Human body motion posture evaluation method and system

Abstract

The invention discloses a human body motion posture evaluation method and system. Low-cost and high-precision human body posture analysis is realized through common camera equipment. The method comprises the following steps: acquiring a human body motion video through camera equipment; detecting key points of a human body by using a deep learning model, and extracting two-dimensional or three-dimensional coordinates of the human body; calculating joint angle, speed and acceleration based on the key point data; and analyzing attitude stability, motion symmetry and fluency characteristics, and generating a comprehensive evaluation result. The system evaluates the motion quality of the user through comparison between dynamic feature extraction and a standard motion template, and generates personalized improvement suggestions based on specific indexes, including direction adjustment and amplitude range adjustment. The method can be widely applied to scenes such as exercise and fitness, rehabilitation medical treatment and competitive training, and the comprehensiveness, the accuracy and the guiding effect of action evaluation are improved. By means of the method, the user can obtain motion quality feedback in real time, the posture is effectively optimized, and the exercise performance or rehabilitation effect is improved.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to a method and system for evaluating human motion posture. Background Art

[0002] In recent years, body motion posture assessment technology has received extensive attention in the fields of sports training, rehabilitation medicine, behavior analysis and intelligent interaction. By analyzing the posture information of the human body during exercise, users can be effectively guided to optimize movement quality, improve sports performance, prevent sports injuries, and assist in medical rehabilitation training. However, the existing technology still has significant deficiencies in theoretical research and practical applications, which limits its popularization and promotion. The following analyzes the main problems and limitations of the existing technology in combination with the specific implementation methods.

[0003] 1. Deficiencies of posture assessment technology based on traditional motion capture systems: Traditional motion capture systems (such as Vicon, OptiTrack, etc.) attach reflective markers to key parts of the human body and use multiple high-precision infrared cameras to capture the three-dimensional motion trajectory of the markers. This is one of the most mature posture assessment technologies. However, this type of technology has the following problems: Expensive equipment and high cost: The hardware equipment of the motion capture system (such as multiple high-precision cameras, sensors, etc.) is expensive, usually costing hundreds of thousands of RMB, and is only suitable for high-budget laboratory research or commercial applications.

[0004] Complex operation and reliance on professionals: Motion capture requires attaching a large number of markers to the user's body, as well as camera calibration and complex data processing. The operation process is cumbersome and requires the assistance of professional technicians, making it difficult for ordinary users to use it independently.

[0005] Obvious environmental limitations: The capture system has strict requirements on light, venue size and background environment. For example, it needs to operate under stable lighting conditions to avoid reflection interference and occlusion problems, which limits its application in ordinary gyms and home environments.

[0006] Poor portability: Since the device requires a multi-camera array and fixed installation, it cannot be made portable and its applicable scenarios are limited.

[0007] 2. Limitations of video-based key point detection technology: In recent years, posture estimation technology based on deep learning has been widely used due to its low cost and no need for markers. This type of technology directly detects the two-dimensional key points of the human body (such as shoulders, elbows, knees, etc.) through video images and uses algorithms to calculate motion features. However, this type of technology still has the following shortcomings in practical applications: Limitations of two-dimensional coordinates: The detection results only provide the plane coordinates (x, y) of the key points of the human body, lacking depth information (z), and cannot fully reflect the true posture of the human body in three-dimensional space. For example, the user's arm moving forward or backward may appear in the same position in the two-dimensional image, resulting in inaccurate analysis results.

[0008] Insufficient ability to capture dynamic features: Existing key point detection technologies are mostly single-frame detection, focusing on the analysis of static postures while ignoring the temporal dimension of actions. The smoothness and rhythmicity of dynamic actions (such as waving or jumping) cannot be achieved through single-frame analysis, resulting in insufficient description of continuous actions.

[0009] Reduced accuracy in complex environments: The accuracy and robustness of keypoint detection will be significantly reduced in multi-person scenes, occluded environments, or conditions with large changes in light. For example, in a crowded gym or natural environment, the detection algorithm may mistake other human keypoints for target keypoints, or fail to detect due to occlusion.

[0010] 3. The singleness and limitations of motion assessment technology: Existing motion assessment technology usually only focuses on indicators of a specific dimension and cannot achieve comprehensive assessment in multiple dimensions. Specifically, Lack of comprehensive posture feature analysis: Most tools only calculate joint angles or motion ranges, while ignoring other key indicators such as posture stability, movement symmetry and smoothness, which are crucial for a comprehensive assessment of movement quality.

[0011] Lack of combined dynamic and static analysis: Static analysis (such as calculation of joint angles in a single frame) and dynamic analysis (such as continuity of movements) are often separated, resulting in one-sided evaluation results. For example, a user may have a correct posture in a certain frame, but the overall movement is not coherent, and existing technologies have difficulty capturing this problem.

[0012] Insufficient personalized feedback: Existing tools mostly provide simple evaluation scores and lack targeted guidance for specific user problems. For example, when insufficient range of motion is detected, no clear direction for improvement is provided (such as “arms need to be raised 15 degrees”), making it difficult for users to optimize their movements.

[0013] 4. Insufficient technology in rehabilitation training: There is a high demand for patient movement posture assessment in the field of rehabilitation medicine, but current technology still has the following problems: Reliance on subjective experience: Many rehabilitation training movements require doctors to judge the patient's movement quality through observation. This subjective evaluation method lacks objectivity and quantitative data, resulting in inaccurate results.

[0014] Poor real-time performance: Patients cannot get real-time feedback during training, and movement deviations cannot be corrected in time, which may delay the rehabilitation process.

[0015] Lack of targeted improvement suggestions: Rehabilitation training usually requires detailed movement adjustments, but the existing system cannot quantify the gap between the patient and the standard movement and provide specific adjustment suggestions.

[0016] 5. Deficiencies in different application scenarios: The limitations of existing technologies in different scenarios further restrict their promotion: Sports and fitness: Ordinary users cannot get real-time movement feedback when exercising at home or in the gym, and improper movements can easily lead to sports injuries.

[0017] Competitive sports: Professional athletes require high-precision posture assessment to optimize performance, but existing technologies are unable to meet their requirements for accuracy and real-time performance.

[0018] Human-computer interaction: In applications such as AR / VR and intelligent robots, real-time capture and dynamic analysis of human posture is still a technical bottleneck, and complex movements (such as squatting and rotating) are difficult to be accurately identified and evaluated.

[0019] Therefore, we urgently need to design a human motion posture assessment method and system to solve the above problems. Summary of the invention

[0020] The invention provides a method for evaluating human body motion posture, which is used to solve the defects in the prior art.

[0021] On the one hand, the present invention provides a method for evaluating human body motion posture, comprising the following steps: S1: Use a camera to collect a human motion video sequence and obtain image data for each frame; S2: Use deep learning models to detect key points of the human body and obtain the two-dimensional coordinates of each joint of the human body ; in: :Indicates the Key points, is the total number of key points; Indicates frame, is the total number of video frames; :are key points In the The horizontal and vertical coordinates of the frame; S3: Calculate the joint angles of adjacent joint points , the formula is:

[0022] in: :Indicates from the joint point Point to joint point A vector of :Indicates from the joint point Point to joint point A vector of : represents a vector The modulus is calculated as: ; S4: Analyze the key point motion trajectory in multiple frames of data and calculate the speed and acceleration , and their formulas are:

[0023]

[0024] in: : Indicates key points In the The frame speed; : Indicates key points In the The acceleration of the frame; : Indicates the time interval between frames.

[0025] S5: Calculate attitude stability index , Movement Symmetry Index and movement fluency index , and generate comprehensive evaluation results.

[0026] On the other hand, the deep learning model in S2 adopts a neural network architecture based on multi-task learning, which can simultaneously predict the coordinates of two-dimensional key points and depth information , specifically including the following steps: Input data collection and preprocessing: collect motion video data including RGB images and depth images through the camera device, and extract the RGB features and depth information of each frame image; Preprocess RGB images and depth images, including normalization, image scaling, and data augmentation.

[0027] Structural design of multi-task deep learning model: adopting multi-task learning neural network architecture, the model includes feature extraction module and multi-task branch: Feature extraction module: use convolutional neural network to extract features of RGB image and depth map; 2D key point detection branch: based on RGB features, generate 2D heat map of each key point, and obtain 2D coordinates through heat map analysis ; Depth information prediction branch: Based on the depth features, predict the depth information of each key point ; Generation of 3D coordinates: Fusion of 2D coordinates and depth information , generate the three-dimensional coordinates of each key point ; Model training: Deep learning model training through multi-task joint loss function:

[0028] in, represents the loss function for 2D keypoint detection, represents the loss function of depth information prediction, and is the weight parameter; Model reasoning: In the model reasoning stage, the RGB image and depth image of each frame are input, and the 3D coordinates of each key point are output. , and used for posture analysis and evaluation.

[0029] On the other hand, the attitude stability index in S5 The calculation formula is:

[0030] in: :Indicates the The key point is The coordinates of the frame; : Indicates the key point value A stability index used to quantify the degree of motion jitter. The smaller the value, the higher the movement stability; Indicates the total number of frames.

[0031] On the other hand, when the stability index Exceeding the set threshold When the system is running, a prompt message indicating insufficient stability is generated, where: It is the threshold for judging the stability of the action, which is adjusted dynamically according to the action type.

[0032] On the other hand, the action symmetry index in S5 The calculation formula is:

[0033] in: :For key points Distance to the origin; Represents the action symmetry index, The closer the value is to 1, the higher the symmetry of the action; Indicates the total number of key points; :Represents and key points Symmetric keypoint numbering.

[0034] On the other hand, for key point pairs with obvious asymmetry, motion improvement suggestions are generated, including: the adjustment direction of the target joint; the adjustment amplitude range.

[0035] On the other hand, the action fluency index in S5 By analyzing the smoothness of speed changes, the formula is:

[0036] in: Indicator of the smoothness of the action. The closer the value is to 1, the smoother the action; : Indicates the speed variation between consecutive frames.

[0037] On the other hand, the movement fluency index The threshold is adjusted according to different sports types. If it is lower than the set threshold, a prompt message is generated to indicate the frame range with insufficient smoothness.

[0038] The present invention also proposes a human body motion posture assessment system, comprising the following modules: Data acquisition module: used to collect video sequences of human body movements; Posture detection module: used to extract the two-dimensional or three-dimensional coordinates of key points of the human body; Dynamic feature analysis module: used to calculate joint angle, velocity, acceleration and motion characteristics; Posture Assessment Module: used to generate indicators of posture stability, symmetry and fluency; Feedback generation module: used to generate action quality assessment reports and improvement suggestions.

[0039] On the other hand, the posture detection module combines the Transformer-based deep learning model and the timing analysis network (such as LSTM or timing convolutional network) to achieve dynamic key point detection and tracking.

[0040] Beneficial effects: Compared with the prior art, this patent provides a method and system for evaluating human motion posture through ordinary camera equipment, avoiding the reliance of traditional motion capture systems on high equipment costs and complex installation environments. High-precision posture detection and evaluation can be completed with only ordinary RGB cameras or smart phones. This technology lowers the entry threshold for users, making it suitable for scenarios such as home fitness, rehabilitation training, and sports guidance, greatly expanding the scope of application. At the same time, the use of artificial intelligence-based algorithms to automatically analyze video data eliminates the need for professional technicians to intervene, allowing ordinary users and non-professional environments to quickly deploy and use the technology.

[0041] This patent innovatively implements a multi-dimensional comprehensive evaluation of human motion posture through the technical solution in claim 1, including joint angle analysis, posture stability detection, movement symmetry evaluation and fluency analysis. It clarifies the calculation methods of posture stability index, symmetry index and fluency index, overcoming the limitation of single feature evaluation in the prior art. For example, by calculating the stability index , can identify the jitter amplitude of key points and prompt users to keep the action stable; through the symmetry index , detect the deviation of the left and right limb movement amplitudes and provide correction directions; through the fluency index ,capture speed fluctuation points and help users optimize action rhythm.,This comprehensive evaluation method makes the analysis of action quality more comprehensive and scientific, and can cover various application scenarios from static postures to dynamic actions.

[0042] This patent not only provides a quantitative assessment of the quality of the action, but also generates personalized improvement suggestions through precise algorithm analysis, such as providing adjustment direction and amplitude prompts for asymmetry, or providing optimization suggestions for the action rhythm for insufficient fluency. Compared with the defects of traditional methods that only provide scores or simple evaluations, the present invention provides users with specific and executable guidance, which significantly improves the actual use effect. At the same time, the solution has great applicability and can be widely used in sports training (such as fitness guidance, competitive sports optimization), rehabilitation medicine (such as patient posture correction, action training guidance) and intelligent human-computer interaction (such as motion capture and analysis of AR / VR scenes). BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 The present invention provides a flow chart of a method for evaluating human body motion posture. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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.

[0046] The following is combined with Figure 1 With reference to the accompanying drawings and a number of embodiments, specific embodiments of the present invention are described in detail.

[0047] The present invention proposes a human body motion posture assessment system, comprising the following modules: Data acquisition module: used to collect video sequences of human body movements; Posture detection module: used to extract the two-dimensional or three-dimensional coordinates of key points of the human body; the posture detection module combines the Transformer-based deep learning model and the timing analysis network to realize dynamic key point detection and tracking.

[0048] Dynamic feature analysis module: used to calculate joint angle, velocity, acceleration and motion characteristics; Posture Assessment Module: used to generate indicators of posture stability, symmetry and fluency; Feedback generation module: used to generate action quality assessment reports and improvement suggestions.

[0049] The evaluation method includes the following steps: S1: Use a camera to collect a human motion video sequence and obtain image data for each frame; S2: Use deep learning models to detect key points of the human body and obtain the two-dimensional coordinates of each joint of the human body , the deep learning model uses a multi-task network architecture, which simultaneously predicts the coordinates of two-dimensional key points and depth information , to generate three-dimensional coordinates .

[0050] in: :Indicates the Key points, is the total number of key points; Indicates frame, is the total number of video frames; :are key points In the The horizontal and vertical coordinates of the frame; S3: Calculate the joint angles of adjacent joint points , the formula is: , in: :Indicates from the joint point Point to joint point A vector of :Indicates from the joint point Point to joint point A vector of : represents a vector The modulus is calculated as: , S4: Analyze the key point motion trajectory in multiple frames of data and calculate the speed and acceleration , and their formulas are: , , in: : Indicates key points In the The frame speed; : Indicates key points In the The acceleration of the frame; : Indicates the time interval between frames.

[0051] S5: Calculate attitude stability index , Movement Symmetry Index and movement fluency index , and generate comprehensive evaluation results, attitude stability index The calculation formula is:

[0052] in: :Indicates the The key point is The coordinates of the frame; : Indicates the key point value A stability index used to quantify the degree of motion jitter. The smaller the value, the higher the movement stability; Indicates the total number of frames; when the stability indicator Exceeding the set threshold When the system is running, a prompt message indicating insufficient stability is generated, where: It is the threshold for judging the stability of the action, which is dynamically adjusted according to the action type; the action symmetry index The calculation formula is:

[0053] in: :For key points Distance to the origin; Represents the action symmetry index, The closer the value is to 1, the higher the symmetry of the action; Indicates the total number of key points; :Represents and key points Symmetric keypoint numbering; For key point pairs with obvious asymmetry, motion improvement suggestions are generated, including: adjustment direction of target joints; adjustment range; movement fluency index By analyzing the smoothness of speed changes, the formula is:

[0054] in: Indicator of the smoothness of the action. The closer the value is to 1, the smoother the action; : Indicates the speed variation between consecutive frames; Movement Fluency Index The threshold is adjusted according to different sports types. If it is lower than the set threshold, a prompt message is generated to indicate the frame range with insufficient smoothness.

[0055] The following is a specific description in conjunction with several embodiments: Embodiment 1: Motion posture evaluation method based on ordinary camera; 1. Data acquisition and preprocessing: Use an ordinary RGB camera (such as the camera on a laptop or a smartphone) to collect human motion videos. Set the resolution to 1920×1080, the frame rate to 30 frames per second, the total acquisition time to 15 seconds, and the data storage format to MP4.

[0056] Decompose the captured video into a series of static image frames, generate 30 frames per second, and mark the time stamp ,in is the total number of frames. The image frame data is preprocessed by OpenCV. The main steps include: image denoising; image size scaling to 640×480; conversion to floating point format to adapt to deep learning model input.

[0057] 2. Posture key point detection calls the OpenPose model based on deep learning to detect 17 key points of the human body frame by frame, including the head, shoulders, elbows, knees, etc., and then outputs the two-dimensional coordinates of each key point:

[0058] in Indicates Key points, Indicates frame.

[0059] The output keypoint coordinates are stored in matrix form:

[0060] 3. Posture feature extraction: Joint angle calculation: Select key parts such as elbows and knees, and calculate the joint angles of adjacent key points. Taking the knee joint as an example, the joint angle formula is:

[0061] in: is the knee joint vector; is the vector norm, representing the length of the knee joint.

[0062] Output the angle sequence of the knee joint in each frame And record the fluctuation range.

[0063] Speed ​​and acceleration calculation: Extract the motion trajectory of the knee joint and calculate its speed and acceleration:

[0064]

[0065] in Second.

[0066] 4. Data analysis and report generation: Joint angle fluctuation analysis: If the angle change of the knee joint does not meet expectations (such as lower than 90 degrees or higher than 180 degrees), the system prompts the user "The knee flexion and extension angle is insufficient, please increase the range of motion."

[0067] Speed ​​analysis: If the speed fluctuates too much (for example, the speed difference between consecutive frames exceeds 20%), the system prompts "The movement speed is uneven, please adjust the rhythm."

[0068] Output report: Action completion: for example, "Action completion: 80%"; Improvement suggestions: For example, "The elbow movement is not standard, please increase the flexion and extension angle."

[0069] Example 2: Posture stability analysis based on three-dimensional coordinates; 1. Data collection and processing: Use an RGB-D camera (such as Intel RealSense) to collect human motion videos and their depth information (Z-axis coordinates). The depth information resolution is 640×480 and the frame rate is 30 frames per second. The depth data is fused with the two-dimensional coordinates to generate the three-dimensional coordinates of the key points of the human body:

[0070] in Indicates The depth of the key point.

[0071] 2. Posture stability analysis: Calculate the stability index of each key point:

[0072] in: Represents the displacement of key points in consecutive frames; if , prompting "The motion jitter of this key point is too large". Output a stability chart and mark the parts with larger jitter.

[0073] Embodiment 3: Movement symmetry assessment in fitness coaching; 1. Data collection: The user uses a mobile phone to shoot a squat exercise video and sets the frame rate to 25 frames per second.

[0074] The key target points are the left and right knees, left and right shoulders, and left and right hips.

[0075] 2. Movement symmetry calculation: The symmetry formula is:

[0076] like , prompting the user that the left and right movements are asymmetrical; Specific suggestions for improvement include: "The right knee has insufficient range of motion, please increase it by 15 degrees."

[0077] Example 4: Fluency assessment in rehabilitation training; 1. Application scenario: In knee joint rehabilitation training, detect the smoothness of the patient's lower limb flexion and extension movements.

[0078] 2. Movement Fluency Calculation: Calculate the fluency index based on the knee joint velocity sequence:

[0079] like , prompting the patient that "the rhythm of the movement is uneven, please adjust the speed."

[0080] Output report: Movement score: such as "Movement fluency score: 75 / 100"; Improvement suggestion: "Slow down the knee speed in the 3rd to 5th seconds."

[0081] Embodiment 4: Three-dimensional key point detection method based on multi-task learning; 1. Data collection and preprocessing: Use an RGB-D camera (such as Intel RealSense or Azure Kinect) to collect video data of human motion. The frame rate of the acquisition is 30 frames per second, and each frame image contains: RGB image: The resolution is 1920×1080, providing color and texture information of the human appearance; Depth image: The resolution is 640×480, and the grayscale value of each pixel represents the depth distance from the camera.

[0082] Data preprocessing includes: resizing: scaling the RGB image to 640×480 and aligning it with the depth image; normalization: normalizing the RGB image pixel values ​​to the range [0,1] by channel; data enhancement: performing random rotation, flipping, and cropping to improve the generalization ability of the model.

[0083] 2. Model architecture design: feature extraction module: Use a multi-task learning neural network to simultaneously predict the 2D coordinates and depth values ​​of key points. Specifically, the following modules use a convolutional neural network (CNN) based on ResNet as the backbone network, taking RGB images and depth maps as input. Extract multi-scale features, fuse RGB image features with depth map features at the feature layer, and generate a shared feature map.

[0084] 2D key point detection branch: Based on the shared feature map, a stacked convolutional layer is used to generate a heat map of 17 key points (each key point corresponds to a channel). Each heat map represents the probability distribution of the key point on the image plane, and the maximum position of the heat map is taken as the 2D coordinate .

[0085] Depth prediction branch: 3D coordinate generation module: Input the shared feature map into the deep regression network and output the depth value of each key point The depth information is predicted using a global regression method, and the depth value is normalized and mapped back to the actual depth range.

[0086] The two-dimensional key point coordinates The corresponding depth value Combined, the three-dimensional coordinates of 17 key points of the human body in each frame are generated:

[0087] 3. Model training: Dataset preparation: Select a public human posture dataset containing depth information, annotate the data, and provide the two-dimensional coordinates and depth values ​​of the key points in each frame.

[0088] Loss function design: Use multi-task joint loss function to optimize the model:

[0089] in: Mean squared error loss for 2D keypoint detection; ; and : are the weights of the two-dimensional detection and depth prediction losses, respectively, with an initial value of 1.0.

[0090] Training details: The Adam optimizer is used with an initial learning rate of 0.001; during training, RGB images and depth maps are input for each batch, and the losses of 2D heat maps and depth predictions are optimized simultaneously.

[0091] Reasoning phase: Data input: 3D coordinate generation: In the inference phase, the real-time acquired RGB images and depth maps are input into the trained model.

[0092] The model outputs the two-dimensional coordinates of each key point in each frame image and depth value Fusion of 2D coordinates and depth values ​​to obtain 3D coordinates of key points . Output the three-dimensional coordinates as structured data for subsequent posture analysis and motion evaluation.

[0093] 4. Application scenarios: Sports guidance: In fitness scenarios, the user's three-dimensional posture is detected in real time to evaluate whether the action conforms to the standard template. For example, in a squat action, the height and angle of the knee joint are detected to see if they meet the specifications, and adjustment suggestions are provided.

[0094] Rehabilitation training: In rehabilitation scenarios, three-dimensional coordinate data is used to monitor the patient's range of motion and degree of recovery, assisting doctors in developing personalized rehabilitation plans.

[0095] Competitive sports optimization: In competitive sports, athletes are provided with detailed 3D posture data to analyze movement efficiency and technical defects.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for evaluating human body motion posture, characterized in that: The following steps are involved: S1: Use a camera to collect a human motion video sequence and obtain image data for each frame; S2: Use deep learning models to detect key points of the human body and obtain the two-dimensional coordinates of each joint of the human body ; in: :Indicates the Key points, is the total number of key points; Indicates frame, is the total number of video frames; :are the key points In the The horizontal and vertical coordinates of the frame; S3: Calculate the joint angles of adjacent joint points , the formula is: in: :Indicates from the joint point Point to joint point A vector of :Indicates from the joint point Point to joint point A vector of : represents a vector The modulus is calculated as: ; S4: Analyze the key point motion trajectory in multiple frames of data and calculate the speed and acceleration , and their formulas are: in: : Indicates key points In the The frame speed; : Indicates key points In the The acceleration of the frame; : Indicates the time interval between frames. S5: Calculate attitude stability index , Movement Symmetry Index and movement fluency index , and generate comprehensive evaluation results.

2. The method for evaluating human body motion posture according to claim 1, characterized in that: The deep learning model in S2 adopts a neural network architecture based on multi-task learning, which can simultaneously predict the coordinates of two-dimensional key points and depth information , specifically including the following steps: Input data collection and preprocessing: collect motion video data including RGB images and depth images through the camera device, and extract the RGB features and depth information of each frame image; Preprocess RGB images and depth images, including normalization, image scaling, and data augmentation. Structural design of multi-task deep learning model: adopting multi-task learning neural network architecture, the model includes feature extraction module and multi-task branch: Feature extraction module: use convolutional neural network to extract features of RGB image and depth map; 2D key point detection branch: based on RGB features, generate 2D heat map of each key point, and obtain 2D coordinates through heat map analysis ; Depth information prediction branch: Based on the depth features, predict the depth information of each key point ; Generation of 3D coordinates: Fusion of 2D coordinates and depth information , generate the three-dimensional coordinates of each key point ; Model training: Deep learning model training through multi-task joint loss function: in, represents the loss function for 2D keypoint detection, represents the loss function of depth information prediction, and is the weight parameter; Model reasoning: In the model reasoning stage, the RGB image and depth image of each frame are input, and the 3D coordinates of each key point are output. , and used for posture analysis and evaluation.

3. The method for evaluating human body motion posture according to claim 1, characterized in that: The S5 attitude stability index The calculation formula is: in: :Indicates the The key point is The coordinates of the frame; : Indicates the key point value A stability index used to quantify the degree of motion jitter. The smaller the value, the higher the movement stability; Indicates the total number of frames.

4. The method for evaluating human body motion posture according to claim 3, characterized in that: When the stability index Exceeding the set threshold When the system is running, a prompt message indicating insufficient stability is generated, where: It is the threshold for judging the stability of the action, which is adjusted dynamically according to the action type.

5. The method for evaluating human body motion posture according to claim 1, characterized in that: Symmetry index of the actions in S5 The calculation formula is: in: :For key points Distance to the origin; Represents the action symmetry index, The closer the value is to 1, the higher the symmetry of the action; Indicates the total number of key points; :Represents and key points Symmetric keypoint numbering.

6. The method for evaluating human body motion posture according to claim 5, characterized in that: For key point pairs with obvious asymmetry, motion improvement suggestions are generated, including: the adjustment direction of the target joint; the adjustment range.

7. The method for evaluating human body motion posture according to claim 1, characterized in that: The movement fluency index in S5 By analyzing the smoothness of speed changes, the formula is: in: Indicator of the smoothness of the action. The closer the value is to 1, the smoother the action; : Indicates the speed variation between consecutive frames.

8. The method for evaluating human body motion posture according to claim 7, characterized in that: Movement Fluency Index The threshold is adjusted according to different sports types. If it is lower than the set threshold, a prompt message is generated to indicate the frame range with insufficient smoothness.

9. A human body motion posture assessment system, based on the human body motion posture assessment method according to any one of claims 1 to 8, characterized in that: Includes the following modules: Data acquisition module: used to collect video sequences of human body movements; Posture detection module: used to extract the two-dimensional or three-dimensional coordinates of key points of the human body; Dynamic feature analysis module: used to calculate joint angle, velocity, acceleration and motion characteristics; Posture Assessment Module: used to generate indicators of posture stability, symmetry and fluency; Feedback generation module: used to generate action quality assessment reports and improvement suggestions.

10. A human body motion posture assessment system according to claim 9, characterized in that: The posture detection module combines the Transformer-based deep learning model and the timing analysis network to achieve dynamic key point detection and tracking.

Images