An artificial intelligence power operation risk supervision and identification system and method

Abstract

The present invention discloses an artificial intelligence power operation risk supervision and identification system and method, which utilizes artificial intelligence image recognition technology to perform hierarchical identification of image data at the power operation site, and firstly performs preliminary identification to determine whether there is a high risk according to personnel identity, equipment category and scene, and determines whether to perform secondary fine identification, thereby improving identification efficiency and saving computing resources. In the secondary identification process, according to different personnel identity categories and scenes, the identification targets and adopted models are also different, thereby further controlling the calculation amount of the model, thereby performing rapid risk identification and reminder.

Description

Technical Field

[0001] The present invention relates to the field of safety management, and in particular to an artificial intelligence power operation risk supervision and identification system and method. Background Art

[0002] With the development of the power industry, the risks in power operations are also increasing. Artificial intelligence can identify potential risk factors through the analysis and processing of power operation-related data, and provide prediction and early warning functions to help power companies and relevant departments take timely measures to reduce the probability and impact of accidents. Common technologies for identifying risks in power operations in the prior art include: Sensor technology: using various sensor devices to monitor and collect data on power equipment and operation sites in real time to identify equipment failures and operation risks. However, there are the following shortcomings: Although the use of sensors to identify on-site data can accurately detect equipment, environment and other data, it is insufficient to identify the operating behavior of on-site operators. In the prior art, there are also wearable devices to collect data related to operators, but there are also problems such as inconvenience in wearing and affecting the normal operation of workers. Image recognition and video monitoring technology: The operation site situation is automatically collected through the power operation recorder. Based on image recognition and discrimination technology, the recognition and alarm of personnel behavior image data during the operation process are studied. The image recognition method can simultaneously identify the operating behavior of on-site operators and equipment-related data, and the abnormal behavior of power operations judged by artificial intelligence algorithms, including human target detection, action representation and action recognition. Behavior recognition needs to be applied to power operation scenarios. In combination with power business experts to define abnormal behavior and normal behavior, data scientists construct unstructured data to form a positive and negative sample library, which serves as an important data basis for training abnormal behavior judgment algorithms for power operations. On the basis of action representation extraction, action feature recognition is completed in the fields of spatial sequence and time series to achieve action classification through data analysis. For example, Chinese patent application CN113111728A discloses an intelligent identification method for risks of power production operations in substations. In power production operations in substations, video surveillance equipment is used to identify and track operators and identify power equipment signs near locations. However, as the number of identified targets increases, the computational cost to ensure recognition accuracy also increases. Summary of the invention

[0003] In order to solve the above problems, the purpose of the present invention is to provide an artificial intelligence power operation risk supervision and identification system and method.

[0004] In order to achieve the above objectives, the following technical solutions are provided:

[0005] An artificial intelligence power operation risk supervision identification method, wherein:

[0006] Step S10: Collecting real-time image data of the current power operation site, and preliminarily extracting the characteristics of personnel, equipment and scenes at the operation site;

[0007] Step S20: judging the identity category of the personnel at the work site according to the personnel characteristics to obtain the corresponding first personnel risk level, and obtaining the corresponding environmental risk level according to the equipment characteristics and the scene characteristics; when the first personnel risk level is lower than the environmental risk level, directly proceeding to step S40, otherwise proceeding to step S30;

[0008] Step S30: performing secondary extraction of the personnel features and the equipment features according to the personnel identity category and the scene features; wherein the secondary extracted personnel features and the equipment features are different according to different personnel identity categories and scene features; determining the corresponding second personnel risk level according to the secondary extracted personnel features and equipment features; when the second personnel risk level is higher than a threshold, proceeding to step S40, otherwise returning to step S10;

[0009] Step S40: Providing high-risk warnings to corresponding personnel.

[0010] As a further improvement of the present invention,

[0011] In the step S10, the initially extracted on-site personnel features include helmet color, work clothes type features and text features, and the equipment features include equipment categories;

[0012] In step S20, the personnel identity categories include outsiders, managers, patrol personnel and maintenance personnel;

[0013] In the step S30, the secondarily extracted personnel features include personnel skeleton features and safety equipment features, and the secondarily extracted equipment features include equipment opening and closing features.

[0014] As a further improvement of the present invention,

[0015] If the identity category of the person is an outsider or a manager, the second extracted personnel features include personnel skeleton features, tracking the center of gravity position of the personnel skeleton, and determining the risk level of the second person in combination with the authorization scope corresponding to the personnel identity category;

[0016] If the identity category of the person is a patrol officer and the scene is not high-altitude, the second extracted person features include the person skeleton features and the equipment opening and closing features, the person posture is identified based on the person skeleton features, and the second person risk level is determined in combination with the equipment opening and closing status;

[0017] If the personnel identity category is maintenance personnel, or the personnel identity category is patrol personnel and the scene is high-altitude, the secondary extracted personnel features include personnel skeleton features and safety equipment features. The personnel posture is identified based on the personnel skeleton features, and the risk level of the second personnel is determined in combination with the wearing of safety equipment and the opening and closing status of the equipment.

[0018] As a further improvement of the present invention,

[0019] The YOLO convolutional neural network is used to build a model to extract image features, and the number of convolution kernels is adjusted according to different situations.

[0020] As a further improvement of the present invention,

[0021] In the step S10, for preliminary feature extraction, the number of convolution kernels in the convolutional neural network model is N1;

[0022] In the step S30, for the secondary extraction of features, if the identity category of the personnel is a foreigner or a manager, the number of convolution kernels in the secondary extraction convolution neural network model is N2; if the identity category of the personnel is a patrol personnel, the number of convolution kernels in the secondary extraction convolution neural network model is N3; if the identity category of the personnel is a maintenance personnel, the number of convolution kernels in the secondary extraction convolution neural network model is N4;

[0023] Among them N1 <N2<N3<N4。

[0024] An artificial intelligence power operation risk supervision and identification system, wherein:

[0025] Information collection module: used to collect real-time image data of the current power operation site and preliminarily extract the characteristics of personnel, equipment and scenes at the operation site;

[0026] A first determination module is used to determine the identity category of the personnel at the work site according to the personnel characteristics to obtain the corresponding first personnel risk level, and obtain the corresponding environmental risk level according to the equipment characteristics and the scene characteristics; when the first personnel risk level is lower than the environmental risk level, a high-risk prompt is given to the corresponding personnel;

[0027] Second determination module: performing secondary extraction of the personnel features and the equipment features according to the personnel identity category and the scene features; wherein the secondary extracted personnel features and the equipment features are different according to different personnel identity categories and scene features; judging the corresponding second personnel risk level according to the secondary extracted personnel features and equipment features; and providing a high risk prompt to the corresponding personnel when the second personnel risk level is higher than a threshold;

[0028] Risk warning module: used to provide high-risk warnings to relevant personnel.

[0029] As a further improvement of the present invention,

[0030] In the information collection module, the initially extracted on-site personnel features include helmet color, work clothes type features and text features, and the equipment features include equipment category;

[0031] In the first determination module, the personnel identity categories include outsiders, managers, patrol personnel and maintenance personnel;

[0032] In the second determination module, the second extracted personnel features include personnel skeleton features and safety equipment features, and the second extracted equipment features include equipment opening and closing features.

[0033] As a further improvement of the present invention,

[0034] The second determination module also includes:

[0035] If the identity category of the person is an outsider or a manager, the second extracted personnel features include personnel skeleton features, tracking the center of gravity position of the personnel skeleton, and determining the risk level of the second person in combination with the authorization scope corresponding to the personnel identity category;

[0036] If the identity category of the person is a patrol officer and the scene is not at high altitude, the second extracted person features include the person skeleton features and the equipment opening and closing features, the person posture is identified based on the person skeleton features, and the second person risk level is determined in combination with the equipment opening and closing status;

[0037] If the personnel identity category is maintenance personnel, or the personnel identity category is patrol personnel and the scene is high-altitude, the secondary extracted personnel features include personnel skeleton features and safety equipment features. The personnel posture is identified based on the personnel skeleton features, and the risk level of the second personnel is determined in combination with the wearing of safety equipment and the opening and closing status of the equipment.

[0038] As a further improvement of the present invention,

[0039] The YOLO convolutional neural network is used to build a model to extract image features, and the number of convolution kernels is adjusted according to different situations.

[0040] As a further improvement of the present invention,

[0041] In the information acquisition module, for preliminary feature extraction, the number of convolution kernels in the convolutional neural network model is N1;

[0042] In the second determination module, for the secondary extraction of features, if the identity category of the personnel is a foreigner or a manager, the number of convolution kernels in the secondary extraction convolution neural network model is N2; if the identity category of the personnel is a patrol personnel, the number of convolution kernels in the secondary extraction convolution neural network model is N3; if the identity category of the personnel is a maintenance personnel, the number of convolution kernels in the secondary extraction convolution neural network model is N4;

[0043] Among them N1 <N2<N3<N4。

[0044] The beneficial effects of the present invention are:

[0045] 1. Using artificial intelligence technology, the image data of the power operation site is graded and identified. The initial identification only involves personnel identity, equipment category and scene, and the model calculation is low. According to different personnel identity categories and scene characteristics, the secondary identification involves more detailed personnel posture, safety equipment, equipment opening and closing identification, etc., and the model calculation is high. This hierarchical identification method is used to improve image processing efficiency.

[0046] 2. In the secondary recognition process, different models are used according to different personnel identity categories and scenarios. This difference is not only reflected in the difference in recognition targets, but also in the different number of convolution kernels, thereby further controlling the computational complexity of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and should not be regarded as limiting the scope.

[0048] Figure 1 It is a flow chart of an artificial intelligence power operation risk supervision and identification method of the present invention;

[0049] Figure 2 It is a structural diagram of an artificial intelligence power operation risk supervision and identification system of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention. Embodiment 1

[0051] Figure 1 This is a flow chart of an artificial intelligence power operation risk supervision and identification method of the present invention. Figure 1 As shown, an artificial intelligence power operation risk supervision and identification method provided by Embodiment 1 of the present invention includes:

[0052] Step S10: Collect real-time image data of the current power operation site and preliminarily extract the characteristics of personnel, equipment and scene at the operation site.

[0053] Preferably, the initially extracted work site personnel features include helmet color, work clothes type features and text features, and the equipment features include equipment categories.

[0054] In the power system, the meaning of the color of the hard hat is usually as follows: White hard hat: usually indicates management personnel, supervisors or visitors. They may be responsible for supervising, managing or inspecting the operation of the power system. Blue hard hat: usually indicates technicians or engineers. They may be professionals responsible for the maintenance, installation or technical support of power equipment. Yellow hard hat: usually indicates power system operators. They are personnel directly involved in the operation, inspection or construction of the power system. Red hard hat: usually indicates high-voltage electricians or power maintenance personnel. They are personnel who are specifically responsible for the maintenance, repair or construction of high-voltage power equipment and need to have special skills and knowledge. Similarly, the color of work clothes is also related to the different identities of the personnel. In addition, some work clothes will have relevant text features, such as job title, name and work number, etc., combined with the color of the hard hat, the type of work clothes and the text features, the identity category of the personnel can be comprehensively determined.

[0055] Step S20: judging the identity category of the personnel at the work site according to the personnel characteristics to obtain the corresponding first personnel risk level, and obtaining the corresponding environmental risk level according to the equipment characteristics and the scene characteristics; when the first personnel risk level is lower than the environmental risk level, directly proceeding to step S40, otherwise proceeding to step S30;

[0056] Obviously, different personnel have different identities and responsibilities, and their risk levels are also different. In the present invention, the identity categories of personnel are specifically divided into outsiders, managers, inspectors, and maintenance personnel. In addition, the risks brought by different environments are also different. For example, high-voltage equipment and high-altitude areas belong to high-risk environments, while control rooms, auxiliary equipment, and safety areas belong to low-risk environments.

[0057] Preferably, when the risk level of a person is low risk, but the risk level of the environment in which he is located is high risk, a risk warning is directly issued to avoid misoperation of risky equipment, accidental entry into risky areas, etc., and because this step only involves preliminary image recognition, it saves computing resources while achieving the effect of risk warning in certain situations.

[0058] Step S30: performing secondary extraction of the personnel features and the equipment features according to the personnel identity category and the scene features; wherein the secondary extracted personnel features and the equipment features are different according to different personnel identity categories and scene features; determining the corresponding second personnel risk level according to the secondary extracted personnel features and equipment features; when the second personnel risk level is higher than a threshold, proceeding to step S40, otherwise returning to step S10;

[0059] Preferably, the secondarily extracted personnel features include personnel skeleton features and safety equipment features, and the secondarily extracted equipment features include equipment opening and closing features.

[0060] Preferably, if the identity category of the person is a foreigner or a manager, the second extracted personnel features include personnel skeleton features, the center of gravity position of the personnel skeleton is tracked, and the risk level of the second person is determined in combination with the authorization scope corresponding to the personnel identity category;

[0061] Preferably, if the identity category of the person is a patrol person and the scene is not at high altitude, the second extracted person features include person skeleton features and equipment opening and closing features, the person posture is identified according to the person skeleton features, and the second person risk level is determined in combination with the equipment opening and closing state;

[0062] Preferably, if the personnel identity category is maintenance personnel, or the personnel identity category is patrol personnel and it is a high-altitude scene, the secondary extracted personnel features include personnel skeleton features and safety equipment features. The personnel posture is identified based on the personnel skeleton features, and the risk level of the second personnel is determined in combination with the wearing of safety equipment and the opening and closing status of the equipment.

[0063] Posture recognition is to judge the posture or action of the human body by analyzing the characteristics of the human skeleton. In the field of computer vision and machine learning, posture recognition usually uses deep learning models and computer vision algorithms to extract and analyze the skeleton characteristics of the human body. For example, the posture of the human body is judged by joint angles, joint positions, joint movement trajectories, joint speeds and accelerations, joint connection relationships, etc. Human posture recognition is used to judge whether the staff has followed the safety operating procedures and correctly worn safety protection devices. For example, approaching or touching high-voltage lines, operating high-risk equipment, not using safety belts or tools correctly when climbing electric poles, and leaning too much when working at heights.

[0064] In the process of identifying the posture of personnel, safety equipment, and the opening and closing of equipment, as the number of identification targets increases, the recognition accuracy requirements for the model also increase accordingly. Preferably, the depth of the neural network can be increased to improve the recognition accuracy. Increasing the number of convolution kernels can increase the representation ability and learning ability of the model, because more convolution kernels can capture more feature information, but at the same time, it will also increase the calculation amount of the model. Therefore, in the secondary recognition process, according to different personnel identity categories and scenes, such as for outsiders or managers, non-high-altitude scene patrol personnel, maintenance personnel or high-altitude scene patrol personnel, their risk levels are obviously different and gradually increase. For example, for outsiders or managers, it is only necessary to pay attention to whether they exceed the safe area or the authorized area. There are few types of targets to be identified, and the recognition accuracy requirements for the model are not high. However, for maintenance personnel or high-altitude scene patrol personnel, it is necessary to pay attention to personnel posture, safety protection equipment, the opening and closing status of high-voltage equipment, etc. There are many types of targets that need to be identified, and the recognition accuracy requirements for the model are high. Therefore, different models are used according to different personnel identity categories and scenarios. This difference is not only reflected in the difference in recognition targets, but also in the different number of convolution kernels, thereby further controlling the computational complexity of the model.

[0065] Preferably, a YOLO convolutional neural network is used to build a model to extract image features, and the number of convolution kernels is adjusted according to different situations.

[0066] Preferably, for the preliminary feature extraction, the number of convolution kernels in the convolution neural network model is N1; for the secondary feature extraction, if the identity category of the personnel is outsiders or management personnel, the number of convolution kernels in the secondary extraction convolution neural network model is N2; if the identity category of the personnel is patrol personnel, the number of convolution kernels in the secondary extraction convolution neural network model is N3; if the identity category of the personnel is maintenance personnel, the number of convolution kernels in the secondary extraction convolution neural network model is N4; wherein N1 <N2<N3<N4。

[0067] Step S40: Providing high-risk warnings to corresponding personnel. Embodiment 2

[0068] The second embodiment of the present invention provides an artificial intelligence power operation risk supervision and identification system, which is used to implement the artificial intelligence power operation risk supervision and identification method described in the first embodiment, wherein:

[0069] Information collection module: used to collect real-time image data of the current power operation site and preliminarily extract the characteristics of personnel, equipment and scenes at the operation site;

[0070] A first determination module is used to determine the identity category of the personnel at the work site according to the personnel characteristics to obtain the corresponding first personnel risk level, and obtain the corresponding environmental risk level according to the equipment characteristics and the scene characteristics; when the first personnel risk level is lower than the environmental risk level, a high-risk prompt is given to the corresponding personnel;

[0071] Second determination module: performing secondary extraction of the personnel features and the equipment features according to the personnel identity category and the scene features; wherein the secondary extracted personnel features and the equipment features are different according to different personnel identity categories and scene features; judging the corresponding second personnel risk level according to the secondary extracted personnel features and equipment features; and providing a high risk prompt to the corresponding personnel when the second personnel risk level is higher than a threshold;

[0072] Risk warning module: used to provide high-risk warnings to relevant personnel.

[0073] Preferably, in the information collection module, the initially extracted on-site personnel features include helmet color, work clothes type features and text features, and the equipment features include equipment category;

[0074] In the first determination module, the personnel identity categories include outsiders, managers, patrol personnel and maintenance personnel;

[0075] In the second determination module, the second extracted personnel features include personnel skeleton features and safety equipment features, and the second extracted equipment features include equipment opening and closing features.

[0076] Preferably, the second determination module further includes:

[0077] If the identity category of the person is an outsider or a manager, the second extracted personnel features include personnel skeleton features, tracking the center of gravity position of the personnel skeleton, and determining the risk level of the second person in combination with the authorization scope corresponding to the personnel identity category;

[0078] If the identity category of the person is a patrol officer and the scene is not at high altitude, the second extracted person features include the person skeleton features and the equipment opening and closing features, the person posture is identified based on the person skeleton features, and the second person risk level is determined in combination with the equipment opening and closing status;

[0079] If the personnel identity category is maintenance personnel, or the personnel identity category is patrol personnel and the scene is high-altitude, the secondary extracted personnel features include personnel skeleton features and safety equipment features. The personnel posture is identified based on the personnel skeleton features, and the risk level of the second personnel is determined in combination with the wearing of safety equipment and the opening and closing status of the equipment.

[0080] Preferably, a YOLO convolutional neural network is used to build a model to extract image features, and the number of convolution kernels is adjusted according to different situations.

[0081] Preferably, in the information acquisition module, for preliminary feature extraction, the number of convolution kernels in the convolutional neural network model is N1;

[0082] In the second determination module, for the secondary extraction of features, if the identity category of the personnel is a foreigner or a manager, the number of convolution kernels in the secondary extraction convolution neural network model is N2; if the identity category of the personnel is a patrol personnel, the number of convolution kernels in the secondary extraction convolution neural network model is N3; if the identity category of the personnel is a maintenance personnel, the number of convolution kernels in the secondary extraction convolution neural network model is N4;

[0083] Among them N1 <N2<N3<N4。

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An artificial intelligence power operation risk supervision and identification method, characterized by: Step S10: Collecting real-time image data of the current power operation site, and preliminarily extracting the characteristics of personnel, equipment and scenes at the operation site; Step S20: determining the identity category of the personnel at the work site according to the personnel characteristics to obtain a corresponding first personnel risk level, and obtaining a corresponding environmental risk level according to the equipment characteristics and the scene characteristics; When the first personnel risk level is lower than the environmental risk level, directly proceed to step S40, otherwise proceed to step S30; Step S30: performing secondary extraction of the personnel features and the equipment features according to the personnel identity category and the scene features; wherein the secondary extracted personnel features and the equipment features are different according to different personnel identity categories and scene features; determining the corresponding second personnel risk level according to the secondary extracted personnel features and equipment features; when the second personnel risk level is higher than a threshold, proceeding to step S40, otherwise returning to step S10; Step S40: Prompt high-risk information to corresponding personnel; In the step S10, the initially extracted on-site personnel features include helmet color, work clothes type features and text features, and the equipment features include equipment categories; In step S20, the personnel identity categories include outsiders, managers, patrol personnel and maintenance personnel; In the step S30, the secondarily extracted personnel features include personnel skeleton features and safety equipment features, and the secondarily extracted equipment features include equipment opening and closing features; The step S30 further includes: If the identity category of the person is an outsider or a manager, the second extracted personnel features include personnel skeleton features, tracking the center of gravity position of the personnel skeleton, and determining the risk level of the second person in combination with the authorization scope corresponding to the personnel identity category; If the identity category of the person is a patrol officer and the scene is not at high altitude, the second extracted person features include the person skeleton features and the equipment opening and closing features, the person posture is identified based on the person skeleton features, and the second person risk level is determined in combination with the equipment opening and closing status; If the identity category of the person is a maintenance personnel, or the identity category of the person is a patrol personnel and the scene is high-altitude, the second extracted personnel features include personnel skeleton features and safety equipment features. The personnel posture is identified based on the personnel skeleton features, and the risk level of the second person is determined in combination with the wearing of safety equipment and the opening and closing status of the equipment; Among them, recognizing a person's posture based on the person's skeleton features includes: The posture of the human body is judged by the joint angle, joint position, joint motion trajectory, joint speed and acceleration, and joint connection relationship; The YOLO convolutional neural network is used to build a model to extract image features, and the number of convolution kernels is adjusted according to different situations; In the step S10, for preliminary feature extraction, the number of convolution kernels in the convolutional neural network model is N1; In the step S30, for the secondary extraction of features, if the identity category of the personnel is a foreigner or a manager, the number of convolution kernels in the secondary extraction convolution neural network model is N2; if the identity category of the personnel is a patrol personnel, the number of convolution kernels in the secondary extraction convolution neural network model is N3; if the identity category of the personnel is a maintenance personnel, the number of convolution kernels in the secondary extraction convolution neural network model is N4; Among them N1 <N2<N3<N4。 2. An artificial intelligence power operation risk supervision and identification system, used to implement the method described in claim 1, characterized in that: include: Information collection module: used to collect real-time image data of the current power operation site and preliminarily extract the characteristics of personnel, equipment and scenes at the operation site; A first determination module: used to determine the identity category of the personnel at the work site according to the personnel characteristics to obtain the corresponding first personnel risk level, and to obtain the corresponding environmental risk level according to the equipment characteristics and the scene characteristics; When the first personnel risk level is lower than the environmental risk level, a high-risk warning is given to the corresponding personnel; Second determination module: performing secondary extraction of the personnel features and the equipment features according to the personnel identity category and the scene features; wherein the secondary extracted personnel features and the equipment features are different according to different personnel identity categories and scene features; judging the corresponding second personnel risk level according to the secondary extracted personnel features and equipment features; and providing a high risk prompt to the corresponding personnel when the second personnel risk level is higher than a threshold; Risk warning module: used to provide high-risk warnings to relevant personnel; In the information collection module, the initially extracted on-site personnel features include helmet color, work clothes type features and text features, and the equipment features include equipment category; In the first determination module, the personnel identity categories include outsiders, managers, patrol personnel and maintenance personnel; In the second determination module, the second extracted personnel features include personnel skeleton features and safety equipment features, and the second extracted equipment features include equipment opening and closing features; The second determination module also includes: If the identity category of the person is an outsider or a manager, the second extracted personnel features include personnel skeleton features, tracking the center of gravity position of the personnel skeleton, and determining the risk level of the second person in combination with the authorization scope corresponding to the personnel identity category; If the identity category of the person is a patrol officer and the scene is not at high altitude, the second extracted person features include the person skeleton features and the equipment opening and closing features, the person posture is identified based on the person skeleton features, and the second person risk level is determined in combination with the equipment opening and closing status; If the identity category of the person is a maintenance personnel, or the identity category of the person is a patrol personnel and the scene is high-altitude, the second extracted personnel features include personnel skeleton features and safety equipment features. The personnel posture is identified based on the personnel skeleton features, and the risk level of the second person is determined in combination with the wearing of safety equipment and the opening and closing status of the equipment; Among them, recognizing a person's posture based on the person's skeleton features includes: The posture of the human body is judged by the joint angle, joint position, joint motion trajectory, joint speed and acceleration, and joint connection relationship; The YOLO convolutional neural network is used to build a model to extract image features, and the number of convolution kernels is adjusted according to different situations; In the information acquisition module, for preliminary feature extraction, the number of convolution kernels in the convolutional neural network model is N1; In the second determination module, for the secondary extraction of features, if the identity category of the personnel is a foreigner or a manager, the number of convolution kernels in the secondary extraction convolution neural network model is N2; if the identity category of the personnel is a patrol personnel, the number of convolution kernels in the secondary extraction convolution neural network model is N3; if the identity category of the personnel is a maintenance personnel, the number of convolution kernels in the secondary extraction convolution neural network model is N4; Among them N1 <N2<N3<N4。

Images