Human body working posture evaluation method and system based on deep learning

Abstract

The invention discloses a human body working posture evaluation method and system based on deep learning, and the method comprises the steps: uploading a working posture image of an operator, detecting the working posture image through a key point detection model, and obtaining key point information; calculating a reliability index of the working attitude image, and judging whether the image is reliable according to the reliability index; calculating a human body part included angle by using a cosine law according to the key point information, and setting an empirical coefficient for optimizing a human body part included angle value; removing values exceeding a preset range in the human body part included angle, and calculating an ergonomic score of the working posture according to the human body part included angle by using a working posture evaluation method; the working posture with the ergonomic score larger than a preset value is found, and an optimization scheme for preventing musculoskeletal diseases and a scheme for relieving pain are provided. Therefore, the efficiency and the accuracy of human body working posture evaluation are improved. The deep learning-based human body working posture evaluation system is used for realizing the method, and is convenient for a user to operate and use.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and more specifically, to a method and system for evaluating human working posture based on deep learning. Background Art

[0002] The symptoms of musculoskeletal disorders (MSDs) are defined as pain in one or more body areas. They are the accumulation of minor injuries caused by poor posture, frequent and long-term repetitive movements, and maintaining the same posture for a long time in the workplace. With the extension of human life expectancy and the trend of population aging, the incidence of MSDs will increase year by year and become a serious occupational health disease. At present, the body posture analysis of people in the workplace and the investigation and research of potential MSD patients can be carried out through the MSDs assessment method, so as to detect and correct employees' poor postures at an early stage and formulate preventive measures for high-risk postures obtained from research and analysis, so as to reduce the incidence of MSDs.

[0003] The MSDs evaluation methods in the prior art mainly include: self-assessment method, direct measurement evaluation method and work observation evaluation method. The self-assessment method collects data on employees' body pain, fatigue and other discomfort parts by interviewing employees and filling out questionnaires. However, there are problems such as complex data collection and data analysis of questionnaires and poor data accuracy; the direct measurement evaluation method uses sensors to directly measure and collect data from employees at work through instruments such as motion capture devices. However, there are problems such as the movement of employees will be affected when they wear instruments at work, and the measurement instruments are expensive and have high economic costs; the work observation evaluation method is that observers observe, record and analyze the adverse factors such as posture, load and duration of employees during work. However, there are problems such as the complexity of the observer's evaluation process and the observation of the workers' working postures one by one, resulting in low evaluation efficiency.

[0004] In summary, how to provide a method for efficiently and accurately evaluating human working posture in a variety of working scenarios is a problem that needs to be solved urgently by those skilled in the art. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide a method and system for evaluating human working posture, which can utilize a deep learning model to efficiently and accurately evaluate human working posture in a variety of work scenarios, and provide solutions for preventing musculoskeletal diseases based on working posture, which is easy to operate and use.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] The present invention provides a method for evaluating human working posture based on deep learning, comprising the following steps:

[0008] Step S1, uploading a working posture image of an operator, wherein the working posture image includes a working posture video and a working posture picture, and detecting the working posture image through a key point detection model to obtain key point information, wherein the key point information includes a horizontal coordinate, a vertical coordinate, and a confidence level of the key point;

[0009] Step S2, calculating the reliability index of the working posture image, and proceeding to the next step when the reliability index is greater than or equal to a preset value, and returning to step S1 when the reliability index is less than the preset value;

[0010] Step S3, calculating the angle of the human body part using the cosine theorem according to the key point information, and setting an empirical coefficient, wherein the empirical coefficient is used to optimize the value of the angle of the human body part;

[0011] Step S4, removing the values ​​of the human body part angles that exceed a preset range, and calculating the ergonomic score of the working posture according to the human body part angles using a work posture evaluation method, wherein the ergonomic score includes a human body part score and an overall score;

[0012] Step S5, finding the working posture whose ergonomic score is greater than a preset value, providing an optimization plan for the working posture, wherein the optimization plan is used to prevent musculoskeletal diseases and provide a pain relief plan for the human body area corresponding to the musculoskeletal disease.

[0013] Optionally, the reliability index can be selectively set to a confidence index and a shooting angle index, the confidence index is used to determine whether the reliability of the key point detection model's evaluation and prediction of the working posture image meets the requirements, and the shooting angle index is used to determine whether the angle of the human body part calculated based on the working posture image meets the requirements.

[0014] Optionally, an average value of all the confidence levels is calculated; when the average value is greater than or equal to the confidence index, the working posture image meets the requirements; and when the average value is less than the confidence index, the working posture image is uploaded again.

[0015] Optionally, the key points include a right shoulder and a left shoulder, and the angle between the vectors of the right shoulder and the left shoulder and the optical axis of the image is set as a shooting angle; when the shooting angle is greater than or equal to the shooting angle index, the working posture image meets the requirements; when the shooting angle is less than the shooting angle index, the working posture image is uploaded again.

[0016] Optionally, the empirical coefficient is set according to the error between the actual angle of the human body part and the angle of the human body part calculated by the key point detection model, and the empirical coefficient includes a neck empirical coefficient, an upper arm empirical coefficient, a forearm empirical coefficient, a back empirical coefficient and a leg empirical coefficient. The neck empirical coefficient is multiplied by the neck angle, the upper arm empirical coefficient is multiplied by the upper arm angle, the forearm empirical coefficient is multiplied by the forearm angle, the back empirical coefficient is multiplied by the back angle, and the leg empirical coefficient is multiplied by the leg angle.

[0017] Optionally, the key point detection model includes an improved YOLOv8-pose model, and the improved YOLOv8-pose model is specifically implemented as follows: the backbone network 100 introduces shifted convolution to improve the feature extraction block 101, forming a feature extraction block 102 containing shifted convolution, the neck network 110 introduces a star block 111 for feature fusion, and the head network 120 introduces a shared detection head 121, and the shared detection head 121 includes a shared convolution block 1211.

[0018] Optionally, the optimization scheme reduces the ergonomic score, the corresponding human body areas include the neck, back, legs, arms and wrists, and the pain relief scheme includes a shoulder and neck relief scheme, a waist and back relief scheme, a leg relief scheme, an arm relief scheme, a wrist relief scheme and a whole body relief scheme.

[0019] The present invention also provides a human working posture evaluation system based on deep learning, which is used to implement the human working posture evaluation method based on deep learning, including an input module, a human key point detection module, an input image judgment module, an ergonomic score calculation module and a pain prevention and relief module; the input module inputs a working posture image and working parameters to obtain information for human working posture evaluation; the human key point detection module is used to obtain the human key point information of the working posture image; the input image judgment module calculates a reliability index, and the reliability index is used to judge whether the working posture image is reliable; the ergonomic score calculation module calculates the angle of human body parts according to the human key point information, and calculates the ergonomic score of the working posture according to the angle of human body parts using a work posture evaluation method; the prevention and pain relief module judges whether there is a problem with the working posture according to the ergonomic score, and provides a solution for preventing musculoskeletal diseases if there is a problem.

[0020] Optionally, the input module includes: an image input unit for inputting an image of a working posture; a working parameter input unit for inputting numerical values ​​including workload and whether the neck, waist, and wrist are twisted during the working action; and an empirical coefficient input unit for selectively setting the numerical value of the empirical coefficient within a preset range.

[0021] Optionally, the prevention and pain relief module includes a work posture optimization unit and a pain relief unit, the work posture optimization unit provides a solution to reduce the ergonomic score, and the pain relief unit provides a pain relief solution for the human body area corresponding to the ergonomic score greater than a preset score.

[0022] The human working posture evaluation method based on deep learning provided by the present invention uploads the working posture image of the operator, detects the working posture image through a key point detection model, obtains key point information, and calculates the reliability index of the working posture image, so that the input working posture image can be guaranteed to have high reliability and can be used for the next step of calculation; the cosine theorem is used to calculate the angle of human body parts, and the empirical coefficient for optimizing the angle of human body parts is calculated, so that the angle of human body parts obtained based on the key point detection model is more accurate; the working posture evaluation method is used to calculate the ergonomic score of the working posture according to the angle of human body parts, and the working posture with an ergonomic score greater than a preset value is found, an optimization plan for preventing musculoskeletal diseases is provided, and a plan for relieving pain is provided, so that the working posture with hidden dangers can be quickly found, and an effective solution can be provided at the same time.

[0023] The human working posture assessment system based on deep learning provided by the present invention includes an input module, a human key point detection module, an input image judgment module, an ergonomic score calculation module and a pain prevention and relief module. It can use the human posture assessment model to efficiently and accurately assess the human working posture in a variety of work scenarios, and can set different parameters according to the actual working posture, which is convenient for user operation and use, and has good practicality and economy. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0025] Figure 1 A flowchart of the steps of the method for evaluating human working posture based on deep learning provided by the present invention;

[0026] Figure 2 A schematic diagram of the improved YOLOv8-pose model provided by the present invention;

[0027] Figure 3 for Figure 2 Schematic diagram of the provided head network detection head;

[0028] Figure 4 A schematic diagram of a human body working posture assessment system based on deep learning provided by the present invention;

[0029] Figure 5 A flowchart of a human body working posture assessment system based on deep learning provided by the present invention;

[0030] in:

[0031] 100-backbone network, 101-feature extraction block, 102-feature extraction block with shifted convolution, 103-convolution block, 104-fast spatial pyramid pooling block, 110-neck network, 111-star block, 112-upsampling, 113-feature splicing, 120-head network, 121-shared detection head, 1211-shared convolution block, 1212-feature map, 1213-channel number conversion, 200-human working posture evaluation system, 210-input module, 220-human key point detection module, 230-input image judgment module, 240-ergonomic score calculation module, 250-pain prevention and relief module. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] In order to enable those skilled in the art to better understand the scheme of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0034] The human body working posture assessment method based on deep learning provided by the embodiment of the present invention can be found in the attached manual. Figure 1 , including the following steps:

[0035] Step S1, upload the working posture image of the operator, the working posture image includes the working posture video and the working posture picture, detect the working posture image through the key point detection model, and obtain the key point information, the key point information includes the horizontal coordinate, the vertical coordinate and the confidence of the key point; Step S2, calculate the reliability index of the working posture image, when the reliability index is greater than or equal to the preset value, proceed to the next step, when the reliability index is less than the preset value, return to step S1; Step S3, use the cosine theorem to calculate the angle of the human body part according to the key point information, set the empirical coefficient, and the empirical coefficient is used to optimize the value of the angle of the human body part; Step S4, remove the value of the human body part angle that exceeds the preset range, and use the work posture evaluation method to calculate the ergonomic score of the working posture according to the human body part angle, and the ergonomic score includes the human body part score and the overall score; Step S5, find the working posture with an ergonomic score greater than the preset value, and provide an optimization plan for the working posture. The optimization plan is used to prevent musculoskeletal diseases and provide a pain relief plan for the corresponding human body area of ​​musculoskeletal diseases.

[0036] The key point detection model divides the working posture video into pictures frame by frame, detects the key point information of the human body for each frame, and obtains the key point information of each picture. The reliability index is used to determine whether the input working posture image is reliable. The reliability index can be set according to the confidence inferred by the key point detection model to determine whether the detected key points can be used for the next calculation. It can also be set according to the angle of the input working posture image to determine whether the ergonomic score calculated from the image at this angle is reliable. The reliability index is compared with the preset value, which is set according to the actual situation. The angle of the human body part calculated by the human key point detection model is compared with the angle of the human body part in actual work to obtain the empirical coefficient, so that the angle of the human body part calculated based on the model is closer to the actual angle. The empirical coefficient is set according to the actual comparison situation.

[0037] The cosine theorem formula for calculating the angle between human body parts is:

[0038]

[0039] In formula 1, the coordinates of the three key points A, B, and C are (x 1 ,y 1 )、(x 2 ,y 2 ) and (x 3 ,y 3 ), x 1 、x 2 、x 3 is the horizontal coordinate of the three key points, y 1 ,y 2 ,y 3are the vertical coordinates of the three key points. Since the range of motion of most people's joints is usually limited, the angle of motion of each joint is within a certain range. By removing and prompting the values ​​of the angles of human body parts that exceed the preset range, the error of key point detection can be reduced. The value of the preset range refers to the angle range of common human joints. For example, the range of the shoulder joint to raise the arm forward is approximately between 0° and 180°, the range of the backward extension is approximately between 0° and 50°, the range of the elbow joint to extend and flex the forearm is approximately between 0° and 150°, the range of the knee joint bending is approximately between 0° and 135°, and the range of the cervical spine to raise and lower the head is approximately between 0° and 45°.

[0040] The work posture assessment method is that the observer observes, records and analyzes the adverse factors such as the posture, load and duration of the employees during the work process to reflect the risk level of musculoskeletal diseases in the work. Specifically, it includes the Ovako working posture analysing system (OWAS), rapid upper limb assessment (RULA) and rapid entire body assessment (REBA). The scores of human body parts include the neck score, back score, leg score, forearm score, upper arm score, wrist score, etc. The overall score is obtained according to the scores of human body parts. When the incoming work posture image is a video, a time series diagram corresponding to the angle of each part of the human body and the ergonomic score can be drawn to facilitate the identification of bad postures and parts of the human body with health risks. The optimization plan is obtained by reducing the range of human body parts and thus reducing the ergonomic score. When the score of a human body part is higher than the preset value, the corresponding part of the human body may have hidden dangers, leading to muscle pain and other problems. The pain relief plan can provide an effective solution to reduce muscle pain.

[0041] The reliability index can be optionally set as the confidence index and the shooting angle index. The confidence index is used to determine whether the reliability of the key point detection model's evaluation and prediction of the working posture image meets the requirements. When the uploaded image is a working posture image, the confidence of the image prediction is directly compared with the confidence index. When the uploaded image is a working posture video, the average of all confidences is calculated based on the confidence of each frame of the image, and the average confidence is compared with the confidence index. When the confidence is greater than or equal to the confidence index, the working posture image meets the requirements. When the confidence is less than the confidence index, the working posture image is re-uploaded. The confidence is the degree of certainty of the model in the prediction result, that is, the probability that the model believes that the prediction is correct. Usually, the confidence value is between 0 and 1, and the confidence index can be set according to the actual situation, such as 0.85.

[0042] The shooting angle index is used to determine whether the angle of the human body part calculated based on the working posture image meets the requirements. The key points include the right shoulder and the left shoulder. The angle between the vectors of the right shoulder and the left shoulder and the optical axis of the image is set as the shooting angle. When the shooting angle is greater than or equal to the shooting angle index, the working posture image meets the requirements. When the shooting angle is less than the shooting angle index, the working posture image is uploaded again.

[0043] The reliability of the shooting angle in calculating the angle between human body parts is determined by the four main parts: neck, back, arms and legs. The calculation formula of the shooting angle is:

[0044]

[0045] In formula 2 The vectors representing the right and left shoulders, The optical axis is represented by the vectors of the right shoulder and the left shoulder, which are calculated based on the coordinates of the right shoulder and the left shoulder. In a two-dimensional image, the optical axis can be represented by the vector between the center of the image and the center of the human body, and the vector between the center of the image and the center of the human body is calculated based on the coordinates of the center of the image and the center of the human body. In a three-dimensional image, if a binocular camera is used to obtain three-dimensional depth information, or through three-dimensional model reconstruction, the spatial position of the optical axis can be further calculated. The main view of the working posture, that is, shooting directly at the worker, has a shooting angle of 90°, and the side view of the working posture, that is, shooting the left and right sides of the worker, has a shooting angle of 0°. Usually, when the angle is between 45° and 90°, the accuracy of the angle of the human body parts calculated based on the human key point detection model is higher. The purpose of calculating the shooting angle is to accurately measure the posture of the worker, so that unnecessary low-reliability human posture data can be shielded and the accuracy of the evaluation method can be improved.

[0046] The shooting angle index can also be set according to the distance between the right shoulder and the left shoulder and the ratio between the left shoulder and the left hip or the right shoulder and the right hip. The shooting angle is 90° and the ratio is smaller, while the shooting angle is 0° and the ratio is larger. There are many ways to set the shooting angle index, which are not limited in this article.

[0047] The empirical coefficient is set according to the error between the actual angle of the human body part and the angle of the human body part calculated by the key point detection model. The empirical coefficients include the neck empirical coefficient, the upper arm empirical coefficient, the forearm empirical coefficient, the back empirical coefficient and the leg empirical coefficient. The neck empirical coefficient is multiplied by the neck angle, the upper arm empirical coefficient is multiplied by the upper arm angle, the forearm empirical coefficient is multiplied by the forearm angle, the back empirical coefficient is multiplied by the back angle, and the leg empirical coefficient is multiplied by the leg angle.

[0048] The purpose of setting the empirical coefficient is to make the angle of the human body parts calculated by the deep learning-based human work posture assessment method closer to the actual angle of the human body parts. There are many ways to calculate the empirical coefficient, such as using motion capture instruments and other instruments to directly measure and collect work posture data using sensors, calculate the actual angle of the human body parts, compare the actual value and calculated value of the angle of different human body parts, and obtain the empirical coefficient of different parts. The empirical coefficient can be set as a constant and multiplied with the calculated value to obtain a value closer to the actual value. Of course, the empirical coefficient can be calculated and set according to the actual situation, and this article does not make specific restrictions.

[0049] As the instruction manual Figure 2-3 As shown, the key point detection model includes an improved YOLOv8-pose model, and the improved YOLOv8-pose model is specifically implemented as follows: the backbone network 100 introduces shifted convolution to improve the feature extraction block 101, forming a feature extraction block 102 containing shifted convolution, the neck network 110 introduces a star block 111 for feature fusion, and the head network 120 introduces a shared detection head 121, and the shared detection head 121 includes a shared convolution block 1211.

[0050] As the instruction manual Figure 2 As shown, the improved YOLOv8-pose model includes a backbone network 100, a neck network 110 and a head network 120. The backbone network 100 is mainly responsible for extracting basic features from the input image, and is composed of a feature extraction block 101, a feature extraction block 102 containing a shifted convolution, a convolution block 103 and a fast spatial pyramid pooling block 104. The shifted convolution replaces the convolution set in the feature extraction block 101, and constructs a feature extraction block 102 containing a shifted convolution, which can improve the feature extraction capability of the backbone network 100. The neck network 110 has the function of multi-scale feature fusion, and is composed of a star block 111, an upsampling 112 and a feature splicing 113, wherein the star block 111 can improve the efficiency of feature fusion. The head network 120 is provided with a shared detection head 121, which performs target detection and key point detection tasks, and performs bounding box positioning and key point regression.

[0051] As the instruction manual Figure 3 As shown, feature maps 1212 of different scales first undergo channel number conversion 1213 to convert the number of channels of the feature map 1212 to the same value, and then enter the shared convolution block 1211. This can effectively reduce the number of model parameters, reduce computational complexity, make the detection head more lightweight, and improve the detection speed of the key point detection model.

[0052] The optimization plan reduces the ergonomic score. The corresponding human body areas include the neck, back, legs, arms and wrists. The pain relief plans include shoulder and neck relief plan, waist and back relief plan, leg relief plan, arm relief plan, wrist relief plan and full body relief plan.

[0053] The optimization plan is set according to the actual working posture. For example, by adjusting the bad working posture, you can avoid bad postures such as excessive flexion, twisting or lateral bending at work, and reduce the pressure on the spine and joints; choose an ergonomic chair that supports the waist and can adjust the seat height, backrest angle, etc. to keep the body in the appropriate posture when working; place items within the reach of the arm to reduce the frequency of overstretching or bending over to get items; use auxiliary tools to work, reduce the burden on the body, and control working hours and rest.

[0054] Pain relief programs include stretching exercises for the neck, back, legs, arms and wrists, massage and hot compresses for the painful areas, acupuncture or massage according to the TCM acupoints corresponding to the painful areas, and the use of instruments and equipment such as cervical vertebrae therapeutic devices, ergonomic cushions and backrests to relieve pain. When pain exists in multiple areas of the human body, full-body relief can be achieved, including full-body stretching through yoga, doing aerobics such as Ba Duan Jin, Wu Qin Xi, Yi Jin Jing, and full-body massage using a massage chair.

[0055] The human body working posture assessment system based on deep learning provided by the embodiment of the present invention is used to implement the human body working posture assessment method based on deep learning, as shown in the attached manual. Figure 4-5 As shown, the human working posture evaluation system 200 includes an input module 210, a human key point detection module 220, an input image judgment module 230, an ergonomic score calculation module 240 and a prevention and pain relief module 250; the input module 210 inputs a working posture image and working parameters to obtain information for human working posture evaluation; the human key point detection module 220 is used to obtain the human key point information of the working posture image; the input image judgment module 230 calculates a reliability index, and the reliability index is used to determine whether the working posture image is reliable; the ergonomic score calculation module 240 calculates the angle of human body parts according to the human key point information, and calculates the ergonomic score of the working posture according to the angle of human body parts using the working posture evaluation method; the prevention and pain relief module 250 determines whether there is a problem with the working posture according to the ergonomic score, and provides a solution for preventing musculoskeletal diseases if there is a problem.

[0056] The human body key point detection module 220 uses the key point detection model to infer and predict the input image. The key point detection model can include a two-dimensional model and a three-dimensional model. The reliability index can be set according to the actual situation, and this article does not make specific restrictions. The ergonomic score calculation module 240 calculates according to the input working parameters, and can optionally calculate the scores of different human body parts and the overall score to facilitate judgment of which human body parts have the highest health risks in bad postures.

[0057] The input module 210 includes: an image input unit for inputting an image of a working posture; a working parameter input unit for inputting numerical values ​​including workload and whether the neck, waist, and wrist are twisted during the working action; and an empirical coefficient input unit for selectively setting the numerical value of the empirical coefficient within a preset range.

[0058] In order to improve the accuracy of the human working posture evaluation system, the input working posture image can be selectively set to an image taken from multiple views, such as a left view, a right view, a main view, a 45° tilted view, etc. Images from different views can better represent the corresponding human body parts, wherein the human body parts include right shoulder, right elbow, right wrist, left shoulder, left elbow, left wrist, right hip, right knee, right ankle, left hip, left knee, left ankle, top of the head and neck. The left view can better represent the left shoulder, left elbow, left wrist, left hip, left knee and left ankle among the human body parts, and the right view can better represent the right shoulder, right elbow, right wrist, right hip, right knee and right ankle among the human body parts. In this way, a suitable view can be selected according to the actual working posture to calculate the scores of the human body parts, and then the scores of the human body parts calculated from different views are sorted, and finally the overall score is calculated for the one with high reliability, thereby improving the accuracy of the ergonomic score calculation, and preventing the inaccurate recognition of some key points due to the occlusion of a single-angle image, thereby causing the problem of inaccurate ergonomic score calculation.

[0059] Since the key point detection model cannot accurately detect the twisting and deflection of human body parts and cannot accurately calculate the ergonomic score, it is possible to perform supplementary calculations by inputting work parameters. The input work parameters are set according to the actual work posture assessment method selected. Taking the whole body rapid assessment as an example, the work parameters include whether the head is twisted, whether the head is deflected, whether the waist is twisted, whether the waist is deflected, workload, whether the wrist is slightly turned upside down (upside down within 15°), whether the wrist is severely turned upside down (upside down more than 15°), whether the wrist is deflected left and right, whether the upper arm is abducted, whether the arm is supported or the body is tilted, and the convenience of holding the object (the object has a handle and is easy to grasp, the hand can grasp the object, the hand cannot grasp the object steadily, and it is difficult to grasp the object without a handle).

[0060] The pain prevention and relief module 250 includes a work posture optimization unit and a pain relief unit. The work posture optimization unit provides a solution for reducing the ergonomic score, and the pain relief unit provides a pain relief solution for the human body area corresponding to the ergonomic score greater than the preset score.

[0061] The pain prevention and relief module 250 first determines whether the input working posture is a bad posture according to the ergonomic score. If the ergonomic score is greater than a preset value, it is determined to be a bad posture, wherein the preset value is set according to the actual working posture evaluation method used. The working posture optimization unit and the pain relief unit can provide optimization solutions and pain relief solutions for human body parts with higher scores among the scores of human body parts, which are specifically set according to actual conditions and are not specifically limited herein.

[0062] The interactive terminal provided by the present invention is described in detail above. The principle and implementation method of the present invention are described in detail using specific examples herein. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. It should be pointed out that for ordinary technicians in this technical field, the present invention can also be improved and modified without departing from the principle of the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A human work posture assessment method based on deep learning, characterized in that: The following steps are involved: Step S1, uploading a working posture image of an operator, wherein the working posture image includes a working posture video and a working posture picture, and detecting the working posture image through a key point detection model to obtain key point information, wherein the key point information includes a horizontal coordinate, a vertical coordinate, and a confidence level of the key point; Step S2, calculating the reliability index of the working posture image, and proceeding to the next step when the reliability index is greater than or equal to a preset value, and returning to step S1 when the reliability index is less than the preset value; Step S3, calculating the angle of the human body part using the cosine theorem according to the key point information, and setting an empirical coefficient, wherein the empirical coefficient is used to optimize the value of the angle of the human body part; Step S4, removing the values ​​of the human body part angles that exceed a preset range, and calculating the ergonomic score of the working posture according to the human body part angles using a work posture evaluation method, wherein the ergonomic score includes a human body part score and an overall score; Step S5, finding the working posture whose ergonomic score is greater than a preset value, providing an optimization plan for the working posture, wherein the optimization plan is used to prevent musculoskeletal diseases and provide a pain relief plan for the human body area corresponding to the musculoskeletal disease.

2. The method for evaluating human working posture based on deep learning according to claim 1, characterized in that: The reliability index can be optionally set to a confidence index and a shooting angle index. The confidence index is used to determine whether the reliability of the key point detection model's evaluation and prediction of the working posture image meets the requirements, and the shooting angle index is used to determine whether the angle of the human body part calculated based on the working posture image meets the requirements.

3. The method for evaluating human working posture based on deep learning according to claim 2, characterized in that: Calculate the average value of all the confidences. When the average value is greater than or equal to the confidence index, the working posture image meets the requirements. When the average value is less than the confidence index, re-upload the working posture image.

4. The method for evaluating human working posture based on deep learning according to claim 2, characterized in that: The key points include the right shoulder and the left shoulder. The angle between the vectors of the right shoulder and the left shoulder and the optical axis of the image is set as the shooting angle. When the shooting angle is greater than or equal to the shooting angle index, the working posture image meets the requirements. When the shooting angle is less than the shooting angle index, the working posture image is uploaded again.

5. The method for evaluating human working posture based on deep learning according to claim 1, characterized in that: The empirical coefficient is set according to the error between the actual angle of the human body part and the angle of the human body part calculated by the key point detection model. The empirical coefficient includes a neck empirical coefficient, an upper arm empirical coefficient, a forearm empirical coefficient, a back empirical coefficient and a leg empirical coefficient. The neck empirical coefficient is multiplied by the neck angle, the upper arm empirical coefficient is multiplied by the upper arm angle, the forearm empirical coefficient is multiplied by the forearm angle, the back empirical coefficient is multiplied by the back angle, and the leg empirical coefficient is multiplied by the leg angle.

6. The method for evaluating human working posture based on deep learning according to claim 1, characterized in that: The key point detection model includes an improved YOLOv8-pose model, and the improved YOLOv8-pose model is specifically implemented as follows: The backbone network (100) introduces shifted convolution to improve the feature extraction block (101), forming a feature extraction block (102) containing shifted convolution. The neck network (110) introduces a star block (111) for feature fusion. The head network (120) introduces a shared detection head (121), and the shared detection head (121) includes a shared convolution block (1211).

7. The method for evaluating human working posture based on deep learning according to claim 1, characterized in that: The optimization plan reduces the ergonomic score, the corresponding human body areas include the neck, back, legs, arms and wrists, and the pain relief plans include shoulder and neck relief plan, waist and back relief plan, leg relief plan, arm relief plan, wrist relief plan and whole body relief plan.

8. A human body working posture assessment system based on deep learning, used to implement the human body working posture assessment method based on deep learning according to any one of claims 1 to 7, characterized in that: It comprises an input module (210), a human body key point detection module (220), an input image judgment module (230), an ergonomic score calculation module (240) and a pain prevention and relief module (250); The input module (210) inputs a working posture image and working parameters to obtain information for evaluating a human working posture; The human body key point detection module (220) is used to obtain the human body key point information of the working posture image; The input image judgment module (230) calculates a reliability index, wherein the reliability index is used to judge whether the working posture image is reliable; The ergonomic score calculation module (240) calculates the angle of human body parts according to the human body key point information, and calculates the ergonomic score of the working posture using a work posture evaluation method according to the angle of human body parts; The pain prevention and relief module (250) determines whether there is a problem with the working posture according to the ergonomic score, and provides a solution for preventing musculoskeletal diseases if there is a problem.

9. The human body working posture assessment system based on deep learning according to claim 8, characterized in that: The input module (210) comprises: An image input unit, used for inputting an image of a working posture; A work parameter input unit, used to input numerical values ​​including workload, and whether the neck, waist, and wrist are twisted during the work action; The empirical coefficient input unit can selectively set the numerical value of the empirical coefficient within a preset range.

10. The human body working posture assessment system based on deep learning according to claim 8, characterized in that: The prevention and pain relief module (250) comprises a work posture optimization unit and a pain relief unit, wherein the work posture optimization unit provides a solution for reducing the ergonomic score, and the pain relief unit provides a pain relief solution for a human body area corresponding to an ergonomic score greater than a preset score.

Images