A method for detecting a violation of an operating tool, an embedded terminal, and a storage medium

Abstract

The application relates to a method for detecting illegal operation of a tool, an embedded terminal and a storage medium. The application uses a target detection algorithm to detect whether a scene image contains a human body and a tool; if yes, the scene image is input into a lightweight OpenPose-CA key point detection algorithm to perform real-time human body and tool key point detection, and the key point set of the human wrist and the tool in each scene image is output; each key point in the key point set in the continuous scene image is connected into a straight line, and the consistency of the hand and the tool action is determined according to whether the shapes and directions of the straight lines are similar, and then whether the human body holds the tool is determined based on the action consistency and the distance between the wrist and the tool key point; the Euclidean distance between the tool key point and the center coordinates of a dangerous area is calculated, and the relationship between the Euclidean distance and a preset threshold is analyzed to determine whether the operation is standard. The method, the embedded terminal and the storage medium provided by the application are lightweight under the condition of ensuring detection accuracy, and are suitable for industrial embedded scenes.

Description

Technical Field

[0001] This invention relates to the field of computer vision operation detection, and more particularly to a method for detecting tools used in violation of operating procedures, an embedded terminal, and a storage medium. Background Technology

[0002] The rapid development of industry has greatly boosted national economic growth, while the construction industry has provided employment opportunities for many. However, due to the large number of industrial equipment and the complexity of operating methods, industrial safety accidents occur frequently. These accidents are the result of a combination of factors from various aspects. In total quality management, the causes of industrial accidents are typically categorized into four aspects: production violations, production equipment, operating methods, and the production environment. Among these, two crucial factors influencing accident occurrence are production violations and operating methods. Therefore, strengthening the standardization and supervision of workers' operating methods is essential for protecting personal safety, reducing property damage, and minimizing accidents.

[0003] Currently, my country's artificial intelligence (AI) technology is developing rapidly, data and computing resources are becoming increasingly abundant, and application scenarios are constantly expanding, laying a solid foundation for innovative applications of AI. In current industrial production processes, some equipment requires workers to operate tools in hazardous areas. Applying target detection and key point detection algorithms to monitor violations of operating procedures is a highly effective way to avoid safety hazards. Chen Guodong et al. proposed a construction site hazardous behavior monitoring method based on YOLOv3 and OpenPose. This method uses YOLOv3 to detect personnel and OpenPose to detect key points, thereby determining whether personnel are in a safe area. However, YOLOv3 has low accuracy, and the OpenPose model is deployed in the cloud, resulting in high cost, too many weight parameters, and long inference time, making it unsuitable for industrial embedded scenarios and failing to meet real-time detection standards. Xu Linhai et al. proposed an embedded lightweight driver leg pose estimation method based on an improved OpenPose, replacing the OpenPose feature extraction network with the lightweight MobileNetV1 network. While this solution achieves lightweight design and is more suitable for embedded scenarios, it leads to a decrease in accuracy. Summary of the Invention

[0004] To address, or at least partially address, the aforementioned technical problems, this invention provides a method for detecting unauthorized tools, an embedded terminal, and a storage medium. The feature extraction network of the OpenPose model is modified by replacing it with a lightweight VarGNet network suitable for embedded systems, reducing model complexity and improving detection efficiency. Simultaneously, to ensure detection accuracy, this paper first performs object detection, and then adds a coordinate attention network suitable for terminals to the OpenPose model to capture more information on feature channels and location, further improving model accuracy.

[0005] In a first aspect, the present invention provides a method for detecting tools used in violation of operating procedures, comprising:

[0006] Extract scene images frame by frame from continuous video of the monitored area from the surveillance camera, draw the dangerous area of ​​the equipment, and obtain the center coordinates of the dangerous area;

[0007] Use object detection algorithms to detect whether there are human bodies and tools in a scene image;

[0008] If human bodies and tools are detected, the scene image is input into the lightweight OpenPose-CA keypoint detection algorithm to perform real-time detection of human body and tool keypoints, and outputs a set of keypoints of human wrists and tools in each scene image;

[0009] Connect each key point in the set of key points in a continuous scene image with a straight line. Determine the consistency of hand and tool movements based on the similarity of the shape and direction of each straight line. Then, determine whether the human body is holding a tool based on the consistency of movements and the distance between the wrist and tool key points.

[0010] The Euclidean distance between the key points of the calculation tool and the center coordinates of the danger zone is calculated, and the relationship between the Euclidean distance and the preset threshold is analyzed to determine whether the operation is in accordance with regulations.

[0011] Furthermore, a quadrilateral danger zone is manually drawn in the monitoring area of ​​the surveillance camera, and the coordinates of the upper left and lower right corners of the danger zone are obtained. Then, the center coordinates of the danger zone are calculated based on the upper left and lower right corner coordinates.

[0012] Furthermore, the lightweight OpenPose-CA keypoint detection algorithm replaces the feature extraction network in the existing OpenPose algorithm with a lightweight VarGNet network based on depthwise separable variable group convolutions. Lightweight coordinate attention (CA) is added to the lightweight VarGNet network to obtain relevant information between feature channels and in space. The lightweight coordinate attention is added after the last convolutional layer of the Normal Block of the lightweight VarGNet network.

[0013] Furthermore, the lightweight coordinate attention performs global average pooling on the input feature map in both the width and height directions to obtain a first width-direction feature map and a first height-direction feature map, respectively. The width-direction and height-direction feature maps are concatenated and fed into a shared convolutional layer to reduce the dimensionality of the concatenated feature map, followed by batch normalization and a sigmoid activation function to obtain the target feature map. The target feature map is then convolved in both the height and width directions to obtain a second height-direction feature map and a second width-direction feature map with the same number of channels and size as the first height-direction and first width-direction feature maps. The second height-direction and second width-direction feature maps are then activated by a sigmoid activation function to obtain attention weights in the height and width directions, respectively.

[0014] The input feature map is multiplied and weighted with height and width attention weights to obtain a feature map with attention weights.

[0015] Furthermore, the set of key points includes the coordinates of the left wrist, the right wrist, both ends of the tool, and the center coordinates.

[0016] Furthermore, by connecting each keypoint in the keypoint set of the wrist and tool within the continuous scene images with a straight line, the displacements of the left wrist, right wrist, both ends of the tool, and the center are obtained. Based on the consistency of the displacements of the left wrist and / or right wrist and the tool, it is preliminarily determined whether the human is holding a tool. For all wrists and tools for which a holding relationship is preliminarily determined, the Euclidean distance D between the left and right wrists and the two endpoints of the tool keypoints of all tools is calculated. L1 D L2 D L3 and D L4 Analysis of D L1 D L2 D L3 and D L4 If the smallest value is less than the set first threshold, then the tool is determined to be held. The two points corresponding to the smallest value are the wrist point holding the tool and the tool point near the wrist.

[0017] Furthermore, the target detection algorithm adopts the YOLO target detection algorithm.

[0018] Furthermore, the Euclidean distance between the key points of the calculation tool and the center coordinates of the danger zone is analyzed, and the relationship between the Euclidean distance and the preset threshold is analyzed to determine whether the operation is in accordance with regulations, including:

[0019] Calculate the Euclidean distance D between the two endpoints of the tool and the center point of the hazardous area. T1 and D T2 Analyze D T1 and D T2 The tool determines whether the smaller value is less than the set second threshold and whether it is located in a danger zone.

[0020] Secondly, the present invention provides an embedded terminal for detecting tools used in violation of operating procedures, comprising: an image acquisition unit, a storage unit, a processing unit, and a bus unit, wherein the bus unit connects the image acquisition unit, the storage unit, and the processing unit, the storage unit stores a computer program, and the computer program, when executed by the processing unit, implements the method for detecting tools used in violation of operating procedures.

[0021] Thirdly, the present invention provides a storage medium for implementing a method for detecting tools used in violation of regulations, the storage medium storing a computer program, which, when executed by a processor, implements the method for detecting tools used in violation of regulations.

[0022] The technical solutions provided in the embodiments of the present invention have the following advantages compared with the prior art:

[0023] This application introduces a lightweight OpenPose-CA keypoint detection algorithm suitable for embedded systems. This lightweight OpenPose-CA algorithm reduces the complexity of the OpenPose model and improves detection speed; it eliminates the need for cloud configuration and can achieve real-time and efficient detection of unauthorized tools solely through embedded terminal devices. The lightweight OpenPose-CA keypoint detection algorithm replaces the feature extraction network in the existing OpenPose algorithm with a lightweight VarGNet network based on depthwise separable variable-group convolutions. Lightweight coordinate attention (CA) is added to the VarGNet lightweight network to obtain spatial and inter-channel correlation information. Specifically, the lightweight coordinate attention is added after the last convolutional layer of the VarGNet lightweight network's Normal Block. This reduces network complexity while focusing attention on important parts of the features by obtaining spatial and inter-channel correlation information, thereby further improving the algorithm's recognition accuracy.

[0024] This invention determines the consistency of hand and tool movements by comparing the similarity in shape and direction formed by connecting key points such as the wrist and tool within continuous scene images. It further determines whether a person or worker is holding a tool by combining the distance between the wrist and tool key points with the consistency of hand and tool movements. Finally, it judges whether the operation is standardized based on the Euclidean distance between the tool's key points and its center coordinates and a preset threshold. By combining target detection and improved key point detection methods, the accuracy of detecting improper tool operations is effectively guaranteed, reducing the false negative rate and achieving precise supervision. Attached Figure Description

[0025] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 A flowchart illustrating a method for detecting tools used in violation of regulations, provided as an embodiment of the present invention;

[0028] Figure 2 A schematic diagram illustrating the target detection algorithm for recognizing human bodies and tools provided in an embodiment of the present invention;

[0029] Figure 3 A schematic diagram of the VarGNet lightweight network provided in an embodiment of the present invention;

[0030] Figure 4 A schematic diagram of the lightweight coordinate attention network structure provided in an embodiment of the present invention;

[0031] Figure 5 This is a schematic diagram showing that each key point in the key point set of the wrist and tool in a continuous scene image provided in an embodiment of the present invention is connected by a straight line.

[0032] Figure 6 This is a schematic diagram of an embedded terminal for detecting unauthorized operations, provided as an embodiment of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0035] Example 1

[0036] See Figure 1 As shown in the figure, an embodiment of the present invention provides a method for detecting tools used in violation of regulations, comprising:

[0037] S1. Extract scene images frame by frame from the continuous video of the monitored area from the surveillance camera, manually draw the danger zone of the equipment, and obtain the center coordinates of the danger zone; manually draw a quadrilateral danger zone in the monitored area of ​​the surveillance camera, and obtain the coordinates of the upper left corner (x1, y1) and the lower right corner (x2, y2) of the danger zone, and then calculate the average of the x and y coordinates of the upper left and lower right corners to obtain the center coordinates (x0, y0) of the danger zone:

[0038]

[0039] S2, use an object detection algorithm to detect whether there are human bodies and tools in the scene image; in specific implementation, the object detection algorithm adopts the YOLO object detection algorithm, preferably YOLOv7, and trains YOLOv7 to detect human bodies and tools, see [reference]. Figure 2 As shown, YOLOv7 selects the area where the human body is located and the area where the tool is sitting in the scene image using bounding boxes. The bounding box of the human body area is labeled with the "person" tag, and the bounding box of the tool area is labeled with the "tool" tag. If a human body and a tool are detected in the scene image, step S3 is executed.

[0040] S3, input the scene image into the lightweight OpenPose-CA key point detection algorithm to perform real-time human and tool key point detection, and output the key point set of human wrist and tool in each scene image; the key point set includes the coordinates of the left wrist, the coordinates of the right wrist, the coordinates of both ends of the tool and the center coordinates.

[0041] The OpenPose algorithm consists of a feature extraction network, a partial affinity field (PAF), and a confidence matrix. Figure 3 The existing OpenPose algorithm consists of several parts. The feature extraction network is the VGG19 network, which only uses a 12-layer backbone convolutional network. The remaining network is parallel. The first branch is the PAF part, which is used to pre-estimate the position of each joint in the human body, solving the problem of encoding only the position and too many support areas. The second branch is the confidence map part, which is used to pair and group the joints according to the number of human bodies based on the confidence of each part.

[0042] The lightweight OpenPose-CA keypoint detection algorithm replaces the feature extraction network in the existing OpenPose algorithm with a lightweight VarGNet network based on depthwise separable variable-group convolutions. This adapts to embedded terminals and improves detection speed. Lightweight coordinate attention (CA) is added to the VarGNet network to obtain spatial and inter-channel correlation information. Specifically, lightweight coordinate attention is added after the last convolutional layer of the VarGNet's Normal Block. The introduction of lightweight coordinate attention focuses attention on important parts of the features, further improving the algorithm's recognition accuracy.

[0043] For details, please refer to Figure 3 As shown, in the VarGNet lightweight network's variable-group convolutions, the number of input channels in each group is fixed and can be adjusted as a hyperparameter, unlike the fixed number of groups in group convolutions. The advantages of this include: from the compiler's perspective, a fixed number of input channels is more suitable for optimization because it has a more consistent computational pattern and data layout; furthermore, it allows for a larger network capacity in variable-group convolutions, thus allowing for smaller channel numbers, which helps reduce the time consumption of off-chip communication.

[0044] For details, please refer to Figure 4 As shown, in order to acquire attention in both the width and height of the image and encode precise location information, the lightweight coordinate attention performs global average pooling on the input feature map in both the width and height directions. Figure 3 In the diagram, X Avg Pool and Y Avg Pool represent global average pooling along the height direction and global average pooling along the width direction, respectively. The first height-direction feature map and the first width-direction feature map are obtained respectively:

[0045]

[0046] Where, x c Let h represent the height of the input feature map, w represent the width of the input feature map, H represent the total height of the input feature map, and W represent the total width of the input feature map.

[0047] The width-direction feature map z w and height-direction feature map z h The concatenated feature maps are then fed into a shared convolutional layer with a 1×1 kernel to reduce their dimensionality. After batch normalization and a sigmoid activation function, the target feature map is obtained. Figure 3 In this application, BatchNorm represents batch normalization, and Non-linear represents non-linearity. The application uses the non-linear Sigmoid activation function.

[0048] f=σ(B(cov([z h ,z w ]))), where σ represents the Sigmoid activation function, B represents batch normalization, and cov represents convolution.

[0049] The target feature map is convolved along the height and width directions to obtain a second height direction feature map F with the same number of channels and size as the first height direction feature map and the first width direction feature map, respectively. h Second width direction feature map F w The second height-direction feature map and the second width-direction feature map are respectively processed by the Sigmoid activation function to obtain the attention weights in the height and width directions:

[0050] g h =σ(F h );

[0051] g w =σ(F w ).

[0052] The input feature map is multiplied and weighted with height and width attention weights to obtain a feature map with attention weights:

[0053] y c (i,j)=x c (i,j)×g h (i)×g w (j).

[0054] S4 connects each keypoint in the keypoint set within the continuous scene image with a straight line. The similarity in shape and direction of these lines determines the consistency of hand and tool movements. Furthermore, based on this consistency and the distance between the wrist and tool keypoints, it is determined whether the person is holding the tool. For specific implementation details, please refer to [link / reference]. Figure 5 As shown, by connecting each keypoint in the keypoint set of the wrist and tool within a continuous scene image with a straight line, the displacements of the left wrist, right wrist, both ends of the tool, and the center are obtained. Based on the consistency of the displacements of the left wrist and / or right wrist and the tool, it is preliminarily determined whether the human is holding a tool. For all wrists and tools for which a holding relationship is preliminarily determined, the Euclidean distance D between the left and right wrists and the two endpoints of the tool keypoints of all tools is calculated. L1 D L2 D L3 and D L4 Analysis of D L1 D L2 D L3 and D L4If the smallest value is less than the set first threshold, then the tool is determined to be held. The two points corresponding to the smallest value are the wrist point holding the tool and the tool point near the wrist.

[0055] S5, calculate the Euclidean distance between the key points of the tool and the center coordinates of the hazardous area, and analyze the relationship between the Euclidean distance and the preset threshold to determine whether the operation is standardized. In specific implementation, calculate the Euclidean distance D between the two endpoints of the tool and the center coordinates of the hazardous area. T1 and D T2 Analyze D T1 and D T2 The tool determines whether the smaller value is less than the set second threshold and whether it is located in a danger zone.

[0056] Example 2

[0057] See Figure 6 As shown, this embodiment of the invention provides an embedded terminal for detecting tools used in violation of operating procedures, comprising: an image acquisition unit, a storage unit, a processing unit, and a bus unit, wherein the bus unit connects the image acquisition unit, the storage unit, and the processing unit, the storage unit stores a computer program, and the computer program, when executed by the processing unit, implements the method for detecting tools used in violation of operating procedures.

[0058] Example 3

[0059] This invention provides a storage medium for implementing a method for detecting tools used in violation of operating procedures. The storage medium stores a computer program, which, when executed by a processor, implements the method for detecting tools used in violation of operating procedures as described above.

[0060] This application introduces a lightweight OpenPose-CA keypoint detection algorithm suitable for embedded systems. This lightweight OpenPose-CA algorithm reduces the complexity of the OpenPose model and improves detection speed; it eliminates the need for cloud configuration and can achieve real-time and efficient detection of unauthorized tools solely through embedded terminal devices. The lightweight OpenPose-CA keypoint detection algorithm replaces the feature extraction network in the existing OpenPose algorithm with a lightweight VarGNet network based on depthwise separable variable-group convolutions. Lightweight coordinate attention (CA) is added to the VarGNet lightweight network to obtain spatial and inter-channel correlation information. Specifically, the lightweight coordinate attention is added after the last convolutional layer of the VarGNet lightweight network's Normal Block. This reduces network complexity while focusing attention on important parts of the features by obtaining spatial and inter-channel correlation information, thereby further improving the algorithm's recognition accuracy.

[0061] This invention determines the consistency of hand and tool movements by comparing the similarity in shape and direction formed by connecting key points such as the wrist and tool within continuous scene images. It further determines whether a person or worker is holding a tool by combining the distance between the wrist and tool key points with the consistency of hand and tool movements. Finally, it judges whether the operation is standardized based on the Euclidean distance between the tool's key points and its center coordinates and a preset threshold. By combining target detection and improved key point detection methods, the accuracy of detecting improper tool operations is effectively guaranteed, reducing the false negative rate and achieving precise supervision.

[0062] In the embodiments provided by this invention, it should be understood that the disclosed structures and methods can be implemented in other ways. For example, the structural embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, structures, or units, and may be electrical, mechanical, or other forms.

[0063] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0064] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0065] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for detecting improperly operated tools, characterized in that, include: Extract scene images frame by frame from continuous video of the monitored area from the surveillance camera, draw the dangerous area of ​​the equipment, and obtain the center coordinates of the dangerous area; Use object detection algorithms to detect whether there are human bodies and tools in a scene image; If a human body and a tool are detected, the scene image is input into the lightweight OpenPose-CA keypoint detection algorithm for real-time detection of human body and tool keypoints, outputting a set of keypoints for the human wrist and tools in each scene image. The lightweight OpenPose-CA keypoint detection algorithm replaces the feature extraction network in the existing OpenPose algorithm with a lightweight VarGNet network based on depthwise separable variable-group convolutions. Lightweight coordinate attention (CA) is added to the VarGNet network to obtain spatial and inter-channel information. The lightweight coordinate attention is added after the last convolutional layer of the VarGNet network's Normal Block. The lightweight coordinate attention performs global average pooling on the input feature map in both width and height directions, obtaining feature maps in the first width and first height directions, respectively. The width and height feature maps are concatenated and fed into a shared convolutional layer to reduce the dimensionality of the concatenated feature map, followed by batch normalization and sigmoid. The activation function yields the target feature map; the target feature map is convolved along the height and width directions to obtain a second height-direction feature map and a second width-direction feature map with the same number of channels and size as the first height-direction feature map and the first width-direction feature map; the second height-direction feature map and the second width-direction feature map are then activated by the Sigmoid activation function to obtain attention weights in the height and width directions respectively; the input feature map is then multiplied and weighted by the height-direction attention weights and the width-direction attention weights to obtain a feature map with attention weights; Connecting each keypoint in a set of keypoints within a continuous scene image into a straight line, the consistency of hand and tool movements is determined based on the similarity of the shape and direction of these lines. Then, based on this consistency and the distance between the wrist and tool keypoints, it is determined whether the person is holding a tool. This process includes: connecting each keypoint in the set of wrist and tool keypoints within the continuous scene image into a straight line to obtain the displacements of the left wrist, right wrist, both ends of the tool, and its center; initially determining whether the person is holding a tool based on the consistency of the left wrist and / or right wrist and tool displacements; and for all wrists and tools with preliminarily determined holding relationships, calculating the Euclidean distance D between the left and right wrists and the two endpoints of the tool keypoints for all tools. L1 D L2 D L3 and D L4 Analysis of D L1 D L2 D L3 and D L4 If the smallest value is less than the set first threshold, then the handheld tool is determined. The two points corresponding to the smallest value are the wrist point holding the tool and the end point of the tool near the wrist, respectively. The Euclidean distance between the key points of the calculation tool and the center coordinates of the danger zone is calculated, and the relationship between the Euclidean distance and the preset threshold is analyzed to determine whether the operation is in accordance with regulations.

2. The method for detecting improperly operated tools according to claim 1, characterized in that, A quadrilateral danger zone is manually drawn in the monitoring area of ​​the surveillance camera, and the coordinates of the upper left and lower right corners of the danger zone are obtained. Then, the center coordinates of the danger zone are calculated based on the upper left and lower right corner coordinates.

3. The method for detecting improperly operated tools according to claim 1, characterized in that, The set of key points includes the coordinates of the left wrist, the right wrist, both ends of the tool, and the center.

4. The method for detecting improperly operated tools according to claim 1, characterized in that, The target detection algorithm used is the YOLO target detection algorithm.

5. The method for detecting improperly operated tools according to claim 1, characterized in that, Calculate the Euclidean distance between the key points of the calculation tool and the center coordinates of the danger zone, and analyze the relationship between the Euclidean distance and the preset threshold to determine whether the operation is in compliance with regulations. Calculate the Euclidean distance D between the two endpoints of the tool and the center point of the hazardous area. T1 and D T2 Analyze D T1 and D T2 The tool determines whether the smaller value is less than the set second threshold and whether it is located in a danger zone.

6. An embedded terminal for detecting tools used in violation of operating procedures, characterized in that, include: The system includes an image acquisition unit, a storage unit, a processing unit, and a bus unit, wherein the bus unit connects the image acquisition unit, the storage unit, and the processing unit, the storage unit stores a computer program, and the computer program, when executed by the processing unit, implements the detection method for the illegal operation tool as described in any one of claims 1-5.

7. A storage medium for implementing a method for detecting tools used in violation of operating procedures, the storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the detection method for the illegal operation tool as described in any one of claims 1-5.

Images