is captured.

[0059] S13, image preprocessing: preprocessing the extracted image, including scaling, denoising, brightness adjustment, etc., to improve the effect of subsequent feature extraction. Commonly used preprocessing methods include histogram equalization, filtering, etc.

[0060] S14. Image feature extraction: Use machine learning models to extract operator image features.

[0061] Furthermore, the machine learning model is configured as a convolutional neural network model.

[0062] In this embodiment, the use of a convolutional neural network (CNN) model to extract operator image features is an efficient and accurate computer vision technology that is widely used in the fields of safety monitoring and behavior analysis. First of all, data collection is the first step in this process, and a large number of operation-related images or videos need to be obtained from the real-time monitoring system. At this stage, it is crucial to ensure the diversity and representativeness of the data to cover different work scenarios, personnel postures, and lighting conditions. Next, the collected image data is preprocessed, including adjusting the image size, normalizing pixel values, and data enhancement. The data set is expanded by means of rotation, flipping, or random cropping to improve the robustness and generalization ability of the model.

[0063] Constructing a convolutional neural network model is the next key step. Generally speaking, the structure of a CNN consists of multiple convolutional layers, activation layers, and pooling layers. The main function of the convolutional layer is to extract the spatial features of the image through a local perception mechanism, while the activation layer (usually using the ReLU activation function) introduces nonlinearity, allowing the network to learn complex features. Pooling layers (such as maximum pooling or average pooling) not only improve computational efficiency, but also effectively reduce the size of the feature map, thereby avoiding the overfitting problem.

[0064] In the process of training the model, the preprocessed image is input into CNN. After a series of convolution and pooling operations, feature vectors for identifying operators can be obtained. These feature vectors contain the appearance information and posture features of operators in a high-dimensional form, which can provide a sufficient data basis for subsequent classification and recognition. In order to improve the accuracy of feature extraction, the transfer learning method can be used to use the CNN model pre-trained on a large data set (such as VGG16, ResNet, etc.) to fine-tune it to adapt to specific application scenarios.

[0065] After feature extraction is completed, the model can classify and identify operators. By inputting the extracted features into the fully connected layer and using the softmax function for multi-category classification, the system can identify operators in the monitoring screen in real time. In addition, these features can be input into other machine learning algorithms for further analysis, such as behavior recognition and anomaly detection, to improve overall monitoring efficiency.

[0066] Randomly divided training and test sets can help evaluate the performance of the model, and the cross-validation method is used to ensure the stability and reliability of the model. Ultimately, the fully trained and tuned CNN model can efficiently extract the image features of operators in actual monitoring applications, achieve accurate identity recognition and timely safety warnings, and thus provide strong support for ensuring the safety of the working environment.

[0067] S2. Eliminate the monitoring video stream images with unclear or incomplete image features of the operator;

[0068] In the surveillance video stream, it is crucial to eliminate images with unclear or incomplete image features of operators to ensure data quality and the accuracy of subsequent analysis. Compared with single-view monitoring, the use of multi-view monitoring systems has significant advantages and can capture and analyze the behavior and status of operators more comprehensively. First, the multi-view system can effectively reduce the impact of blind spots, provide clearer images of operators, and reduce blur or incomplete situations caused by poor angles. Through the image quality assessment tool, each frame of the image is automatically analyzed, and combined with indicators such as contrast, clarity and noise level, it can more accurately identify and filter poor quality images. Secondly, the multi-view system can capture the complete features of the operator from different angles, and use target detection algorithms (such as YOLO or SSD) to effectively identify the status of the operator without being limited by a single view. In addition, the analysis of lighting conditions can also better evaluate the image quality at different viewing angles, ensuring that clear monitoring images can be obtained regardless of environmental changes. When applying face or body recognition technology, multi-view data can improve the accuracy of recognition and reduce the rate of false recognition. In summary, compared with a single perspective, multi-perspective monitoring significantly improves the comprehensiveness and reliability of data analysis through rich perspective information and high-quality images, providing a more solid foundation for decision-making.

[0069] Furthermore, in step S2, the step of removing the monitoring video stream images whose image features of the operators are unclear or incomplete comprises the following steps:

[0070] S21. Use computer vision technology to identify the face, key body parts and posture of the operator's image features to determine whether the operator's features in the image are complete;

[0071] S22, evaluating whether the image is clear using image quality assessment indicators, where the image quality assessment indicators include clarity, contrast, and brightness;

[0072] In this embodiment, certain threshold standards are set for image quality evaluation indicators such as clarity, contrast and brightness. If the image quality is lower than the threshold, it means that the image does not meet the clarity requirement, and then it can be determined that the image feature is unclear or incomplete.

[0073] S23. Based on the quality assessment results, images that meet the standards are screened out from the video stream, and images that do not meet the clarity standards or have incomplete staff features are removed.

[0074] S3. Determine the coordinates of the operator through the Beidou system or UWB system, and determine the spatial position of the operator based on the distance between the operator and multiple monitoring devices;

[0075] In this embodiment, the coordinates of the operator are determined by the Beidou high-precision positioning system or the UWB high-precision positioning system or a system with both, which can achieve accurate positioning in a normal environment. Combined with the installation positions and real-time monitoring data of multiple surveillance cameras, the spatial position acquisition accuracy of the operator can be further improved. First, the positioning service provided by the Beidou high-precision positioning system and the UWB high-precision positioning system enables each operator to obtain its latitude and longitude coordinates in real time and prepare basic data for subsequent analysis. By equipping each operator with a high-precision positioning device, continuous tracking and updating of its position can be ensured. Secondly, the coordinates of the operator are combined with the installation position data of multiple surveillance cameras. Each camera has a fixed field of view. Using the principle of triangulation, the distance between the operator and each camera can be calculated to accurately determine its spatial position in the monitoring area. Specifically, with the help of image data obtained by multiple surveillance cameras, combined with real-time positioning information, real-time tracking can be performed to ensure that the operator within the monitoring range is always within the visible range. By combining the positioning information of the Beidou system or UWB system with the surveillance camera data, it not only improves the positioning accuracy of the operators in space, but also monitors their behavior in real time to ensure safety and efficiency. The application of this technology can significantly reduce the potential risks caused by unclear positions or strong electromagnetic interference, and provide data support for more refined management. At the same time, the system can also use automatically generated virtual fences to promptly alert operators of deviations from their positions, further improving the level of intelligent on-site management.

[0076] Further, in step S3, the coordinates of the operator are determined by the Beidou system or the UWB system, and the spatial position of the operator is determined in combination with the distance between the operator and multiple monitoring devices, including the following steps:

[0077] S31. System preparation: Equip the operators with Beidou positioning modules or UWB positioning modules, and arrange the required base stations or signal transmitters in the operation area to ensure signal coverage and stability;

[0078] It should be noted that the Beidou positioning module or UWB positioning module equipped by the operators needs to ensure that the equipment is calibrated before departure to improve positioning accuracy.

[0079] S32, coordinate positioning: obtain the coordinates of the operator in space through the Beidou system or UWB system;

[0080] In this embodiment, the operator's device receives signals from satellites or base stations through the Beidou system or the UWB system. The Beidou system uses multiple satellites for triangulation, while the UWB system calculates distance by measuring the time difference between the sent signal and the received signal. With this information, the device can calculate the operator's exact coordinates in space.

[0081] S33, distance calculation: after determining the coordinates of the operator, calculate the distance between the operator and multiple surveillance video points;

[0082] In this embodiment, after determining the coordinates of the operator, the distance between the operator and multiple monitoring devices needs to be calculated. For the UWB system, the device can accurately measure the distance to each monitoring device by the received signal strength and time difference. For the Beidou system, it is necessary to combine satellite data for triangulation, while considering the influence of terrain and obstacles to obtain a more accurate distance.

[0083] S34, spatial position determination: determining the current spatial position of the operator according to the coordinates of the operator and the distances between the operator and the multiple monitoring video points;

[0084] In this embodiment, the coordinates of the operator are combined with the coordinates of the monitoring device, and the spatial position of the operator can be determined using a geometric calculation method. If the distance data of multiple monitoring devices are available, further calculations can be performed to optimize the spatial position identification of the operator using a positioning algorithm (such as the least squares method) to improve the accuracy and reliability of positioning.

[0085] S35. Real-time monitoring and data updating: The spatial position of the operators is transmitted to the monitoring system in real time for monitoring and analysis, and the operators are tracked in real time by continuously acquiring and updating the position data.

[0086] In this embodiment, the location data of the operator is transmitted to the monitoring system in real time for monitoring and analysis. By continuously acquiring and updating the location data, the operator can be tracked in real time to ensure their safety and improve work efficiency. At the same time, the system can set a safety zone. Once the operator exceeds the predetermined range, the system can immediately send an alarm.

[0087] S4. Determine the current operation process and link according to the spatial position of the operator;

[0088] In the embodiment, the current operation process and link of the operator is determined according to the spatial position of the operator, and first of all, the positioning data and operation process information need to be integrated. The operation area is usually pre-defined, and each process is clearly divided in space. The operator is tracked in real time using a positioning system (such as the Beidou system and the UWB system), and the current operation process can be automatically identified when the operator enters a specific area. For example, when the operator enters the "grounding location area", the system can judge it as the "grounding preparation process" in real time. In addition, combined with real-time monitoring data, the inspector can obtain the progress status of the operation to ensure that the personnel operate in accordance with the regulations and issue an alarm in time when they encounter deviations. At the same time, by analyzing the operation efficiency and operation quality of different links, managers can identify bottlenecks and optimize the operation process. The whole process can also be recorded to form historical data for subsequent review and analysis, and further improve the safety and standardization of the operation. By effectively using spatial location data, not only can the process status of the operator be grasped in real time, but also the improvement of work efficiency and the reduction of accident risks can be achieved in dynamic management.

[0089] Furthermore, the current operation process and link is determined according to the spatial position of the operator, specifically: the operation areas of different operation processes and links are preset, and the current operation process and link is determined according to the operation area where the spatial position of the operator is located.

[0090] S5. Determine whether the operator has any risky behavior in the current operation process and link based on the operator's image features and the operator's spatial position;

[0091] In this embodiment, for different operation procedures and links, operators have corresponding operation risk behaviors, including unqualified personnel qualifications, wrong operation objects, mistakenly entering intervals, safety crossing lines, proximity to live objects, wrong hanging or missing ground wires, wrong or missing ground cutters, climbing without safety belts, climbing without escalators, staying under the boom, large appliances and equipment are too close, and not wearing safety helmets. Once these behaviors occur, they may lead to safety hazards. First, unqualified personnel qualifications indicate that the operator is not familiar with safety regulations and procedures, does not have general or special operation qualifications, or has a history of violations and has been included in the blacklist, which poses a great safety risk. This embodiment can perform face recognition on the operator image to determine whether the current operator has relevant qualifications. Wrong operation object means that the operator may have mistakenly entered a non-working area during the operation, which may cause an accident. Mistakenly entering intervals and safety crossing lines indicate that the operator may have entered a safe area or a restricted area, and this behavior will increase the possibility of accidents. The risk of being close to live objects needs to be paid special attention to, because this behavior may cause the risk of electric shock and endanger personnel safety. Incorrectly hanging or missing ground wires, or incorrectly or missing grounding switches, means that workers do not use safety tools or connection equipment correctly during electrical work, which may cause electrical failures or accidents. Not wearing a safety belt when working at heights, or working on an unmanned ladder, significantly increases the risk of falling from heights. Staying under the boom, or keeping large equipment too close to equipment, are both improper behaviors when operating near heavy equipment, which may cause objects to fall or collide. Finally, not wearing a safety helmet is the most basic lack of safety protection measures, which greatly increases the risk of being hit on the head by objects.

[0092] Furthermore, the risky operational behaviors include unqualified personnel, wrong operating objects, entering the wrong interval, crossing the safety line, being close to live objects, wrongly hanging or missing ground wires, wrongly closing or missing ground switches, climbing without wearing safety belts, climbing with unmanned ladders, staying under the boom, large appliances and equipment being too close, and not wearing safety helmets.

[0093] Furthermore, judging whether the operator has any operation risk behavior in the current operation process and link according to the operator image features and the operator spatial position includes the following steps:

[0094] S51. Data collection: Collect large-scale image features of operators in historical periods and corresponding spatial location data samples of operators;

[0095] S52, data annotation: annotating the types of operation risk behaviors existing in the image features of the operators;

[0096] S53, using the image features and spatial positions of operators as explanatory variables and the corresponding operation risk behavior types as response variables, training a random forest model;

[0097] S54. Input the current operator image features and the operator spatial position into the random forest model to determine the operation risk behavior type of the current operation process and link.

[0098] In this embodiment, the random forest model is a machine learning method based on ensemble learning, which consists of multiple decision trees. Each decision tree is constructed independently, and the diversity of the model is increased by random sampling and feature selection of input data. The final prediction result is obtained by integrating the prediction results of all decision trees.

[0099] Furthermore, the random forest model includes the following construction steps:

[0100] S531, using data fusion technology to process the image features of the operator and the spatial position data of the operator to obtain the operator behavior feature vector;

[0101] In this embodiment, this step aims to extract the identity, posture and action features of the operator in the image through a deep learning model, and at the same time parse the position coordinates to construct a structured feature vector. Subsequently, a weighted algorithm and a deep learning algorithm are used to effectively fuse different features to form a comprehensive feature representation. This integrated feature vector is used to identify and classify the behavior patterns of the operators, such as a certain link in a certain process. Finally, a deep learning relationship can be established with the type of operational risk behavior to achieve a predictive effect.

[0102] Furthermore, the data fusion technology specifically includes:

[0103] Data encoding is performed based on the image features of the operator to obtain the behavior code;

[0104] In this embodiment, data encoding based on the image features of the operator is to assign letters or numbers to the identity, posture and action features of the operator in the image to obtain a feature code (behavior code) having a sequence of letters or numbers.

[0105] The spatial position weight is calculated according to the spatial position of the operator, specifically: a coordinate system is established with the symmetry center of the specific working area of ​​the operator as the origin, and the spatial position weight is calculated according to the distance between the spatial position of the operator and the symmetry center;

[0106] In this embodiment, the spatial position weight is calculated based on the spatial position of the operator. The symmetry center of the operator's specific work area can be used as the origin to establish coordinates, and the weight of each coordinate is calculated to indicate the closeness between the operator and his work area, which is used to determine whether the operator is in a specific work area.

[0107] Furthermore, the spatial position weight is calculated as follows:

[0108]

[0109] Where W d is the spatial position weight of the operator; D s D is the distance between the intersection point and the symmetry center generated by drawing the line connecting the operator's spatial position and the symmetry center and the boundary of the specific work area; i It is the distance between the operator's spatial position and the center of symmetry.

[0110] The behavior code and spatial position weight are combined to form the operator behavior feature vector.

[0111] In this embodiment, the operator behavior feature vectors of multiple data samples will form a data matrix with behavior codes and spatial positions as columns (variables) and operation risk behavior types as rows (objects or samples). The trained random forest model determines the operation risk behavior type under the comprehensive effect of multiple variables.

[0112] S532, taking the operator behavior feature vector as input and the operation risk behavior type as target output, training a random forest model;

[0113] S533. Use confusion matrix to evaluate the accuracy of random forest model in predicting the type of operational risk behavior.

[0114] S6. When operators engage in risky operational behaviors in the current operational procedures and links, the smart safety helmets worn by the operators or the voice broadcasting system in the substation will automatically broadcast the current risky operational behaviors.

[0115] Furthermore, in step S6, the current operation risk behavior type is automatically broadcasted, and when the operator does not stop the operation risk behavior within the specified time, an alarm message is reported to the management personnel.

[0116] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for identifying substation operation risks based on multi-view video and high-precision positioning, characterized in that: The following steps are involved: S1. Obtain surveillance video stream images of multiple surveillance video points, and extract image features of operators through the surveillance video stream images; S2. Eliminate the monitoring video stream images with unclear or incomplete image features of the operator; S3. Determine the coordinates of the operator through the Beidou system or UWB system, and determine the spatial position of the operator based on the distance between the operator and multiple monitoring devices; S4. Determine the current operation process and link according to the spatial position of the operator; S5. Determine whether the operator has any risky behavior in the current operation process and link based on the operator's image features and the operator's spatial position; S6. When operators engage in risky operational behaviors in the current operational procedures and links, the smart safety helmets worn by operators or the voice broadcast system installed in the substation will automatically broadcast the current risky operational behaviors.

2. According to claim 1, a method for identifying substation operation risks based on multi-view video and high-precision positioning is characterized by: In step S1, the acquisition of monitoring video stream images of multiple monitoring video points and the extraction of operator image features through the monitoring video stream images include the following steps: S11, by installing cameras at different locations of the substation, obtaining monitoring video stream images from multiple angles; S12, video stream processing: use video processing software or framework to process the surveillance video stream and extract single frame images from it; S13, image preprocessing: preprocessing the extracted image, wherein the preprocessing includes scaling, denoising and brightness adjustment; S14. Image feature extraction: Use machine learning models to extract operator image features.

3. The method for identifying substation operation risks based on multi-view video and high-precision positioning according to claim 1 is characterized in that: In step S2, the step of removing the monitoring video stream images whose image features of the operators are unclear or incomplete comprises the following steps: S21. Use computer vision technology to identify the face, key body parts and posture of the operator's image features to determine whether the operator's features in the image are complete; S22, evaluating whether the image is clear using image quality assessment indicators, where the image quality assessment indicators include clarity, contrast, and brightness; S23. Based on the quality assessment results, images that meet the standards are screened out from the video stream, and images that do not meet the clarity standards or have incomplete staff features are removed.

4. The method for identifying substation operation risks based on multi-view video and high-precision positioning according to claim 1 is characterized in that: In step S3, the coordinates of the operator are determined by the Beidou system or the UWB system, and the spatial position of the operator is determined in combination with the distance between the operator and multiple monitoring devices, including the following steps: S31. System preparation: Equip the operators with Beidou positioning modules or UWB positioning modules, and arrange the required base stations or signal transmitters in the operation area to ensure signal coverage and stability; S32, coordinate positioning: obtain the coordinates of the operator in space through the Beidou system or UWB system; S33, distance calculation: after determining the coordinates of the operator, calculate the distance between the operator and multiple surveillance video points; S34, spatial position determination: determining the current spatial position of the operator according to the coordinates of the operator and the distances between the operator and the multiple monitoring video points; S35. Real-time monitoring and data updating: The spatial position of the operators is transmitted to the monitoring system in real time for monitoring and analysis, and the operators are tracked in real time by continuously acquiring and updating the position data.

5. The method for identifying substation operation risks based on multi-view video and high-precision positioning according to claim 1 is characterized in that: In step S4, the current operation process and link are determined according to the spatial position of the operator, specifically: the operation areas of different operation processes and links are preset, and the current operation process and link are determined according to the operation area where the spatial position of the operator is located.

6. The method for identifying substation operation risks based on multi-view video and high-precision positioning according to claim 1 is characterized in that: In step S5, the risky operations include unqualified personnel, wrong operation objects, mistakenly entering intervals, crossing safety lines, proximity to live objects, wrong or missing ground wires, wrong or missing ground switches, climbing without safety belts, climbing with unmanned escalators, staying under the boom, large appliances and equipment being too close, and not wearing safety helmets.

7. The method for identifying substation operation risks based on multi-view video and high-precision positioning according to claim 1 is characterized in that: In step S5, judging whether the operator has any operation risk behavior in the current operation process and link according to the operator image features and the operator spatial position includes the following steps: S51. Data collection: Collect large-scale image features of operators in historical periods and corresponding spatial location data samples of operators; S52. Data labeling: labeling the types of operation risk behaviors existing in the image features of the operators; S53, using the image features of the operator and the spatial position of the operator as explanatory variables and the corresponding operation risk behavior type as the response variable, training a random forest model; S54. Input the current operator image features and the operator spatial position into the random forest model to determine the operation risk behavior type of the current operation process and link.

8. The method for identifying substation operation risks based on multi-view video and high-precision positioning according to claim 7 is characterized in that: The random forest model includes the following construction steps: S531, using data fusion technology to process the image features of the operator and the spatial position data of the operator to obtain the operator behavior feature vector; S532, taking the operator behavior feature vector as input and the operation risk behavior type as target output, training a random forest model; S533. Use confusion matrix to evaluate the accuracy of random forest model in predicting the type of operational risk behavior.

9. The method for identifying substation operation risks based on multi-view video and high-precision positioning according to claim 8, characterized in that: The data fusion technology specifically includes: Data encoding is performed based on the image features of the operator to obtain the behavior code; The spatial position weight is calculated according to the spatial position of the operator, specifically: a coordinate system is established with the symmetry center of the specific working area of ​​the operator as the origin, and the spatial position weight is calculated according to the distance between the spatial position of the operator and the symmetry center; The behavior code and spatial position weight are combined to form the operator behavior feature vector.

10. The method for identifying substation operation risks based on multi-view video and high-precision positioning according to claim 1, characterized in that: In step S6, the current operation risk behavior type is automatically reported, and when the operator does not stop the operation risk behavior within the specified time, an alarm message is reported to the management personnel.

Images