CN117173791A - Distribution network constructor violation detection method and system based on action recognition - Google Patents
Distribution network constructor violation detection method and system based on action recognition Download PDFInfo
- Publication number
- CN117173791A CN117173791A CN202311260587.5A CN202311260587A CN117173791A CN 117173791 A CN117173791 A CN 117173791A CN 202311260587 A CN202311260587 A CN 202311260587A CN 117173791 A CN117173791 A CN 117173791A
- Authority
- CN
- China
- Prior art keywords
- safety protection
- image
- frame
- protection equipment
- information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention belongs to the field of computer vision and engineering application, and the method comprises the steps of carrying out graph acquisition by a visible light acquisition system, establishing a corresponding database for matching and linking, receiving transmission data of video monitoring equipment, judging the operation behavior type of an operator by utilizing a behavior information judging model based on skeleton information, and outputting the operation behavior type to a background management system; the background management system judges a safety protection equipment library of a corresponding type, outputs safety protection equipment information to be detected, and detects the information and the operation behavior image through the target detection module; and comparing the anchor frame position information, performing coincidence calculation, and judging whether an operator violates rules or carries safety protection equipment according to regulations. The invention provides a multi-person detection method for realizing construction sites through a target detection and behavior estimation network, which is used for detecting different behavior types and detecting safety protection equipment and effectively detecting targeted safety protection equipment by combining the behavior estimation and the target detection.
Description
Technical Field
The invention belongs to the fields of computer vision and engineering application, and particularly relates to a method for detecting violations of network allocation constructors based on action recognition.
Background
The intelligent monitoring technology and the terminal equipment reflect some problems in the current ground application practice of the on-site operation monitoring of the power distribution network, and mainly comprise the following steps: (1) the system relies heavily on manual monitoring and is inefficient. The existing safety protection device relies on manpower to detect in a large amount, so that the detection effect is poor; (2) Real-time monitoring is difficult to realize, and comprehensive real-time monitoring is difficult to carry out on a multi-person construction scene; (3) The problem that the safety protection equipment is not worn according to the specification exists in the construction process of operators, and the background manager is difficult to detect in real time. With the development of deep learning neural network technology, a real-time high-precision early warning system based on visual driving plays an increasingly important role. The real-time high-precision intelligent detection technology based on AI vision is researched, so that the efficiency and the efficiency of the early warning system are improved. The visual perception capability of the early warning system for the distribution network operation scene is an important core capability of the system, and the capability is based on understanding of the operation scene, and the visual-based video target detection technology is a key technology for realizing the understanding of the operation scene.
The video object detection technique is a technique for solving the problem of how to locate and identify objects appearing in each video, and video detection has a characteristic of high redundancy compared with image object detection. The neural network proposed automatic learning of advanced features in multimedia data since the 2015 s. From this, the target detection based on the deep learning neural network becomes a popular research direction, and the video target detection is closer to the task requirement under the real scene, such as video monitoring, object identification and the like under the power distribution operation scene. With the development of deep learning, video object detection has now enabled the ability to detect objects.
The behavior recognition method is a method for realizing behavior judgment through the key point positions of human bodies in videos or images, and can be used for solving the problem that the operation behaviors are difficult to judge quickly. The method is mainly based on a three-dimensional convolution method, a double-flow structure method and a skeleton key point method, and a skeleton key point-based research method is provided by DU (digital television) and the like at the earliest, is a behavior gesture detection and recognition method with highest accuracy at present, is not easy to be influenced by illumination change, and is suitable for being applied to a construction operation environment. The method can also realize rapid detection of a plurality of different detection targets at the same time, and is beneficial to realizing real-time detection of behaviors of a plurality of constructors in a construction site.
Disclosure of Invention
The present invention has been made in view of the above-described problems occurring in the prior art. The invention discloses a violation detection system suitable for identifying behaviors of operators on a distribution network non-fixed-point operation site, which is realized based on a skeleton behavior identification technology and a target detection technology, wherein the construction operation behavior type of the operators is judged through skeleton characteristic information, the corresponding construction operation behavior type is transmitted to a background management system, the background management system outputs the type of safety protection equipment required to be worn by the operators according to the corresponding operation type, and the corresponding type of safety protection equipment required to be detected is transmitted to the target detection system, so that whether the safety protection equipment of the operators performing operation is illegal or not is realized.
Therefore, the method for detecting the violations of the distribution network constructors based on the action recognition is provided.
In order to solve the technical problems, the invention provides a method for detecting violations of network allocation constructors based on action recognition, which comprises the following steps:
the visible light acquisition system performs graph acquisition, establishes a corresponding database for matching and linking, receives transmission data of the video monitoring equipment, judges the operation behavior type of an operator by utilizing the behavior information judging model based on skeleton information, and outputs the operation behavior type to the background management system; the background management system judges a safety protection equipment library of a corresponding type, outputs safety protection equipment information to be detected, and detects the information and the operation behavior image through the target detection module; and comparing the anchor frame position information, performing coincidence calculation, and judging whether an operator violates rules or carries safety protection equipment according to regulations.
As a preferable scheme of the method for detecting the violations of the distribution network constructors based on the action recognition, the invention comprises the following steps: the graphic acquisition comprises the steps of establishing a visible light acquisition system at a construction site, setting an image acquisition device of a distribution control ball and a visible light camera, configuring video monitoring equipment at the distribution construction site, acquiring videos, intercepting images at a fixed frequency, and transmitting the images to a background management system.
The corresponding databases comprise a logic matching base for establishing different types of operation behavior databases, setting operation behaviors and safety protection equipment, receiving transmission data of the video monitoring equipment, matching and linking the behavior type databases and the safety protection equipment databases in a background management system according to priori knowledge, wherein each type of behavior corresponds to a specific safety protection equipment database.
The behavior information discrimination model adopts an alpha phase framework to realize multi-person gesture recognition estimation and comprises a symmetrical space transformation network, a non-maximum suppression module and a behavior guiding suggestion generator.
As a preferable scheme of the method for detecting the violations of the distribution network constructors based on the action recognition, the invention comprises the following steps: the target detection module comprises detection of safety protection equipment by using target detection to find all operators at a construction site and utilizing rotary target detection.
The method comprises the steps that all operators in a construction site are detected by using target detection, the operators in the image are detected by using a Yolov5s algorithm, central pixel position information of the operators is obtained according to four vertex pixel position information of a detected anchor frame, preprocessing is carried out when the image enters the content of the Yolov5s algorithm, the quality of the image is improved through Mosaic data, then, image features are extracted by using a deep convolutional neural network and are realized by adopting a residual convolutional group with the size of 3x3, then, the size of a feature image is restored to the size of an original image through a pyramid pooling module, and finally, the target detection result image with the anchor frame information of the operators is marked and output by using the anchor frame for a feature object through an output end.
As a preferable scheme of the method for detecting the violations of the distribution network constructors based on the action recognition, the invention comprises the following steps: the method for extracting skeleton key points by adopting the space-time diagram convolutional neural network comprises the steps of setting 14 skeleton key points in a human body, constructing a connection relation of the skeleton key points of the human body, and constructing a space-time diagram according to video information and the connection relation;
the space-time diagram comprises defining a node t in a t-th frame image in the video as v ti Node i in the t+r frame image is defined as v (t+1)i Let v ti ＝(x 1 ，y 1 ,c 1 )，v (t+1)i ＝(x 2 ,y 2 ,c 2 ) Then the vector A of the same node in the T frame image can be expressed asThe method comprises the steps that the vector information of 14 skeleton key points is combined by blue line connection of the skeleton key points between adjacent frames to obtain a key point time diagram, image characteristic information of the key points is obtained by adopting diagram convolution, and a node v is obtained ti Is expressed as:
the spatial map convolution includes defining a hierarchical update rule by feature B and map structure G: the method comprises the steps of adopting a space configuration division strategy to divide a field set of skeleton information points into three subsets, wherein the first subset is skeleton nodes, the second subset is an adjacent node set which is closer to the center of gravity of the whole skeleton than the nodes in space position, and the third subset is an adjacent node set which is farther from the center than the nodes in space position, and the third subset is expressed as:
wherein f represents feature map information, v tj Nodes j, S (v) representing the t-th frame in the space diagram ti ) Representing v ti W represents a weighting function, l in Representing the mapping function, Z ij Representing regularization term Z ij (v ti )＝|{v tk |l ti (v tk )＝l ti (v ij ) Corresponding subset technique in } |l ti (v ti ) Is v ti Mapping under a single frame, d i Is the average distance from the key point i to the center of gravity of all frames in the training set, d j Is node v tj Distance to the center of gravity.
As a preferable scheme of the method for detecting the violations of the distribution network constructors based on the action recognition, the invention comprises the following steps: the multi-person gesture recognition estimation comprises the steps of setting hand and head characteristics as interesting characteristics in a scene of a construction operation site, combining the characteristics by adopting a space-time convolution module to realize behavior characteristic extraction of a video, wherein TCN in a network represents time convolution operation, GCN+TCN in the network is realized by using a residual error network, and the calculation process of graph convolution is as follows:
∑ j A j ＝A+I
wherein f in For inputting a feature map, A is an adjacency matrix for key point connection in a t-th frame image, I is an identity matrix for self-linking, M is a learnable weight matrix for node importance among different bones, W is a weight matrix for map convolution,representing a normalized diagonal matrix.
In the video action recognition, the whole action judgment is completed by utilizing a double-flow structure: the method comprises the steps of realizing two inputs of a space flow and a time flow by using the space-time convolution network, extracting a connection diagram of each skeleton key point in a single frame image by the space flow, enabling the time flow to represent a position change connection diagram of the single skeleton key point in a period of time, superposing and fusing coefficients of probability distribution layers of the two after the space flow and the time flow are connected, superposing the softmax of the space flow and the softmax of the time flow in the space-time convolution network to obtain a final fusion score, and sequencing the fusion scores.
And according to the content, performing skeleton key point measurement and calculation to judge whether a hand lifting action exists, if the hand lifting action does not exist, performing operation by a real person, if the hand lifting action exists, performing operation by the real person, and when the score of one action type is obviously higher than that of other action types according to the fusion score sequencing aiming at the real person in operation, determining that the highest score of the action type is the highest probability of the action, and when the scores of the action types are not different, judging the probability of the action by combining the video content and the construction environment.
As a preferable scheme of the method for detecting the violations of the distribution network constructors based on the action recognition, the invention comprises the following steps: the detection of the safety protection equipment by using the rotating target detection comprises the following steps of using a target detection algorithm RTMDet with a rotating frame to realize the target detection output of the safety protection equipment, wherein the information of the anchor frame position of the operator in the operation behavior image is specifically as follows: and adding a rotation detection head in the regression branch of the Yolov5, firstly adding 1x1 convolution in the regression branch to realize angle prediction, secondly modifying a bounding box modifier, and finally replacing the original GIOU loss function by using a rotation detection loss function.
The rotation detection loss function comprises the steps that a rotation marking frame of a detection diagram uses five parameters to represent a central coordinate point, length, width and deflection angle of the (x, y, w, h, theta) marking frame, and the (x * ，y * ，w * ,h * θ) represents a central coordinate point, length, width, and deflection angle of the prediction frame expressed as:
the calculation of θ is performed by means of direct regression and indirect regression, wherein the direct regression is expressed as F (Δθ) using a model comprising direct predicted angular offset:
the indirect regression includes matching two target vectors of a model prediction frame and a labeling frame, denoted as F (sin theta, cos theta), expressed as:
the calculation process continues to normalize, and according to the calculation process of indirect regression, a formula is set as a rotation loss function, and the function formula is expressed as:
wherein N is pos Refers to the number of all correct anchor frames, b n Marking the nth prediction frame, gt n Refers to an nth real annotation frame a 1 Is a super-ginseng, defaults to {0,01;1, n represents the number of dimensions, t i 、t' i Vector labels representing prediction and labeling frames, Z is the product of the rotation matrix and eigenvalue diagonal matrix of the image and the rotation transpose matrix, V B And (Z) calculating the volume of the corresponding rotation frame based on Kalman filtering to obtain a new Gaussian distribution, and solving the pixel coincidence probability distribution of the rotation labeling frame and the rotation prediction frame.
As a preferable scheme of the method for detecting the violations of the distribution network constructors based on the action recognition, the invention comprises the following steps: the calculating of the contact ratio comprises the steps of comparing the position information of the anchor frame of the safety protection device in the target detection image with the position information of the anchor frame of the operator in the operation behavior image output by the operation detection module, and judging the distance between the position information of the operator in each operation behavior type and the position information of the safety protection device through the calculation of the contact ratio of the anchor frame of the safety protection device and the position information of the operator in each operation behavior type: when the contact ratio of the anchor frame of the operator and the anchor frame of one of the safety protection devices is found to be 0< X <20%, an alarm is sent out, a background management system is prompted to indicate that the operator who is engaged in which operation acts has violations, the anchor frame contact ratio is repeatedly stopped until the contact ratio is more than 20%, whether other people violate the rules is checked at the same time, the synchronous check of multiple people is realized, and when the contact ratio X of the anchor frame of the operator and the anchor frame of one of the safety protection devices is more than or equal to 20%, the operator is considered to carry the safety protection devices, and whether other operators violate the rules is continuously checked.
The invention further aims to provide a system of the method for detecting the violations of the distribution network constructors based on the action recognition, which can effectively recognize the violations of the operation behaviors by monitoring and analyzing the actions of the constructors on the construction site in real time, thereby ensuring the safety of the construction process and improving the construction quality and efficiency.
The system for detecting the violations of the construction personnel of the distribution network based on the action recognition is characterized by comprising an image acquisition module, a data receiving module, a target detection module, an operation behavior recognition module, a background scheduling module and an operation safety protection equipment detection module.
The image acquisition module is used for completing image acquisition by a visible light acquisition system, and the whole monitoring of the construction site is realized by arranging a control ball visible light image acquisition device in the construction site.
And the data receiving module is used for receiving the transmission data of the video monitoring equipment after establishing the operation behavior database and the corresponding safety protection equipment library.
And the target detection module is used for detecting the related safety protection equipment type object to obtain the pixel position information of the safety protection equipment object.
The operation behavior recognition module extracts operation behaviors and specific position information of a plurality of operators on an operation construction site, takes an operation behavior image as output, and outputs a result to the background management system.
And the background dispatching module is used for judging different types of safety protection equipment libraries according to the input information and the equipment database by the background management system and outputting the safety protection equipment information to be detected.
And the operation safety protection equipment detection module obtains operation behavior types of all operators on a construction site after the behavior prediction is completed, and obtains detection types of the operation safety protection equipment according to the linkage of the operation behaviors in the background scheduling module and the operation safety protection equipment library.
A computer device comprising a memory and a processor, said memory storing a computer program, characterized in that said processor, when executing said computer program, implements the steps of a method for detecting violations of a distribution network constructor based on action recognition.
A computer readable storage medium having stored thereon a computer program, characterized in that the computer program when executed by a processor implements the steps of a method for detecting a violation of a distribution network constructor based on action recognition.
The invention has the beneficial effects that: the utility model provides a distribution network operation personnel violation detection system based on operation action, mainly aims at whether the operation personnel is according to operation action regulation, wears relevant safety protection equipment problem. The system can effectively improve the supervision efficiency of the construction operation site. When the system runs, the operation behavior can be judged through the key points of skeleton information, all operators on a construction site are found through target detection, and the detection of safety protection equipment is realized through rotating target detection, so that the operation environment of the operators tends to be safe to a greater extent, and the personal safety of the operators is ensured. The method can effectively maintain the stability of the system operation by executing the steps, so that operators in the operation of the distribution network can timely find and warn against illegal and unworn safety protection equipment
Drawings
For a clearer description of the technical solutions of embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art, wherein:
fig. 1 is a flow chart of a method for detecting violations of a distribution network constructor based on action recognition according to an embodiment of the present invention.
Fig. 2 is a flowchart of an implementation of a method for detecting violations of a distribution network constructor based on action recognition according to an embodiment of the present invention.
Fig. 3 is an alphapore structure diagram of a method for detecting violations of a distribution network constructor based on action recognition according to an embodiment of the present invention.
Fig. 4 is a block diagram of human skeleton key points of a method for detecting violations of distribution network constructors based on motion recognition according to an embodiment of the present invention.
Fig. 5 is a key point time chart of a method for detecting violations of network allocation constructors based on action recognition according to an embodiment of the present invention.
Fig. 6 is a behavior recognition deep learning network structure of a method for detecting violations of distribution network constructors based on motion recognition according to an embodiment of the present invention.
Fig. 7 is a schematic workflow diagram of a system for detecting violations of network operators based on motion recognition according to an embodiment of the present invention.
Detailed Description
So that the manner in which the above recited objects, features and advantages of the present invention can be understood in detail, a more particular description of the invention, briefly summarized above, may be had by reference to the embodiments, some of which are illustrated in the appended drawings. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways other than those described herein, and persons skilled in the art will readily appreciate that the present invention is not limited to the specific embodiments disclosed below.
Further, reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic can be included in at least one implementation of the invention. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.
While the embodiments of the present invention have been illustrated and described in detail in the drawings, the cross-sectional view of the device structure is not to scale in the general sense for ease of illustration, and the drawings are merely exemplary and should not be construed as limiting the scope of the invention. In addition, the three-dimensional dimensions of length, width and depth should be included in actual fabrication.
Also in the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "upper, lower, inner and outer", etc. are based on the orientation or positional relationship shown in the drawings, are merely for convenience of describing the present invention and simplifying the description, and do not indicate or imply that the apparatus or elements referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present invention. Furthermore, the terms "first, second, or third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
The terms "mounted, connected, and coupled" should be construed broadly in this disclosure unless otherwise specifically indicated and defined, such as: can be fixed connection, detachable connection or integral connection; it may also be a mechanical connection, an electrical connection, or a direct connection, or may be indirectly connected through an intermediate medium, or may be a communication between two elements. The specific meaning of the above terms in the present invention will be understood in specific cases by those of ordinary skill in the art.
Example 1
Referring to fig. 1-2, a first embodiment of the present invention provides a method for detecting a rule violation of a distribution network constructor based on action recognition, including:
s1: the visible light acquisition system performs graph acquisition, establishes a corresponding database for matching and linking, receives transmission data of the video monitoring equipment, judges the operation behavior type of an operator by utilizing the behavior information judging model based on skeleton information, and outputs the operation behavior type to the background management system.
Furthermore, the image acquisition equipment of the system is completed by a visible light acquisition system, and the whole monitoring of the construction site is realized by arranging the visible light image acquisition equipment such as a cloth control ball and the like in the construction site. The behavior type database and the protection equipment database are matched and linked in a background management system according to priori knowledge, and each type of behavior needs to correspond to a specific safety protection equipment database, for example, the safety protection equipment database corresponding to the ascending electricity verification behavior is an electricity verification pole, a safety helmet and a safety rope; the safety protection equipment library corresponding to the power distribution welding operation is a safety helmet and a protective suit.
It should be noted that, after the visible light acquisition system is built, the operation behavior database is built and the corresponding safety protection equipment library is built on the construction site, the transmission data of the video monitoring equipment is received, and the operation behavior type of the operator is judged by utilizing the behavior information judging model based on the skeleton information. The model can realize multi-person gesture estimation and position calibration by using an alpha phase framework, can extract the operation behaviors of a plurality of operators and the specific position information of the operators on an operation construction site through the framework, takes images with the operation behaviors as output, and outputs results to a background management system. The alpha phase framework realizes multi-person gesture recognition estimation. The framework is formed by combining a symmetrical space transformation network, a non-maximum suppression module (NMS) and a behavior guidance suggestion generator, and the method is characterized in that a traditional target detection method and a single behavior recognition algorithm are combined, so that the effect of multi-person behavior recognition is realized, and the multi-operator behavior detection on a construction site is facilitated.
It should also be noted that detection of safety protection equipment is achieved by using the rotation target detection by using the target detection to find all operators at the construction site;
the method comprises the steps that all operators in a construction site are detected by using target detection, the operators in the image are detected by using a Yolov5s algorithm, central pixel position information of the operators is obtained according to four vertex pixel position information of a detected anchor frame, preprocessing is carried out when the image enters the content of the Yolov5s algorithm, the quality of the image is improved through Mosaic data, then, image features are extracted by using a deep convolutional neural network and are realized by adopting a residual convolutional group with the size of 3x3, then, the size of a feature image is restored to the size of an original image through a pyramid pooling module, and finally, the target detection result image with the anchor frame information of the operators is marked and output by using the anchor frame for a feature object through an output end.
It should also be noted that behavior type recognition is implemented using a space-time diagram convolution module, which forms a body joint sequence by given 2D coordinates, constructing a node matrix set V { V } ti I t= … T, i= … N }, a space-time diagram g= (V, E) is constructed, where T is the number of frames and N is the number of key points. The generated two-dimensional coordinate information of the human skeleton points is (x, y), and confidence information is introduced into the two-dimensional coordinate information to obtain the key point information of the human skeleton points which is (x, y, z). Accordingly, the human skeleton key point information is set as a feature vector F (v ti ) The feature vector consists of the confidence of the key points of the t frame and the i frame in the coordinate vector and the video information. Based on the above steps, the present patent builds a space-time diagram of a bone sequence in two steps. The first step is to use the boundary relationship between frames to represent the time sequence relationship of the corresponding skeleton points of human body; and secondly, constructing a space diagram in each frame of image according to the skeleton connection relation of the human body. Let the boundary relationship between frames be E, then E is a set of edges consisting of two subsets, E S ＝{v ti v tj |(i,j)∈H}，E F ＝{v ti v (t+1)i }. The first set comprises a connection set of skeleton points in each frame of image, H is a set of skeleton points of a human body, the second subset represents connection between different frames of the same key points, extraction of the skeleton key points is achieved by setting 14 skeleton key points in a human body, and a human skeleton relation is constructedAnd constructing a space-time diagram according to the connection relation of the key points and the video information.
The space-time diagram comprises defining a node t in a t-th frame image in the video as v ti Node i in the t+r frame image is defined as v (t+1)i Let v ti ＝(x 1 ,y 1 ,c 1 )，v (t+1)i ＝(x 2 ,y 2 ,c 2 ) Then the vector A of the same node in the T frame image can be expressed asThe method comprises the steps that the vector information of 14 skeleton key points is combined by blue line connection of the skeleton key points between adjacent frames to obtain a key point time diagram, image characteristic information of the key points is obtained by adopting diagram convolution, and a node v is obtained ti Is expressed as:
the spatial map convolution includes defining a hierarchical update rule by feature B and map structure G: the method comprises the steps of adopting a space configuration division strategy to divide a field set of skeleton information points into three subsets, wherein the first subset is skeleton nodes, the second subset is an adjacent node set which is closer to the center of gravity of the whole skeleton than the nodes in space position, and the third subset is an adjacent node set which is farther from the center than the nodes in space position, and the third subset is expressed as:
wherein f represents feature map information, v tj Nodes j, S (v) representing the t-th frame in the space diagram ti ) Representing v ti W represents a weighting function, l in Representing the mapping function, Z ij Representing regularization term Z ij (v ti )＝|{v tk |l ti (v tk )＝l ti (v ij ) Corresponding subset technique in } |l ti (v ti ) Is thatv ti Mapping under a single frame, d i Is the average distance from the key point i to the center of gravity of all frames in the training set, d j Is node v tj Distance to the center of gravity.
It should be noted that, in the scene of a construction work site, the most frequently occurring work behaviors are the lifting hand and the lowering head behaviors, and the hand and head key points of a human body are rapidly spatially changed in a short time, and the contribution of the hand and the head in the behavior recognition is large, so that the two features are set as interesting features. On the basis, the space-time convolution module is combined to realize the behavior feature extraction of the video, TCN in the network represents the time convolution operation, GCN+TCN in the network is realized by a residual error network, and the calculation process of the graph convolution is as follows:
∑ j A j ＝A+I
Wherein f in For inputting a feature map, A is an adjacency matrix for key point connection in a t-th frame image, I is an identity matrix for self-linking, M is a learnable weight matrix for node importance among different bones, W is a weight matrix for map convolution,representing a normalized diagonal matrix.
In the video action recognition, the whole action judgment is completed by utilizing a double-flow structure: the method comprises the steps of realizing two inputs of a space flow and a time flow by using the space-time convolution network, extracting a connection diagram of each skeleton key point in a single frame image by the space flow, enabling the time flow to represent a position change connection diagram of the single skeleton key point in a period of time, superposing and fusing coefficients of probability distribution layers of the two after the space flow and the time flow are connected, superposing the softmax of the space flow and the softmax of the time flow in the space-time convolution network to obtain a final fusion score, and sequencing the fusion scores.
And according to the content, performing skeleton key point measurement and calculation to judge whether a hand lifting action exists, if the hand lifting action does not exist, performing operation by a real person, if the hand lifting action exists, performing operation by the real person, and when the score of one action type is obviously higher than that of other action types according to the fusion score sequencing aiming at the real person in operation, determining that the highest score of the action type is the highest probability of the action, and when the scores of the action types are not different, judging the probability of the action by combining the video content and the construction environment.
S2: the background management system judges a safety protection equipment library of a corresponding type, outputs safety protection equipment information to be detected, and detects the information and the operation behavior image through the target detection module.
Further, after the behavior prediction is completed, the operation behavior types of all operators on the construction site are obtained, the detection types of the operation safety protection equipment required to be realized are obtained according to the connection between the operation behaviors in the background management system and the operation safety protection equipment library, and the target detection of the safety protection equipment is realized by using a target detection algorithm RTMDet with a rotating frame.
The algorithm can effectively detect the inclined object, can be used for detecting targets of all safety protection devices in the image by using a depth convolution network, can use the anchor frame to carry out position calibration on the object of interest, and can obtain the position information of the central pixel point of the four vertexes of the anchor frame in the image for the anchor frame which carries out position calibration in the image. And obtaining the position information of all the safety protection devices according to the position information of the central pixel point. The principle of the algorithm for realizing the detection of the inclined object is as follows: and adding a rotation detection head in the regression branch of the Yolov5, firstly adding 1x1 convolution in the regression branch to realize angle prediction, secondly modifying a bounding box modifier, and finally replacing the original GIOU loss function by using a rotation detection loss function.
RotatingThe detection loss function is realized based on KFIOU proposed by Yang et al, the method principle is realized based on Gaussian modeling and Kalman filtering, firstly, the rotation labeling frame of the detection diagram can be expressed by five parameters, namely the center coordinate point, the length, the width and the deflection angle of the (x, y, w, h, theta) labeling frame are assumed, and the (x) is used * ,y * ,w * ,h * θ) represents a central coordinate point, length, width, and deflection angle of the prediction frame expressed as:
t x ＝(x-x 0 )/w a ,t y ＝(y-y 0 )/h a
t w ＝log(w/w a ),t h ＝log(h/h a )
and the calculation of theta is performed by adopting a direct regression mode and an indirect regression mode, wherein the direct regression is used for directly predicting the angle deviation by a model, and is expressed as F (delta theta):
the indirect regression is used for matching two target vectors of a model prediction frame and a labeling frame and is expressed as F (sin theta, cos theta) as follows:
meeting requirements according to trigonometric function definition in calculation processThus, the following calculation formula is used:
according to the calculation process of the indirect regression, it can be found that a function formula with continuous boundaries can be obtained by using the indirect regression, and the formula is set as a rotation loss function, and the function formula is expressed as follows:
the functional formula is expressed as:
wherein N is pos Refers to the number of all correct anchor frames, b n Marking the nth prediction frame, gt n Refers to an nth real annotation frame a 1 Is a super-ginseng, defaults to {0,01;1, n represents the number of dimensions, t i 、t' i Vector labels representing prediction and labeling frames, Z is the product of the rotation matrix and eigenvalue diagonal matrix of the image and the rotation transpose matrix, V B And (Z) calculating the volume of the corresponding rotation frame based on Kalman filtering to obtain a new Gaussian distribution, and solving the pixel coincidence probability distribution of the rotation labeling frame and the rotation prediction frame.
S3: and comparing the anchor frame position information, performing coincidence calculation, and judging whether an operator violates rules or carries safety protection equipment according to regulations.
Further, comparing the anchor frame position information of the safety protection equipment in the target detection image with the anchor frame position information of the operator in the operation behavior image output by the operation detection module, and judging the distance between the position information of the operator in each operation behavior type and the position information of the safety protection equipment through the coincidence ratio calculation of the anchor frame position information and the anchor frame position information of the operator in the operation behavior image: when the contact ratio of the anchor frame of the operator and the anchor frame of one of the safety protection devices is found to be 0< X <20%, an alarm is sent out, a background management system is prompted to indicate that the operator who is engaged in which operation acts has violations, the anchor frame contact ratio is repeatedly stopped until the contact ratio is more than 20%, whether other people violate the rules is checked at the same time, the synchronous check of multiple people is realized, and when the contact ratio X of the anchor frame of the operator and the anchor frame of one of the safety protection devices is more than or equal to 20%, the operator is considered to carry the safety protection devices, and whether other operators violate the rules is continuously checked.
Example 2
Referring to fig. 3-6, for one embodiment of the present invention, a method for detecting violations of a distribution network constructor based on motion recognition is provided, and in order to verify the beneficial effects of the present invention, scientific demonstration is performed through experiments.
The adopted case description is a climbing operation description under the power distribution construction operation environment, in the implementation scheme, the visible light acquisition equipment is firstly adopted to carry out video acquisition on the construction site according to the system implementation flow chart in fig. 2, and the setting method is that the visible light cameras are arranged at four vertexes of the construction site, so that the visible light cameras can observe the operation condition of the construction site. Secondly, setting the behavior type of the construction and a corresponding safety protection equipment library in a background management system through priori knowledge, and secondly, acquiring videos by a visible light acquisition device, intercepting images at a fixed frequency, and transmitting the images to the background management system. And then judging the action type of the worker according to the image requirement, and judging by adopting an alpha phase structure, wherein the judging flow is shown in figure 3.
The judging method firstly adopts a target detection algorithm to realize the pixel position information calibration of personnel through an image, secondly adopts a human skeleton key point algorithm used by the patent to judge whether the hand lifting action exists, if yes, judges that the hand lifting action exists in the working process, and adopts the key point time diagram and the dual behavior recognition deep learning network structure described in the figures 5 and 6 in the invention to realize the judging mode, wherein the dual behavior recognition deep learning network structure comprises two groups of algorithms shown in the figure 6, the algorithm structures are the same but the attention points are different, one is the attention space change is called space flow information, and the other is the attention time change is called time flow information. After the operation behaviors of the operators are obtained, the operation behavior types are required to be transmitted to a background system, and the background system finds a safety protection equipment library of the ascending operation behaviors according to the corresponding safety protection equipment library to obtain the types to be detected.
Detecting the type of the safety protection equipment to be detected in the image in a rotation detection mode to obtain a detection anchor frame of the detection equipment, calculating the contact ratio by adopting the detection anchor frame and the detection anchor frame of the operator, and judging that the operator fails to use the safety protection equipment according to the regulation when the contact ratio is less than 20%, and sending out an alarm.
Taking ascending electricity inspection operation and distribution welding operation as examples, continuing a test experiment, and verifying the rule violation detection method of the distribution network constructor:
TABLE 1
The video of the construction site can be acquired in real time through the visible light acquisition equipment and analyzed through the background management system, so that the monitoring efficiency can be greatly improved; the behavior judgment is carried out through the alpha phase structure, and the key point time diagram and the double behavior recognition deep learning network structure are combined, so that the behavior type of an operator can be accurately judged, and the construction safety is ensured; the system can automatically set the behavior type of the construction and the corresponding safety protection equipment library according to priori knowledge, and intelligent management of the construction site is realized.
It should be noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the technical solution of the present invention, which is intended to be covered by the scope of the claims of the present invention.
Example 3
A third embodiment of the present invention, which is different from the first two embodiments, is:
the functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-only memory (ROM), a random access memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
Logic and/or steps represented in the flowcharts or otherwise described herein, e.g., a ordered listing of executable instructions for implementing logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer readable medium may even be paper or other suitable medium on which the program is printed, as the program may be electronically captured, via, for instance, optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.
It is to be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, the various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, may be implemented using any one or combination of the following techniques, as is well known in the art: discrete logic circuits having logic gates for implementing logic functions on data signals, application specific integrated circuits having suitable combinational logic gates, programmable Gate Arrays (PGAs), field Programmable Gate Arrays (FPGAs), and the like.
Example 4
Referring to fig. 7, in a fourth embodiment of the present invention, the embodiment provides a system for detecting violations of a distribution network constructor based on action recognition, which includes an image acquisition module, a data receiving module, a target detection module, a job behavior recognition module, a background scheduling module, and a job safety protection device detection module.
The image acquisition in the image acquisition module is completed by a visible light acquisition system, and the whole monitoring of the construction site is realized by arranging a control ball visible light image acquisition device in the construction site.
And the data receiving module receives the transmission data of the video monitoring equipment after establishing the operation behavior database and the corresponding safety protection equipment library.
The target detection module is used for detecting the related safety protection equipment type object through the target detection module, and pixel position information of the safety protection equipment object is obtained.
The operation behavior recognition module extracts operation behaviors and specific position information of a plurality of operators on an operation construction site, takes an image with the operation behaviors as output, and outputs a result to the background management system.
And a background management system in the background scheduling module judges different types of safety protection equipment libraries according to the input information and the equipment database and outputs the safety protection equipment information to be detected.
It should be noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the technical solution of the present invention, which is intended to be covered by the scope of the claims of the present invention.
Claims (10)
1. A method for detecting rule violations of distribution network constructors based on action recognition is characterized by comprising the following steps: comprising the steps of (a) a step of,
the visible light acquisition system performs graph acquisition, establishes a corresponding database for matching and linking, receives transmission data of the video monitoring equipment, judges the operation behavior type of an operator by utilizing the behavior information judging model based on skeleton information, and outputs the operation behavior type to the background management system;
the background management system judges a safety protection equipment library of a corresponding type, outputs safety protection equipment information to be detected, and detects the information and the operation behavior image through the target detection module;
and comparing the anchor frame position information, performing coincidence calculation, and judging whether an operator violates rules or carries safety protection equipment according to regulations.
2. The method for detecting the violations of the distribution network constructors based on the action recognition according to claim 1, wherein the method comprises the following steps: the graphic acquisition comprises the steps of establishing a visible light acquisition system at a construction site, setting an image acquisition device of a distribution control ball and a visible light camera, configuring video monitoring equipment at the distribution construction site, acquiring videos, intercepting images at a fixed frequency, and transmitting the images to a background management system;
the corresponding databases comprise a logic matching base for establishing different types of operation behavior databases, setting operation behaviors and safety protection equipment, receiving transmission data of the video monitoring equipment, matching and linking the behavior type databases and the safety protection equipment databases in a background management system according to priori knowledge, wherein each type of behavior corresponds to a specific safety protection equipment database;
the behavior information discrimination model adopts an alpha phase framework to realize multi-person gesture recognition estimation and comprises a symmetrical space transformation network, a non-maximum suppression module and a behavior guiding suggestion generator.
3. The method for detecting the violations of the distribution network constructors based on the action recognition according to claim 2, wherein the method comprises the following steps: the target detection module comprises a step of detecting safety protection equipment by using target detection to find all operators on a construction site and utilizing rotary target detection;
The method comprises the steps that all operators in a construction site are detected by using target detection, the operators in the image are detected by using a Yolov5s algorithm, central pixel position information of the operators is obtained according to four vertex pixel position information of a detected anchor frame, preprocessing is carried out when the image enters the content of the Yolov5s algorithm, the quality of the image is improved through Mosaic data, then, image features are extracted by using a deep convolutional neural network and are realized by adopting a residual convolutional group with the size of 3x3, then, the size of a feature image is restored to the size of an original image through a pyramid pooling module, and finally, the target detection result image with the anchor frame information of the operators is marked and output by using the anchor frame for a feature object through an output end.
4. A method for detecting violations of a distribution network constructor based on motion recognition as claimed in claim 3, wherein: the method for extracting skeleton key points by adopting the space-time diagram convolutional neural network comprises the steps of setting 14 skeleton key points in a human body, constructing a connection relation of the skeleton key points of the human body, and constructing a space-time diagram according to video information and the connection relation;
the space-time diagram comprises defining a node t in a t-th frame image in the video as v ti Node i in the t+r frame image is defined as v (t+1)i Let v ti ＝(x 1 ,y 1 ,c 1 )，v (t+1)i ＝(x 2 ,y 2 ,c 2 ) Then the vector A of the same node in the T frame image can be expressed asThe method comprises the steps that the vector information of 14 skeleton key points is combined by blue line connection of the skeleton key points between adjacent frames to obtain a key point time diagram, image characteristic information of the key points is obtained by adopting diagram convolution, and a node v is obtained ti Is expressed as:
the spatial map convolution includes defining a hierarchical update rule by feature B and map structure G: the method comprises the steps of adopting a space configuration division strategy to divide a field set of skeleton information points into three subsets, wherein the first subset is skeleton nodes, the second subset is an adjacent node set which is closer to the center of gravity of the whole skeleton than the nodes in space position, and the third subset is an adjacent node set which is farther from the center than the nodes in space position, and the third subset is expressed as:
wherein f represents feature map information, v tj Nodes j, S (v) representing the t-th frame in the space diagram ti ) Representing v ti W represents a weighting function, l in Representing the mapping function, Z ij Representing regularization term Z ij (v ti )＝|{v tk |l ti (v tk )＝l ti (v ij ) Corresponding subset technique in } |l ti (v ti ) Is v ti Mapping under a single frame, d i Is the average distance from the key point i to the center of gravity of all frames in the training set, d j Is node v tj Distance to the center of gravity.
5. The method for detecting the violations of the distribution network constructors based on the action recognition according to claim 4, wherein the method comprises the following steps: the multi-person gesture recognition estimation comprises the steps of setting hand and head characteristics as interesting characteristics in a scene of a construction operation site, combining the characteristics by adopting a space-time convolution module to realize behavior characteristic extraction of a video, wherein TCN in a network represents time convolution operation, GCN+TCN in the network is realized by using a residual error network, and the calculation process of graph convolution is as follows:
∑ j A j ＝A+I
wherein f in For inputting a feature map, A is an adjacency matrix for key point connection in a t-th frame image, I is an identity matrix for self-linking, M is a learnable weight matrix for node importance among different bones, W is a weight matrix for map convolution,representing a normalized diagonal matrix;
in the video action recognition, the whole action judgment is completed by utilizing a double-flow structure: the method comprises the steps of realizing two inputs of a space flow and a time flow by using the space-time convolution network, extracting a connection diagram of each skeleton key point in a single frame image by the space flow, enabling the time flow to represent a position change connection diagram of the single skeleton key point in a period of time, superposing and fusing coefficients of probability distribution layers of the two after the two are connected, superposing the softmax of the space flow and the softmax of the time flow in the space-time convolution network to obtain a final fusion score, and sequencing the fusion scores;
And according to the content, performing skeleton key point measurement and calculation to judge whether a hand lifting action exists, if the hand lifting action does not exist, performing operation by a real person, if the hand lifting action exists, performing operation by the real person, and when the score of one action type is obviously higher than that of other action types according to the fusion score sequencing aiming at the real person in operation, determining that the highest score of the action type is the highest probability of the action, and when the scores of the action types are not different, judging the probability of the action by combining the video content and the construction environment.
6. The method for detecting the violations of the distribution network constructors based on the action recognition according to claim 5, wherein the method comprises the following steps: the detection of the safety protection equipment by using the rotating target detection comprises the following steps of using a target detection algorithm RTMDet with a rotating frame to realize the target detection output of the safety protection equipment, wherein the information of the anchor frame position of the operator in the operation behavior image is specifically as follows: adding a rotation detection head in a regression branch of YOLOv5, firstly adding 1x1 convolution in the regression branch to realize angle prediction, secondly modifying a boundary box modifier, and finally replacing the original GIOU loss function by using a rotation detection loss function;
The rotation detection loss function comprises the steps that a rotation marking frame of a detection diagram uses five parameters to represent a central coordinate point, length, width and deflection angle of the (x, y, w, h, theta) marking frame, and the (x * ,y * ,w * ,h * θ) represents a central coordinate point, length, width, and deflection angle of the prediction frame expressed as:
the calculation of θ is performed by means of direct regression and indirect regression, wherein the direct regression is expressed as F (Δθ) using a model comprising direct predicted angular offset:
the indirect regression includes matching two target vectors of a model prediction frame and a labeling frame, denoted as F (sin theta, cos theta), expressed as:
the calculation process continues to normalize, and according to the calculation process of indirect regression, a formula is set as a rotation loss function, and the function formula is expressed as:
wherein N is pos Refers to the number of all correct anchor frames, b n Marking the nth prediction frame, gt n Refers to an nth real annotation frame a 1 Is a super-ginseng, defaults to {0,01;1, n represents the number of dimensions, t i 、t' i Vector labels representing prediction and labeling frames, Z is the product of the rotation matrix and eigenvalue diagonal matrix of the image and the rotation transpose matrix, V B And (Z) calculating the volume of the corresponding rotation frame based on Kalman filtering to obtain a new Gaussian distribution, and solving the pixel coincidence probability distribution of the rotation labeling frame and the rotation prediction frame.
7. The method for detecting violations of the distribution network constructors based on action recognition according to claim 6, wherein the method comprises the following steps: the calculating of the contact ratio comprises the steps of comparing the position information of the anchor frame of the safety protection device in the target detection image with the position information of the anchor frame of the operator in the operation behavior image output by the operation detection module, and judging the distance between the position information of the operator in each operation behavior type and the position information of the safety protection device through the calculation of the contact ratio of the anchor frame of the safety protection device and the position information of the operator in each operation behavior type:
when the contact ratio of the anchor frame of the operator and the anchor frame of one of the safety protection devices is found to be 0< X <20%, an alarm is sent out, a background management system is prompted to indicate that the operator who is engaged in which operation acts has violations, the anchor frame contact ratio is repeatedly stopped until the contact ratio is more than 20%, whether other people violate the rules is checked at the same time, the synchronous check of multiple people is realized, and when the contact ratio X of the anchor frame of the operator and the anchor frame of one of the safety protection devices is more than or equal to 20%, the operator is considered to carry the safety protection devices, and whether other operators violate the rules is continuously checked.
8. A system employing the method for detecting violations of network constructors based on motion recognition according to any of claims 1 to 7, characterized in that: the system comprises an image acquisition module, a data receiving module, a target detection module, an operation behavior identification module, a background scheduling module and an operation safety protection equipment detection module;
The image acquisition module is used for completing image acquisition by a visible light acquisition system, and realizing integral monitoring of a construction site by arranging a control ball visible light image acquisition device in the construction site;
the data receiving module is used for receiving transmission data of the video monitoring equipment after establishing an operation behavior database and a corresponding safety protection equipment library;
the target detection module is used for detecting related safety protection equipment type objects to obtain pixel position information of the safety protection equipment objects;
the operation behavior recognition module extracts operation behaviors and specific position information of a plurality of operators on an operation construction site, takes an image with the operation behaviors as output, and outputs a result to the background management system;
the background management system judges different types of safety protection equipment libraries according to the input information and the equipment database and outputs safety protection equipment information to be detected;
and the operation safety protection equipment detection module obtains operation behavior types of all operators on a construction site after the behavior prediction is completed, and obtains detection types of the operation safety protection equipment according to the linkage of the operation behaviors in the background scheduling module and the operation safety protection equipment library.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 7 when the computer program is executed.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311260587.5A CN117173791A (en) | 2023-09-27 | 2023-09-27 | Distribution network constructor violation detection method and system based on action recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311260587.5A CN117173791A (en) | 2023-09-27 | 2023-09-27 | Distribution network constructor violation detection method and system based on action recognition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117173791A true CN117173791A (en) | 2023-12-05 |
Family
ID=88943083
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311260587.5A Pending CN117173791A (en) | 2023-09-27 | 2023-09-27 | Distribution network constructor violation detection method and system based on action recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117173791A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118249517A (en) * | 2024-05-21 | 2024-06-25 | 国网江西省电力有限公司南昌供电分公司 | Multimodal substation operation and maintenance management method and system based on Wi-Fi probe |
| CN118277925A (en) * | 2024-06-04 | 2024-07-02 | 北京警察学院 | Classification security assessment method and device for public security video image system |
| CN118840785A (en) * | 2024-09-20 | 2024-10-25 | 朗坤智慧科技股份有限公司 | Intelligent identification method and system for ascending operation risk based on industrial scene |
| CN119048974A (en) * | 2024-07-25 | 2024-11-29 | 广东伏光新能源科技有限公司 | Construction safety monitoring method and device, storage medium and electronic equipment |
| CN119090877A (en) * | 2024-11-06 | 2024-12-06 | 江苏中车数字科技有限公司 | A method and system for visually monitoring welding operation process and quality |
| CN119090877B (en) * | 2024-11-06 | 2025-04-11 | 江苏中车数字科技有限公司 | A method and system for visually monitoring welding operation process and quality |
-
2023
- 2023-09-27 CN CN202311260587.5A patent/CN117173791A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118249517A (en) * | 2024-05-21 | 2024-06-25 | 国网江西省电力有限公司南昌供电分公司 | Multimodal substation operation and maintenance management method and system based on Wi-Fi probe |
| CN118277925A (en) * | 2024-06-04 | 2024-07-02 | 北京警察学院 | Classification security assessment method and device for public security video image system |
| CN119048974A (en) * | 2024-07-25 | 2024-11-29 | 广东伏光新能源科技有限公司 | Construction safety monitoring method and device, storage medium and electronic equipment |
| CN118840785A (en) * | 2024-09-20 | 2024-10-25 | 朗坤智慧科技股份有限公司 | Intelligent identification method and system for ascending operation risk based on industrial scene |
| CN119090877A (en) * | 2024-11-06 | 2024-12-06 | 江苏中车数字科技有限公司 | A method and system for visually monitoring welding operation process and quality |
| CN119090877B (en) * | 2024-11-06 | 2025-04-11 | 江苏中车数字科技有限公司 | A method and system for visually monitoring welding operation process and quality |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN117173791A (en) | Distribution network constructor violation detection method and system based on action recognition | |
| Weerasinghe et al. | Application of Microsoft Kinect sensor for tracking construction workers | |
| CN109829908B (en) | Binocular image-based method and device for detecting safety distance of ground object below power line | |
| CN112818768A (en) | Transformer substation reconstruction and extension violation behavior intelligent identification method based on meta-learning | |
| Al-Sabbag et al. | Enabling human–machine collaboration in infrastructure inspections through mixed reality | |
| CN113989944B (en) | Operation action recognition method, device and storage medium | |
| Kim et al. | Hybrid DNN training using both synthetic and real construction images to overcome training data shortage | |
| CN111209864A (en) | Target identification method for power equipment | |
| CN113537019B (en) | Detection method for identifying wearing of safety helmet of transformer substation personnel based on key points | |
| CN114494427B (en) | Method, system and terminal for detecting illegal behaviors of person with suspension arm going off station | |
| CN113569801A (en) | Distribution construction site live equipment and live area identification method and device thereof | |
| CN114170686A (en) | Elbow bending behavior detection method based on human body key points | |
| Liu et al. | A dual attention network for automatic metallic corrosion detection in natural environment | |
| Zhang et al. | B2CNet: A Progressive Change Boundary-to-Center Refinement Network for Multi-Temporal Remote Sensing Images Change Detection | |
| CN113361968B (en) | Power grid infrastructure worker safety risk assessment method based on artificial intelligence and big data | |
| CN118887608A (en) | Method and device for monitoring operation of power transmission line, and electronic equipment | |
| CN118968347A (en) | A structural damage adaptive identification method and system | |
| CN118552844A (en) | Knowledge-driven automatic tracking method and device for bridge structure construction progress | |
| Kim et al. | Training a visual scene understanding model only with synthetic construction images | |
| CN116403275B (en) | Method and system for detecting personnel advancing posture in closed space based on multi-vision | |
| CN117392611A (en) | Construction site safety monitoring methods, systems, equipment and storage media | |
| CN115527008A (en) | Safety simulation experience training system based on mixed reality technology | |
| CN118092654B (en) | Virtual reality application method, system, terminal and storage medium for operation and maintenance industry | |
| Liu et al. | Research on an improved yolov5s algorithm for detecting helmets on construction sites | |
| CN112883889A (en) | Method suitable for detecting violation behaviors in power construction scene |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |