Method, device and storage medium for monitoring medical staff's actions and preventing fatigue

Abstract

The present invention discloses a method, device and storage medium for monitoring the actions of medical staff and preventing fatigue, which relates to the technical field of image processing. Among them, the method includes: collecting the posture images of medical staff from different perspectives to construct an original data set; performing data enhancement processing on each image in the original data set; using the labelme software to mark the body key points; inputting the marked image data set into the DNPOSE-YOLOv8 network for training; collecting the working images of medical staff, and using the trained DNPOSE-YOLOv8 network to identify the coordinate information of each key point of the medical staff's body during work; calculating the positional relationship between the key points to judge whether there is a risk in the working posture of the medical staff; generating a reminder message to instruct the medical staff to optimize their working postures or actions. Through the present invention, the rapid and accurate detection of the postures of medical staff can be realized, the bad postures of medical staff can be corrected, and musculoskeletal diseases can be prevented.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a method, device and storage medium for monitoring the actions of medical staff and preventing fatigue. Background Art

[0002] Work-related Musculoskeletal Disorders (abbreviated as WRMSDs) are a common type of occupational injury caused by long-term, repetitive or extreme postures, which seriously affect the health and well-being of workers in all industries. WRMSDs not only cause troubles to the health of patients, but also cause huge economic losses to individuals and society.

[0003] To reduce the harm caused by WMSDs, researchers conduct ergonomic risk assessments for different work tasks. Commonly, it is to determine the angles of parts such as the upper limbs, lower limbs, neck and trunk, as well as factors such as the force on the worker and the standing posture, frame by frame by observing the videos or pictures of workers during work. Finally, the risk assessment of the work task is completed. Although this method can help identify and improve the factors that may cause WRMSDs in the working environment to a certain extent, its operation process is complicated and time-consuming, and the evaluation results are greatly affected by subjective factors, which limits the evaluation accuracy and intervention effect. Summary of the Invention

[0004] Aiming at the problem that the existing technology for monitoring the postures of medical staff is inaccurate and untimely, the present invention provides a method, device and storage medium for monitoring the actions of medical staff and preventing fatigue, so as to achieve rapid and accurate detection of the postures of medical staff, correct the bad postures of medical staff, and prevent musculoskeletal diseases.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] According to one aspect of the present invention, there is provided a method for monitoring the actions of medical staff and preventing fatigue, including: acquiring medical staff posture images from different perspectives collected by a high-precision camera to construct an original data set; performing data augmentation processing on each image in the original data set to increase data diversity and expand the data set; using labelme software to label the human key points on each image; inputting the labeled image data set into the DNPOSE-YOLOv8 network for training to obtain the trained DNPOSE-YOLOv8 network, where the trained DNPOSE-YOLOv8 network is used to identify the target object in the input image and locate the positions of each key point of the target object; collecting real-time medical staff work images and using the trained DNPOSE-YOLOv8 network to identify the coordinate information of each key point of the medical staff's body during work; calculating the positional relationship between the key points according to the coordinate information, and judging whether there is a risk in the medical staff's work posture according to the positional relationship between the key points; generating a reminder message when a risk posture is detected, instructing the medical staff to optimize their work postures or actions to reduce occupational fatigue.

[0007] Optionally, performing data augmentation processing on each image in the original data set includes: taking the clockwise direction as the positive direction, performing rotation processing of ±30°, ±45°, ±60°, ±90° on each image respectively; performing random brightness and darkness change adjustment on each image, and screening the images after the brightness and darkness change to remove the images with unqualified brightness; adding salt-and-pepper noise, Gaussian noise, and speckle noise processing to each image respectively.

[0008] Optionally, before inputting the labeled image data set into the DNPOSE-YOLOv8 network for training, the method further includes: uniformly adjusting the image size in the data set to 640*640 and scaling the pixel values to between 0 and 1.

[0009] Optionally, the DNPOSE-YOLOv8 network is an improvement based on YOLOv8-pose, and the improvement content includes: adopting a more lightweight detection head LW-Pose to speed up the inference process of the network; introducing an asymmetric design in the network structure to reduce the computational complexity and the number of model parameters; using 3×3 depthwise separable convolution DS_Conv instead of the traditional 3×3 convolution in each branch to reduce the number of parameters; adding a feature focus diffusion pyramid structure to the HEAD network structure, where the pyramid structure extracts the features of each layer and performs feature fusion, and then diffuses the fused features to each feature dimension, so that each feature dimension has more comprehensive feature information, thereby realizing the feature focus diffusion of the pyramid structure and improving the detection accuracy.

[0010] Optionally, the pyramid structure adopts a feature focusing module and a feature diffusion mechanism. The feature focusing module aims to capture and integrate feature information from different scales and generate a feature map with context information. Among them, the feature focusing module accepts inputs of three scales and contains an Inception-Style module internally. The Inception-Style module uses a set of parallel depth convolutions to capture information across multiple scales. The feature diffusion mechanism aims to diffuse the context information features generated by the feature focusing module to each detection scale to achieve cross-scale feature fusion and enhancement and improve the detection accuracy.

[0011] Optionally, the positional relationship between the key points includes the distance between the key points. Calculating the positional relationship between the key points according to the coordinate information and judging whether there is a risk in the working posture of the medical staff according to the positional relationship between the key points includes: obtaining the coordinates of the left and right ankles output by the DNPOSE-YOLOv8 network, and calculating the normal distance between the left and right ankles; judging whether the normal distance between the left and right ankles exceeds a preset threshold. If it exceeds, it is considered that there is a risk in the working posture of the medical staff.

[0012] Optionally, the positional relationship between the key points includes the joint angle. Calculating the positional relationship between the key points according to the coordinate information and judging whether there is a risk in the working posture of the medical staff according to the positional relationship between the key points includes:

[0013] For the three key points that form the joint angle 、 、 , where is the joint point, 、 are the key points adjacent to the joint point . The included angle formed by the line segments and is the joint angle of the joint . According to 、 、 coordinate values, calculate the vectors 、 :

[0014] ;

[0015] ;

[0016] In the formula, are the x, y, and z coordinate values of the key point respectively; are the x, y, and z coordinate values of the key point respectively; The x, y, and z coordinate values of the key points respectively ;

[0017] Calculate the cosine value of the angle between vectors and through the dot product and magnitude of the vectors: ω ;

[0018] ;

[0019] Let , then the joint angle of joint ;

[0020] Judge whether the calculated joint angle is within the preset range. If it is not within the preset range, it is considered that there is a risk in the working posture of the medical staff.

[0021] Optionally, the generating a reminder message when a risk posture is detected includes: setting risk levels, including negligible risk, low risk, medium risk, and high risk, where negligible risk corresponds to levels 1-2 in the RULA scale, low risk corresponds to levels 3-4 in the RULA scale, medium risk corresponds to levels 5-6 in the RULA scale, and high risk corresponds to level 7 in the RULA scale; displaying reminder messages corresponding to the risk levels in different colored fonts through the display interface, where negligible risk corresponds to displaying white font, low risk corresponds to displaying yellow font, medium risk corresponds to displaying orange font, and high risk corresponds to displaying red font.

[0022] According to another aspect of the present invention, there is also provided a medical staff movement monitoring and fatigue prevention device, where the medical staff movement monitoring and fatigue prevention device includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the computer program is executed by the processor, it implements the steps of the method described above.

[0023] According to another aspect of the present invention, there is also provided a storage medium, where a computer program is stored on the storage medium, and when the computer program is executed by a processor, it implements the steps of the method described above.

[0024] A method for monitoring the actions of medical staff and preventing fatigue proposed by the present invention realizes the rapid and accurate detection of the postures of medical staff, avoids human subjective errors, can correct the bad postures of medical staff, and prevent musculoskeletal diseases. In addition, the method of the present invention avoids the physiological and psychological discomfort caused to workers by wearing sensors, has higher automation and more detailed and objective evaluation, and reduces the disadvantages of time-consuming and complicated evaluation methods. The present invention has the characteristics of high efficiency, real-time, and accuracy, and can be widely applied to medical scenarios to help medical staff optimize the dispensing postures and actions, reduce occupational fatigue, and improve work efficiency and safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0026] Figure 1 is a flowchart of a method for monitoring the actions of medical staff and preventing fatigue according to an embodiment of the present invention;

[0027] Figure 2 is a schematic diagram of the camera layout;

[0028] Figure 3 is a network structure diagram of DNPOSE-YOLOv8;

[0029] Figure 4 is a structure diagram of LW-Pose;

[0030] Figure 5 is a structure diagram of the feature focus diffusion pyramid;

[0031] Figure 6 is a structure diagram of the feature focus diffusion module;

[0032] Figure 7 is an experimental result diagram of the original YOLOv8-pose model and the DNPOSE-YOLOv8 model. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] In order to enable those skilled in the art to better understand the solution of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application. It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0034] Second, the "one embodiment" or "embodiment" referred to herein means a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The "in one embodiment" that appears in different places in this specification does not all refer to the same embodiment, nor is it an embodiment that is separate or selectively mutually exclusive with other embodiments.

[0035] It should be noted that the terms "first", "second", etc. in the specification, claims, and the above-mentioned drawings of this application are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products, or devices.

[0036] Embodiment 1: In this embodiment, a method for monitoring the actions of medical staff and preventing fatigue is provided, aiming to detect whether there are problems with the action postures of medical staff during the work process, so as to reduce physical fatigue and overwork caused by bad actions.

[0037] Refer to Figure 1 , Figure 1 is a flowchart of a method for monitoring the actions of medical staff and preventing fatigue according to an embodiment of the present invention. As Figure 1 shown, the method includes the following steps:

[0038] S1. Obtain the posture images of medical staff from different perspectives collected by a high-precision camera, and construct an original data set;

[0039] The layout of the high-precision camera can refer to Figure 2 to ensure that the camera shooting range should cover the working area of medical personnel, so as to be able to comprehensively and accurately capture various action postures of medical staff during the work process.

[0040] S2. Perform data augmentation processing on each image in the original data set to increase data diversity and expand the data set;

[0041] The data augmentation processing includes rotation, adjustment of brightness and darkness changes, and adding noise, etc., aiming to simulate various possible shooting conditions and environmental changes, such as light, angle, noise, etc., so that the model can better adapt to various situations in actual applications.

[0042] In the embodiment of the present invention, the data augmentation processing includes:

[0043] S21, with the clockwise direction as the positive direction, rotate each image by ±30°, ±45°, ±60°, and ±90° respectively;

[0044] By rotating the images, the pose changes under different perspectives can be simulated, which helps the model learn the human pose features at different angles. In this embodiment, each original image is rotated into 8 different angles (including the original unrotated version, a total of 9), and the rotated images will be used as new training samples to increase the diversity of the dataset.

[0045] S22, adjust the random brightness and darkness of each image, and screen the images after the brightness and darkness changes to remove the images with unqualified brightness;

[0046] By adjusting the brightness and darkness, the shooting scenes under different lighting conditions can be simulated, which helps the model adapt to the images under different lighting intensities and improve its robustness in the real environment. The random brightness and darkness adjustment can be achieved by randomly adjusting the brightness, contrast, etc. of the image.

[0047] In addition, to avoid the loss of image information caused by overexposure or overdarkness, the embodiment of the present invention also screens the images after the brightness change. By checking the contrast, brightness and other indicators of the image, or whether the key information in the image is still visible, the overexposed or overdark images are removed to ensure that the image quality in the dataset meets the training requirements.

[0048] S23, add salt-and-pepper noise, Gaussian noise, and speckle noise processing to each image respectively.

[0049] Adding noise is to simulate the interference that the image may receive during transmission or acquisition. By adding noise, the model will learn how to accurately recognize human poses in the presence of noise, thereby improving its detection ability in low-quality images and enhancing the robustness of the model. Among them, salt-and-pepper noise, also known as impulse noise, randomly adds black and white noise points in the image, sets some pixels to black or white, simulating random bright or dark points in the image; Gaussian noise randomly changes the pixel values in the image according to the Gaussian distribution (normal distribution), simulating random fluctuations in the image; speckle noise simulates the speckle-like interference in the image, simulating the texture changes in the image.

[0050] Through the above data augmentation steps, the diversity and complexity of the dataset can be increased, which helps the model learn more generalized feature representations and adapt to various changing conditions, thereby improving its stability and reliability in real application scenarios. For example, in a specific embodiment, after the above data augmentation processing, the original dataset is expanded to 3000 images.

[0051] S3, use the labelme software to mark the human key points on each image;

[0052] The human body key points are the main joint points of the human body, such as the head, shoulders, elbows, wrists, hips, knees, and ankles, etc. According to specific application requirements, other key points can also be included, such as the fingertips, toes, or facial feature points, etc. By annotating the human body key points, the model can accurately identify and understand the positions and relationships of various parts of the human body in space.

[0053] In one embodiment, the number of human body key points is set to 13, which are the head (1), neck (1), shoulders (2 on the left and right), elbows (2 on the left and right), wrists (2 on the left and right), hips (1), knees (2 on the left and right), and ankles (2 on the left and right). The labelme software is used to annotate these 13 human body key points for each image in the dataset. These images with annotated key points will be used as model training samples to train the DNPose - YOLOv8 network model, and the model will learn how to identify and locate the human body key points according to the input image.

[0054] Furthermore, before sending the images to the DNPose - YOLOv8 network for training, the images need to be pre - processed, including unifying the image size and normalizing the pixels. By using image scaling algorithms (such as bilinear interpolation, nearest - neighbor interpolation, etc.), the size of the images is uniformly adjusted to 640 * 640 pixels to meet the input requirements of the DNPose - YOLOv8 network; and the image pixel values are scaled to between 0 and 1. The RGB color values of each pixel point (usually in the range of 0 - 255 integers) are converted, and each pixel value is divided by 255 to make it a floating - point number between 0 and 1. After the image pre - processing step, each image in the dataset will have a unified size (640x640) and pixel value range (0 to 1). Such a dataset is more suitable for training the DNPose - YOLOv8 network, and the network can more easily learn and extract features in the images.

[0055] The dataset after unifying the size and normalizing the pixels will be divided into a training set, a validation set, and a test set. The training set is used to train the network model, the validation set is used to evaluate the performance of the model and adjust the hyperparameters during the training process, and the test set is used to evaluate the final performance of the model after training.

[0056] S4, Input the annotated image dataset into the DNPose - YOLOv8 network for training to obtain the trained DNPose - YOLOv8 network. The trained DNPose - YOLOv8 network is used to identify the target object in the input image and locate the positions of each key point of the target object;

[0057] DNPOSE-YOLOv8 is an advanced network model that integrates human pose detection and object detection functions. It can not only efficiently identify the target objects (i.e., medical staff) in the image, but also accurately locate the positions of each key point of the human body.

[0058] Referring to Figure 3 , Figure 3 , as shown in the network structure diagram of DNPOSE-YOLOv8, the DNPOSE-YOLOv8 network in the embodiments of the present invention is an improvement based on YOLOv8-pose, including:

[0059] 1. Adopting a more lightweight detection head LW-Pose ( Figure 4 is the structure diagram of LW-Pose) to accelerate the inference process of the network;

[0060] 2. Introducing an asymmetric design into the network structure to reduce the computational complexity and the number of model parameters;

[0061] The asymmetric design is an optimization method for deep learning network structures, which refers to introducing asymmetry in the convolution kernel, pooling window, or connection mode of the network. By using convolution kernels of different sizes, asymmetric pooling windows, or adjusting the connection mode of network layers, such as using non-square convolution kernels 1*3 or 3*5 instead of the common symmetric design (such as standard 3*3 or 5*5 convolution kernels), the computational amount is reduced, the inference speed and feature expression ability are improved, and the computing resources are utilized more efficiently.

[0062] 3. Using 3×3 depthwise separable convolution DS_Conv instead of the traditional 3×3 convolution in each branch to reduce the number of parameters;

[0063] DS_Conv decomposes the convolution operation into Depthwise Convolution and pointwise convolution. After Depthwise Convolution, the number of feature maps is the same as the number of channels in the input layer, and it is impossible to expand the feature maps. Since this operation is performed independently on each channel of the input layer and cannot effectively utilize the feature information of different channels at the same spatial position, pointwise convolution is needed to combine these feature maps to generate new feature maps. However, since it cannot effectively utilize the feature information of different channels at the same spatial position, pointwise convolution will be used later for dimension expansion. The operation of pointwise convolution is very similar to the conventional convolution operation. By applying a 1×1 convolution kernel between different channels to mix the information between channels, it does not change the spatial dimension of the feature maps but only changes the number of channels. Therefore, the pointwise convolution operation weights and combines the previous maps in the depth direction to generate new feature maps. Secondly, due to the reduction in the number of parameters, depthwise separable convolution is more computationally efficient and suitable for use in environments with limited resources. Finally, by combining the information of different channels through pointwise convolution, a certain feature extraction ability is maintained.

[0064] For example, when the convolution kernel size is D K * D K , the number of input channels is M , the number of output channels is N , and the size of the output feature map is D F * D F , after standard convolution, it can be calculated that:

[0065] The number of parameters is: ;

[0066] The amount of computation is: ;

[0067] While using DS_Conv with the same size in this embodiment, it can be calculated that:

[0068] The number of parameters is: ;

[0069] The amount of computation is: ;

[0070] It can be seen by comparison that in the embodiments of the present invention, depthwise separable convolution is adopted, and the number of parameters and the amount of computation are significantly less than those of standard convolution. Therefore, the model parameters can be effectively reduced and the detection speed can be improved.

[0071] 4. Add a feature focusing and diffusion pyramid structure ( Figure 5 as shown in the figure of the feature focusing and diffusion pyramid structure) to the HEAD network structure. The pyramid structure extracts the features of each layer and performs feature fusion, and then diffuses the fused features to each feature dimension, so that each feature dimension has more comprehensive feature information, thereby realizing the feature focusing and diffusion of the pyramid structure and improving the detection accuracy.

[0072] The feature focusing and diffusion pyramid structure adopts a customized feature focusing module and a feature diffusion mechanism. Among them, Figure 6 as shown in the figure of the feature focusing and diffusion module. The feature focusing module aims to capture and integrate feature information from different scales and generate a feature map with context information. The feature focusing module accepts inputs of three scales, that is, targets of different sizes are detected at different feature map scales. It internally contains an Inception-Style module, and the Inception-Style module uses a group of parallel depth convolutions to capture information across multiple scales; the feature diffusion mechanism aims to diffuse the context information features generated by the feature focusing module to each detection scale to achieve cross-scale feature fusion and enhancement and improve the detection accuracy.

[0073] An ablation experiment comparison is carried out using the original YOLOv8-pose model and the DNPose-YOLOv8 model proposed in the embodiments of the present invention. The experimental results are as Figure 7 shown, Figure 7 where Pose-mAP@0.5 is the mAP (mean average precision) for key point detection of the model at the IoU (Intersection over Union) = 0.5 threshold, Pose-mAP@0.95 is the mAP (mean average precision) for key point detection of the model at the IoU = 0.95 threshold, FLOPs / G represents the complexity of the network, and Params / M is the number of parameters of the network, with the unit of M. Compared with the original YOLOv8-pose model, the DNPose-YOLOv8 model proposed in the embodiments of the present invention has advantages in various indicators, verifying that the improvement is effective.

[0074] S5. Real-time collect the working images of medical staff, and use the trained DNPose-YOLOv8 network to identify the coordinate information of each key point of the medical staff's body during work

[0075] The DNPose - YOLOv8 network generates key points for pose detection through the detection head. The feature map output by the network is converted into the coordinates and sizes of bounding boxes, key points for pose detection, and the probability density distribution is converted into class probabilities to obtain the final bounding box coordinates, class probabilities, confidence scores, and key point coordinate information. After the image passes through the network output, through image cropping technology, each bounding box and the key point coordinates contained in the box are saved to a specified folder. Each file contains an image of a healthcare worker's working pose and the corresponding key point coordinates.

[0076] S6. Calculate the positional relationship between key points according to the coordinate information, and judge whether there is a risk in the working pose of the healthcare worker according to the positional relationship between the key points;

[0077] Traverse the images in the specified folder (i.e., the folder where the output results of the DNPose - YOLOv8 network are located), read each image and its corresponding key point coordinate information, calculate the positional relationship between key points according to the coordinate information of the key points output by the model. The positional relationship includes distance and angle. According to the calculated positional relationship, judge whether there is a risk in the working pose of the healthcare worker.

[0078] In one embodiment, S6 includes risk assessment based on the normal distance between the left and right ankles. Specifically:

[0079] S611. Obtain the coordinates of the left and right ankles output by the DNPose - YOLOv8 network, and calculate the normal distance between the left and right ankles;

[0080] S612. Judge whether the normal distance between the left and right ankles exceeds a preset threshold. If it exceeds, it is considered that there is a risk in the working pose of the healthcare worker.

[0081] The normal distance refers to the distance between the left and right ankles in the direction perpendicular to the ground. This distance can be used to evaluate the standing stability of the healthcare worker or the risk of falling. Real - time monitor the standing situation of the detected healthcare worker, and judge through spatial position information. When the difference in the normal distance between the two ankles is greater than the preset threshold (such as 20 mm), it is determined to be standing on one foot and there is a risk.

[0082] In another embodiment, S6 includes risk assessment based on joint angles. Specifically:

[0083] Use the vector method to calculate the joint angles for the key point information in each image. Vectors have directionality and translatability. Therefore, any two non - overlapping points 、 in the human body coordinate system can be converted into vectors in the reference coordinate system in the form of coordinate vectors through coordinate transformation .

[0084] S621. For the three key points that make up the joint angle 、 、 , where is the joint point, 、 are the key points adjacent to the joint point . The included angle formed by the line segment and is the joint angle of the joint . According to the 、 、 coordinate values, calculate the vectors 、 :

[0085] ;

[0086] ;

[0087] S622. Calculate the cosine value of the included angle 、 between the vectors through the dot product and modulus length of the vectors ω :

[0088] ;

[0089] Let , then the joint angle of the joint ;

[0090] S623. Determine whether the calculated joint angle is within the preset range. If it is not within the preset range, it is considered that there is a risk in the working posture of the medical staff.

[0091] According to medical or ergonomic knowledge, set the preset range for each joint angle. For example, the normal cervical flexion is generally 35 to 45 degrees. If the calculated joint angle is not within the preset range, it is considered that there is a risk in the working posture of the medical staff, such as excessive joint extension or flexion.

[0092] Understandably, in addition to the above-mentioned ankle distance and joint angle, in other embodiments, the positional relationship between the left and right knees can also be calculated, such as whether they overlap, cross, etc., to determine whether there are bad postures such as crossing legs, or the bending amplitudes of the cervical spine and waist can be calculated to determine whether there are postures such as hunchback.

[0093] S7. Generate a reminder message when a risk posture is detected, instructing the medical staff to optimize their working postures or movements to reduce occupational fatigue.

[0094] The reminder information includes text, sound, or images, and can be conveyed to medical staff through means such as a display screen, a speaker, or a mobile device.

[0095] In one embodiment, risk levels are set, including negligible risk, low risk, medium risk, and high risk. Among them, negligible risk corresponds to levels 1 - 2 in the RULA (Rapid Upper Limb Assessment) scale, low risk corresponds to levels 3 - 4 in the RULA scale, medium risk corresponds to levels 5 - 6 in the RULA scale, and high risk corresponds to level 7 in the RULA scale; different risk ranges are delimited. When the calculated positional relationship is at different risk levels, reminder information corresponding to the risk level is displayed in different - colored fonts through the display interface. Among them, negligible risk corresponds to white - colored font display, low risk corresponds to yellow - colored font display, medium risk corresponds to orange - colored font display, and high risk corresponds to red - colored font display. By distinguishing different colors, a more intuitive warning can be given to medical staff.

[0096] Through the above steps, the present invention can efficiently, real - time, and accurately complete the posture assessment of medical staff and remind of risky postures. The present invention has the advantages of high automation, detailed objectivity, and can be widely applied in medical scenarios to help medical staff optimize their working postures and movements (such as dispensing medicine actions), reduce occupational fatigue, and improve work efficiency and safety.

[0097] Embodiment 2: For the same inventive concept, in this embodiment, a device for monitoring medical staff's actions and preventing fatigue is also provided to implement the above - mentioned embodiment and preferred implementation manners. Those that have been described will not be repeated here.

[0098] Specific examples in this embodiment can refer to the examples described in the above - mentioned embodiment and optional implementation manners, and will not be repeated here.

[0099] The embodiment of the present invention also provides a storage medium, in which a computer program is stored. Among them, the computer program is set to execute the steps in any one of the above - mentioned method embodiments when running.

[0100] Optionally, in this embodiment, the above - mentioned storage medium may include, but is not limited to: USB flash drives, read - only memory (ROM for short), random access memory (RAM for short), mobile hard disks, magnetic disks, or optical discs and other various media that can store computer programs.

[0101] The serial numbers of the above - mentioned embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0102] In the above embodiments of the present invention, the descriptions of the respective embodiments have their own focuses. For the parts not elaborated in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0103] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for monitoring the movements of medical staff and preventing fatigue, characterized in that: include: Obtain posture images of medical staff from different perspectives captured by high-precision cameras and construct the original data set; Perform data augmentation on each image in the original dataset to increase data diversity and expand the dataset; Use labelme software to mark the key points of the human body in each image; Input the labeled image data set into the DNPOSE-YOLOv8 network for training to obtain a trained DNPOSE-YOLOv8 network, wherein the trained DNPOSE-YOLOv8 network is used to identify the target object in the input image and locate the position of each key point of the target object; Collecting working images of medical staff in real time, and using the trained DNPOSE-YOLOv8 network to identify coordinate information of key points of the medical staff's body when they are working; Calculate the positional relationship between the key points according to the coordinate information, and determine whether there is a risk in the working posture of the medical staff according to the positional relationship between the key points; Generates reminder information when risky postures are detected, instructing medical staff to optimize their working postures or movements to reduce occupational fatigue; The DNPOSE-YOLOv8 network is an improvement on the YOLOv8-pose. The improvements include: Use a lighter detection head, LW-Pose, to speed up the network’s reasoning process; Introducing asymmetric design into the network structure to reduce computational complexity and model parameter count; In each branch, 3×3 depth-wise separable convolution DS_Conv is used instead of traditional 3×3 convolution to reduce the number of parameters; A feature focused diffusion pyramid structure is added to the HEAD network structure. The pyramid structure extracts features from each layer and performs feature fusion, and then diffuses the fused features to each feature dimension, so that each feature dimension has more comprehensive feature information, thereby realizing feature focused diffusion of the pyramid structure and improving detection accuracy. The pyramid structure adopts a feature focusing module and a feature diffusion mechanism. The feature focusing module is designed to capture and integrate feature information from different scales and generate a feature map with contextual information. The feature focusing module accepts inputs of three scales and contains an Inception-Style module. The Inception-Style module uses a set of parallel deep convolutions to capture information across multiple scales. The feature diffusion mechanism is designed to diffuse the contextual information features generated by the feature focusing module to each detection scale to achieve cross-scale feature fusion and enhancement and improve detection accuracy.

2. The method for monitoring the movement and preventing fatigue of medical staff according to claim 1, characterized in that: The data augmentation process for each image in the original dataset includes: Taking clockwise as the positive direction, each image is rotated by ±30°, ±45°, ±60°, and ±90° respectively; Perform random brightness and darkness adjustment on each image, and screen the images after brightness and darkness change to remove images whose brightness does not meet the requirements; Salt and pepper noise, Gaussian noise, and speckle noise are added to each image.

3. The method for monitoring the movement and preventing fatigue of medical personnel according to claim 1, characterized in that: Before inputting the labeled image dataset into the DNPOSE-YOLOv8 network for training, the method further includes: The images in the dataset are uniformly resized to 640*640 and the pixel values ​​are scaled to between 0 and 1.

4. The method according to claim 1, characterized in that: The positional relationship between the key points includes the distance between the key points, and the calculating the positional relationship between the key points according to the coordinate information, and judging whether there is a risk in the working posture of the medical staff according to the positional relationship between the key points includes: Get the coordinates of the left and right ankles output by the DNPOSE-YOLOv8 network and calculate the normal distance between the left and right ankles; Determine whether the normal distance between the left and right ankles exceeds the preset threshold. If so, it is considered that the medical staff's working posture is risky.

5. The method according to claim 1, characterized in that The positional relationship between the key points includes the joint angles, and the calculating the positional relationship between the key points according to the coordinate information, and judging whether there is a risk in the working posture of the medical staff according to the positional relationship between the key points includes: For the three key points that make up the joint angle , , ,in, For the joint point, , For joint points Adjacent key points, line segments and The angle formed is the joint The joint angles are based on , , Coordinate value calculation vector , : ； ； In the formula, Key points The x, y, and z coordinate values ​​of Key points The x, y, and z coordinate values ​​of Key points The x, y, and z coordinate values ​​of Calculate vectors by vector dot product and modulus and The angle between ω The cosine of : ； make , then the joint Joint angle ; It is determined whether the calculated joint angle is within the preset range. If not, it is considered that the medical staff's working posture is risky.

6. The method according to claim 1, characterized in that The generating of reminder information when a risky posture is detected includes: Set risk levels, including negligible risk, low risk, medium risk and high risk, where negligible risk corresponds to RULA level 1-2, low risk corresponds to RULA level 3-4, medium risk corresponds to RULA level 5-6, and high risk corresponds to RULA level 7; The display interface displays the corresponding risk level reminder information in different color fonts, where negligible risks are displayed in white fonts, low risks are displayed in yellow fonts, medium risks are displayed in orange fonts, and high risks are displayed in red fonts.

7. A medical staff motion monitoring and fatigue prevention device, characterized in that: The medical staff motion monitoring and fatigue prevention device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps of the method described in any one of claims 1 to 6 are implemented.

8. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

Images