AI posture analysis system

Abstract

The invention relates to an AI posture analysis system which comprises a data acquisition module, a posture analysis module, a dynamic posture correction simulation module, a personalized scheme customization module and a user interaction module. According to the method, through algorithm design of a multi-dimensional deviation evaluation model and the like, accurate and comprehensive basic data support is provided for the system. In a feature extraction process, the module not only can capture a numerical value of a single index, but also can ensure that extracted features and healthy posture data of different ages and genders form effective contrast through a standardized reference interval and deviation calculation logic. The limitation that a traditional method only pays attention to the surface posture is avoided, and a more specific and traceable analysis basis is provided for subsequent problem diagnosis. The introduced multi-index collaborative deviation index formula breaks through the inherent mode of single-index independent evaluation in traditional diagnosis, the comprehensiveness of posture problem recognition is remarkably improved, and the diagnosis result is made to better fit the real posture health state.

Description

Technical Field

[0001] This application relates to the technical field of body posture analysis, and in particular to an AI body posture analysis system. Background Technology

[0002] Postural analysis is a scientific method that uses observation and assessment of physiological characteristics such as posture, body symmetry, joint alignment, and muscle tension to determine whether an individual has postural abnormalities or potential health risks. This analysis is commonly used in rehabilitation medicine, sports science, orthopedics, and fitness to identify common postural problems such as rounded shoulders, kyphosis, anterior pelvic tilt, X-shaped legs, or O-shaped legs, and to provide a basis for developing personalized correction and training programs. Postural analysis typically combines static observation, dynamic testing, imaging measurements (such as X-rays and 3D scans), and biomechanical assessments to ensure the accuracy and comprehensiveness of the results. With the increasing prevalence of sedentary lifestyles and poor posture in modern society, postural analysis plays an increasingly important role in preventing sports injuries, improving physical function, enhancing athletic performance, and promoting overall health, and has become an essential component of health management.

[0003] However, current technological methods are inefficient, highly subjective, and the analysis results are greatly affected by human factors. While some smart devices can perform simple body data measurements, such as body fat scales that measure weight and body fat percentage, they cannot perform body shape analysis. Some mobile applications, although having photo analysis functions, can only perform facial recognition or simple body proportion estimations, and cannot generate a visual representation of the expected weight loss outcome, let alone provide a scientific weight loss plan. These solutions suffer from inaccurate analysis, lack of visualization, and lack of personalization, making it difficult to meet users' diverse health management needs. Summary of the Invention

[0004] To address the problem that existing technologies are unable to perform posture analysis or have low accuracy, this application provides an AI posture analysis system.

[0005] This application provides an AI posture analysis system, which adopts the following technical solution: an AI posture analysis system, including a data acquisition module, a posture analysis module, a dynamic posture correction simulation module, a personalized solution customization module, and a user interaction module; The posture analysis module is internally equipped with an image preprocessing module, a posture feature extraction module, and a problem diagnosis module; The dynamic posture correction simulation module is internally equipped with an action library construction module and a dynamic simulation engine module.

[0006] In a preferred embodiment, the data acquisition module comprises an image acquisition unit, a basic information input unit, a health goal setting unit, and a data transmission unit. The image acquisition unit supports multi-device access; users can take real-time full-body side-view photos using their mobile phone camera or upload existing body photos from their local album. The system automatically recognizes the photo resolution and prompts for adjustment to a recommended size (e.g., 1080×1920 pixels) to ensure analysis accuracy. The basic information input unit includes basic fields such as age, gender, height, and weight, and also reserves an interface for external devices to synchronize body fat percentage and muscle mass data from a smart body fat scale, or health parameters such as daily activity levels and resting heart rate from a smart bracelet. The health goal setting unit provides both drop-down menu and free input methods; users can select preset goals (e.g., "improve hunchback" or "lose 5kg") or custom goals (e.g., "reduce pelvic tilt angle by 8°"). The data transmission unit uses the HTTPS encryption protocol to compress image data into JPEG format, convert personal information and goal data into JSON format, and upload them uniformly to the cloud server. When local computing power is sufficient, local storage is used for temporary caching.

[0007] In a preferred embodiment, the image preprocessing module is internally provided with a multi-source image alignment unit, a noise suppression unit, a key region enhancement unit, and a coordinate normalization unit.

[0008] In a preferred embodiment, the dynamic curvature compensation formula of the body feature extraction module is: in: K represents the actual curvature of the spine in the sagittal plane (unit: mm). -1 The larger the value, the more abnormal the curvature; This represents the second derivative of the spinal curve at the arc length parameter s; This represents the first derivative of the spinal curve at the arc length parameter s; α represents the age compensation coefficient; δ age This represents the difference between a user's age and the average age of healthy individuals of the same sex, used to correct for physiological changes in spinal curvature caused by aging.

[0009] This formula, by introducing an age compensation term, avoids misjudging the physiological curvature of the elderly as pathological scoliosis, thus improving the clinical accuracy of posture analysis and better meeting practical application needs than the traditional static curvature formula.

[0010] In a preferred embodiment, the formula for the Multi-Indicator Collaborative Bias Index (SCBI) of the problem diagnosis module is: In the formula: w k The weights of the k-th body posture indicators are: thoracic kyphosis angle w1 = 0.4, pelvic tilt angle w2 = 0.3, and spinal curvature w3 = 0.3. v k This represents the actual measured value of the user's k-th individual physical metric; μ k This represents the health mean of the k-th indicator in the user's age-gender group; σ k This represents the health standard deviation of the k-th indicator in the user's age-gender group; γ represents the synergy coefficient, used to quantify the cumulative effect of deviations from multiple indicators.

[0011] In a preferred embodiment, the motion library construction module internally includes a motion classification unit, a parameter annotation unit, a motion verification unit, and an update and maintenance unit. The motion classification unit is subdivided according to common postural problems, covering shoulder and neck issues, lower back and abdomen issues, and spinal issues, with each category containing 8-12 typical corrective movements. The parameter annotation unit quantifies the standard execution process of each movement from multiple dimensions, recording joint range of motion, muscle activation area, single movement duration, and intervals between sets. The motion verification unit, composed of a review team of rehabilitation medicine experts and fitness coaches, performs dual verification of the safety and corrective effectiveness of the movements. The update and maintenance unit supports dynamic expansion, allowing for the import of new motions from authoritative rehabilitation literature to supplement the library, and also adjusting motion priorities or adding detailed annotations based on user feedback.

[0012] In a preferred embodiment, the dynamic simulation engine module internally includes a data interface unit, a simulation calculation unit, and a video generation unit. The data interface unit is responsible for extracting the three-dimensional coordinates of the user's joints output by the posture analysis module and retrieving the parameters of the corresponding corrective movements from the action library. The simulation calculation unit, based on a human biomechanical model, inputs the user's current posture data and movement parameters into the physical simulation engine, calculating the spatial displacement and angular changes of each joint frame by frame during the execution of the movement. The video generation unit converts the simulation calculation results into dynamic images, uses keyframe interpolation technology to generate continuous animation, synchronously overlays data tags, and allows the user to select a viewing angle for a clearer view of the posture change process.

[0013] In a preferred embodiment, the joint angle-time nonlinear adjustment equation of the dynamic simulation engine module is as follows: In the formula: θ(t) represents the joint target angle at time t; θ0 represents the user's initial abnormal angle; θtarget Indicates the target angle of the action; k represents the muscle elasticity coefficient; t represents the execution time of the action; c represents the muscle fatigue coefficient; m represents the user's muscle mass.

[0014] This formula uses the exponent e -kt Simulates the accelerating effect of muscle elasticity on angle changes (rapid angle change initially, then gradually flattens out), and uses a linear decay term. It reflects the limitations of muscle fatigue on movement execution (the rate of angle change decreases over time), thus more realistically simulating the differences in postural changes when different users perform the same movement.

[0015] In a preferred embodiment, the personalized plan customization module comprises an exercise plan generation unit, a diet plan formulation unit, a daily routine suggestion planning unit, and a plan adjustment feedback unit. The exercise plan generation unit has a built-in posture problem-movement mapping library and automatically allocates exercise types, sets, and intensity based on the user's posture analysis results and available daily time. The diet plan formulation unit generates the intake ratios of protein, carbohydrates, and fats based on the user's basal metabolic rate and target calorie deficit, combined with the Chinese Dietary Guidelines, and recommends specific food combinations. The daily routine suggestion planning unit analyzes the user's lifestyle habits related to posture problems and formulates activity interval reminders and sleep posture guidance. The plan adjustment feedback unit records the user's feedback on plan execution and triggers parameter adjustments.

[0016] In a preferred embodiment, the user interaction module comprises an output display unit, an input feedback unit, and a progress tracking unit. The output display unit presents the analysis results through a multi-column interface: details of posture problems on the left, a dynamic correction simulation video in the middle, and personalized solution cards on the right. The input feedback unit provides text input boxes and sliders, allowing users to input adjustment requests such as "want to increase low-intensity movements," or adjust the target progress using the sliders. The progress tracking unit supports users uploading photos of their posture at different stages; the system automatically compares and analyzes changes in key indicators, generates a line graph to show the improvement trend, and marks "Current progress: 60% of the target completed."

[0017] In summary, this application includes at least one of the following beneficial technical effects: 1. Through the design of algorithms such as the multi-dimensional deviation assessment model, the system is provided with accurate and comprehensive basic data support. During feature extraction, this module not only captures the numerical value of a single indicator but also ensures that the extracted features are effectively compared with health posture data of different ages and genders through standardized reference intervals and deviation calculation logic. This multi-dimensional extraction method avoids the limitations of traditional methods that only focus on surface posture. For example, it is no longer limited to qualitative judgments such as "whether there is kyphosis," but rather to quantitative results accurate to "thoracic kyphosis angle 35°," providing more specific and traceable analytical basis for subsequent problem diagnosis. This transforms the posture assessment of the entire system from "fuzzy description" to "numerical recording," enhancing the reliability of the analysis.

[0018] 2. The introduced multi-indicator synergistic deviation index formula breaks through the inherent pattern of "independent assessment of a single indicator" in traditional diagnosis, significantly improving the comprehensiveness of postural problem identification. This formula, through a combination of weighted summation and product terms, considers both the degree of deviation of individual indicators and captures the synergistic effect of multiple indicator abnormalities. This design enables the system to more sensitively detect situations where "a single indicator is mildly abnormal but the overall posture is significantly unbalanced." For example, when the independent deviations of the thoracic kyphosis angle and pelvic tilt angle do not reach the moderate threshold, the SCBI may still indicate a moderate risk, thus avoiding missed diagnoses caused by single-indicator assessment and making the diagnostic results more consistent with the true state of postural health.

[0019] 3. The dynamic joint angle adjustment equation in the dynamic simulation engine module provides users with a more personalized simulation of the correction process. This formula incorporates personalized parameters such as muscle elasticity coefficient and fatigue coefficient, integrating the user's actual physiological characteristics, including body fat percentage, muscle mass, and daily activity level, into the simulation calculation. This changes the mechanical mode of traditional simulations that focuses on "uniform movement speed and uniform angle changes." For users with poor muscle endurance, when performing the same corrective movement, the system automatically simulates a realistic process of "rapid initial angle changes followed by a decrease in rate due to muscle fatigue," rather than simply linearly achieving the target angle. This personalized simulation not only allows users to more intuitively see the possible trajectory of their own posture adjustment but also helps rehabilitation instructors adjust the intensity of movements based on the simulation results, making the correction plan more targeted and operable. Attached Figure Description

[0020] Figure 1 This is the overall system block diagram of this application; Figure 2 This is a system block diagram of the posture analysis module of this application; Figure 3 This is a system block diagram of the dynamic posture correction simulation module of this application.

[0021] Figure labeling: 1. Data acquisition module; 2. Posture analysis module; 3. Dynamic posture correction simulation module; 4. Personalized solution customization module; 5. User interaction module; 6. Image preprocessing module; 7. Posture feature extraction module; 8. Problem diagnosis module; 9. Action library construction module; 10. Dynamic simulation engine module. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0023] The following is in conjunction with the appendix Figures 1-3 This application will be described in further detail.

[0024] See Figure 1-3 An AI posture analysis system includes a data acquisition module 1, a posture analysis module 2, a dynamic posture correction simulation module 3, a personalized solution customization module 4, and a user interaction module 5. The posture analysis module 2 is internally equipped with an image preprocessing module 6, a posture feature extraction module 7, and a problem diagnosis module 8; the dynamic posture correction simulation module 3 is internally equipped with an action library construction module 9 and a dynamic simulation engine module 10.

[0025] The data acquisition module 1 consists of an image acquisition unit, a basic information input unit, a health goal setting unit, and a data transmission unit. The image acquisition unit supports multi-device access; users can take real-time full-body side-view photos using their mobile phone camera or upload existing body photos from their local album. The system automatically recognizes the photo resolution and prompts for adjustment to the recommended size to ensure analysis accuracy. The basic information input unit includes basic fields such as age, gender, height, and weight, and also reserves an interface for external devices to synchronize body fat percentage and muscle mass data from a smart body fat scale, or daily activity levels and resting heart rate from a smart bracelet. The health goal setting unit offers both drop-down menu and free input methods, allowing users to choose preset or custom goals. The data transmission unit uses the HTTPS encryption protocol, compressing image data into JPEG format and converting personal information and goal data into JSON format, all of which are then uploaded to the cloud server. When local computing power is sufficient, local storage is used for temporary caching.

[0026] The image preprocessing module 6 internally includes a multi-source image alignment unit, a noise suppression unit, a key region enhancement unit, and a coordinate normalization unit. The multi-source image alignment unit integrates image data simultaneously acquired from different acquisition devices (such as RGB cameras and depth cameras), eliminating viewing angle deviations between devices through feature point matching algorithms (such as SIFT feature extraction and RANSAC registration), ensuring that the RGB image and depth image at the same time point are perfectly aligned at the pixel level. The noise suppression unit addresses potential interference during image acquisition (such as salt-and-pepper noise in low light and motion blur), using adaptive median filtering to process random noise, and combining bilateral filtering to preserve edge details while smoothing non-edge areas. The key region enhancement unit uses human pose detection algorithms to locate core posture-related areas such as the shoulders, neck, waist, and hips, and performs local histogram equalization on the images of these areas to improve the contrast of muscle contours and skeletal nodes. The coordinate normalization unit uniformly scales the aligned and enhanced images to a fixed resolution, establishes a local coordinate system based on the human body's center point, and adjusts the image viewing angle to align the front or side view of the human body with the main axis of the image, providing spatially consistent input data for subsequent posture feature extraction.

[0027] The body posture feature extraction module 7 is based on an improved OpenPose deep learning model, combined with 3D coordinate reconstruction and geometric calculation. First, the system preprocesses the user-uploaded full-body frontal / side view images, including grayscale conversion, edge detection, and image normalization, to enhance the recognition accuracy of human contours and key points. The improved OpenPose model, through a multi-stage convolutional network, adds the detection of 8 key posture nodes, including the thoracic vertebrae, lumbar vertebrae, and sacrum, to the traditional 2D key point detection, outputting the 2D pixel coordinates (ui, vi) of 25 key points.

[0028] Subsequently, the system uses the principle of binocular vision (if the user takes two photos from different angles) or monocular depth estimation to map the 2D coordinates to 3D space, and obtain the three-dimensional coordinates (xi,yi,zi) of each joint point, where i is the joint point number, i=1 corresponds to the left shoulder, i=2 corresponds to the right shoulder, i=3 corresponds to the thoracic vertebral vertex, i=4 corresponds to the midpoint of the lumbar vertebra, i=5 corresponds to the left hip, and i=6 corresponds to the right hip; In the calculation of core indicators: Shoulder and neck angle: Calculate the angle between the shoulder line and the vertical line of the cervical spine using the three-dimensional coordinates of the left shoulder (x1,y1,z1), right shoulder (x2,y2,z2), and the cervical vertebra (x7,y7,z7) to determine the degree of hunchback / kyphosis. Spinal curvature: Three points are selected: the thoracic vertebral vertex (x3,y3,z3), the lumbar vertebral midpoint (x4,y4,z4), and the sacral vertex (x8,y8,z8). The sagittal curve of the spine is fitted by cubic spline interpolation, and the radius of curvature of the curve is calculated to quantify the problem of scoliosis or excessive curvature. Pelvic tilt angle: Based on the coordinates of the left hip (x5,y5,z5), right hip (x6,y6,z6) and sacral vertex (x8,y8,z8), a pelvic plane is constructed, the angle between this plane and the horizontal plane is calculated, and anterior or posterior tilt is determined. The dynamic curvature compensation formula addresses the issue of traditional static curvature calculations neglecting individual spinal physiological differences (such as variations in baseline curvature due to age and gender). The formula is as follows: in: K represents the actual curvature of the spine in the sagittal plane (unit: mm). -1 The larger the value, the more abnormal the curvature; This represents the second derivative of the spinal curve at the arc length parameter s (reflecting the degree of curvature of the curve); This represents the first derivative of the spinal curve at the arc length parameter s (reflecting the degree of inclination of the curve); α represents the age compensation coefficient (0.02≤α≤0.05, obtained through training with a large sample of healthy population data); δ age This represents the difference between the user's age and the average age of healthy individuals of the same sex, used to correct for physiological changes in spinal curvature caused by aging (such as the natural decrease in lumbar curvature in the elderly).

[0029] This formula, by introducing an age compensation term, avoids misjudging the physiological curvature of the elderly as pathological scoliosis, thus improving the clinical accuracy of posture analysis and better meeting practical application needs than the traditional static curvature formula.

[0030] Problem diagnosis module 8 is implemented based on a "standard body posture database" and a "multi-dimensional deviation assessment model." First, the standard body posture database integrates large-scale healthy population data to establish "healthy body posture index reference ranges" for each age-gender group. For example, the normal range for thoracic kyphosis angle in women aged 20-30 is 20°-30°, the normal range for pelvic tilt angle is 5°-15°, and the normal range for sagittal curvature of the spine is 0.01-0.03 mm. -1 .

[0031] Once the user completes the extraction of body posture features, the system first matches the user's age and gender to the corresponding reference intervals and calculates the absolute and relative deviations of each indicator. Subsequently, the system uses a "multi-indicator collaborative evaluation model" to comprehensively determine the type of posture problem: if the absolute deviation of the thoracic kyphosis angle is >5° and the relative deviation is >1.5σ (σ is the standard deviation of the reference interval), it is determined to be "kyphosis"; if the absolute deviation of the pelvic tilt angle is >8° and the relative deviation is >2σ, it is determined to be "anterior pelvic tilt".

[0032] In the severity grading, the system uses a three-level classification standard: Mild: Absolute deviation of a single indicator ≤ 10% of the upper limit of the reference interval (e.g., thoracic kyphosis angle 33°, upper limit of reference 30°, deviation 10%); Moderate: Absolute deviation of a single indicator >10% and ≤20% (e.g., thoracic kyphosis angle 36°, deviation 20%); Severe: Absolute deviation of a single indicator >20% (e.g., thoracic kyphosis angle 39°, deviation 30%); If multiple indicators are abnormal at the same time (such as kyphosis + anterior pelvic tilt), the highest level is taken as the final classification (such as moderate kyphosis and severe anterior pelvic tilt, which is ultimately judged as severe). To address the shortcomings of traditional diagnostic methods that rely on independent assessment of single indicators while neglecting the correlation with postural issues (e.g., kyphosis is often accompanied by anterior pelvic tilt), the system introduces the Multi-Indicator Coordination Bias Index (SCBI), with the following formula: In the formula: w k The weights of the k-th body posture indicators are: thoracic kyphosis angle w1 = 0.4, pelvic tilt angle w2 = 0.3, and spinal curvature w3 = 0.3. v k This represents the actual measured value of the user's k-th body posture index (e.g., thoracic kyphosis angle 35°); μ k This represents the health mean of the k-th indicator in the user's age-gender group (e.g., the mean thoracic kyphosis angle of women aged 20-30 is 25°). σ k This represents the health standard deviation of the k-th indicator in the user's age-gender group (e.g., the standard deviation of the thoracic kyphosis angle for women aged 20-30 is 5°). γ represents the synergy coefficient (0.1≤γ≤0.3, obtained through training with large sample abnormal body posture data), used to quantify the superposition effect of multiple index biases.

[0033] The motion library construction module 9 internally includes motion classification, parameter annotation, motion verification, and update / maintenance units. The motion classification unit is subdivided according to common postural problems, covering shoulder and neck, lower back and abdomen, and spine issues, with each category containing 8-12 typical corrective movements. The parameter annotation unit quantifies the standard execution process of each movement from multiple dimensions, recording joint range of motion, muscle activation area, single-repetition duration, and intervals between sets. The motion verification unit, reviewed by a team of rehabilitation medicine experts and fitness coaches, performs dual verification of the safety and corrective effectiveness of the movements. The update / maintenance unit supports dynamic expansion, allowing for the import of new movements from authoritative rehabilitation literature to supplement the library, and also adjusting movement priorities or adding detailed annotations based on user feedback.

[0034] The dynamic simulation engine module 10 internally includes a data interface unit, a simulation calculation unit, and a video generation unit. The data interface unit is responsible for extracting the three-dimensional coordinates of the user's joints output by the posture analysis module and retrieving the parameters of the corresponding corrective movements from the action library. The simulation calculation unit, based on the human biomechanical model, inputs the user's current posture data and movement parameters into the physical simulation engine, calculating the spatial displacement and angular changes of each joint frame by frame during the execution of the movement. The video generation unit converts the simulation calculation results into dynamic images, uses keyframe interpolation technology to generate continuous animation, synchronously overlays data tags, and allows users to select the viewing angle for a clearer view of the posture change process.

[0035] The data interface unit of the dynamic simulation engine module obtains the three-dimensional coordinates of the user's key joints from the posture analysis module and extracts the current abnormal indicators; at the same time, it retrieves the standard parameters of the selected corrective action from the action library.

[0036] The simulation calculation unit is based on a multi-rigid-body dynamics model of the human body, simplifying the human body into a system of 12 rigid bodies connected by ball joints. The mass, center of mass position, and moment of inertia of each rigid body are calculated from basic information such as the user's height and weight. For the selected corrective action, the system first solves the inverse kinematics: with the target angle of the action as a constraint, it calculates the expected angular velocity and angular acceleration of each joint; then, through forward dynamics simulation, it inputs muscle force, gravity, and inertial force into the Newton-Euler equations, and calculates the spatial displacement and angular changes of the joint points frame by frame.

[0037] The video generation unit converts the joint coordinates and angle values ​​at each time point obtained from simulation calculations into keyframes of the 3D model, and generates continuous animation through cubic spline interpolation; at the same time, it overlays real-time data labels and allows users to select the viewing angle to clearly present the body adjustment process.

[0038] The dynamic simulation engine introduces a joint angle-time nonlinear adjustment equation, the formula of which is: In the formula: θ(t) represents the joint target angle (e.g., hunchback angle, unit: °) at time t; θ0 represents the user's initial abnormal angle (e.g., 35°). θ target Indicates the standard target angle for the action (e.g., 15°). k represents the muscle elasticity coefficient (0.1≤k≤0.3, obtained by fitting user body fat percentage and muscle mass data; the higher the body fat percentage, the smaller k is). t represents the execution time of the action (unit: s); c represents the muscle fatigue coefficient (0.02≤c≤0.05, determined by the user's daily activity level data; the lower the activity level, the larger c is). m represents the user's muscle mass (unit: kg, obtained synchronously via body fat scale).

[0039] This formula uses the exponent e -kt Simulates the accelerating effect of muscle elasticity on angle changes (rapid angle change initially, then gradually flattens out), and uses a linear decay term. It reflects the limitations of muscle fatigue on movement execution (the rate of angle change decreases over time), thus more realistically simulating the differences in postural changes when different users perform the same movement.

[0040] The personalized plan customization module 4 consists of an exercise plan generation unit, a diet plan formulation unit, a daily routine suggestion planning unit, and a plan adjustment feedback unit. The exercise plan generation unit has a built-in posture problem-movement mapping library, automatically allocating exercise types, sets, and intensity based on the user's posture analysis results and daily available time. The diet plan formulation unit generates the intake ratios of protein, carbohydrates, and fats based on the user's basal metabolic rate and target calorie deficit, combined with the Chinese Dietary Guidelines, and recommends specific food combinations. The daily routine suggestion planning unit analyzes the user's lifestyle habits related to posture problems and sets activity interval reminders and sleep posture guidance. The plan adjustment feedback unit records the user's feedback on plan execution and triggers parameter adjustments.

[0041] User interaction module 5 consists of an output display unit, an input feedback unit, and a progress tracking unit. The output display unit presents the analysis results through a split-screen interface: the left side displays details of posture problems, the middle displays a dynamic correction simulation video, and the right side displays personalized solution cards. The input feedback unit provides text input boxes and sliders, allowing users to input adjustment requests such as "want to increase low-intensity movements" or adjust the target progress using the sliders. The progress tracking unit allows users to upload photos of their posture at different stages, and the system automatically compares and analyzes changes in key indicators, generating a line graph to show the improvement trend and marking "Current progress: 60% of the target completed."

[0042] From the above, we can conclude that: In this invention, a multi-dimensional deviation assessment model and other algorithmic designs provide the system with accurate and comprehensive basic data support. During feature extraction, this module not only captures the values ​​of single indicators (such as thoracic kyphosis angle and pelvic tilt angle), but also ensures that the extracted features are effectively compared with healthy body posture data of different ages and genders through standardized reference intervals and deviation calculation logic. This multi-dimensional extraction method avoids the limitations of traditional methods that only focus on surface posture. For example, it is no longer limited to qualitative judgments such as "whether there is kyphosis," but rather to quantitative results accurate to "thoracic kyphosis angle 35°," providing more specific and traceable analytical basis for subsequent problem diagnosis. This transforms the entire system's posture assessment from "fuzzy description" to "numerical recording," enhancing the reliability of the analysis.

[0043] This invention introduces the Multi-Indicator Coordination Bias Index (SCBI) formula, which breaks through the inherent pattern of "independent assessment of a single indicator" in traditional diagnosis, significantly improving the comprehensiveness of postural problem identification. This formula, through a combination of weighted summation and product terms, considers both the degree of deviation of individual indicators (such as abnormal values ​​in the thoracic kyphosis angle) and the synergistic effect of multiple indicator abnormalities (such as the correlation between kyphosis and anterior pelvic tilt). This design enables the system to more sensitively detect situations where "a single indicator is slightly abnormal but the overall posture is significantly unbalanced." For example, when the independent deviations of the thoracic kyphosis angle and the pelvic tilt angle do not reach the moderate threshold, SCBI may still indicate a moderate risk, thus avoiding the problem of missed diagnoses caused by single indicator assessment and making the diagnostic results more consistent with the true state of postural health.

[0044] In this invention, the joint angle dynamic adjustment equation of the dynamic simulation engine module provides users with a more personalized simulation of the correction process. This formula incorporates personalized parameters such as muscle elasticity coefficient and fatigue coefficient, integrating the user's actual physiological characteristics, including body fat percentage, muscle mass, and daily activity level, into the simulation calculation. This changes the mechanical mode of traditional simulations that focuses on "uniform movement speed and uniform angle changes." For example, when a user with poor muscle endurance performs the same correction movement, the system automatically simulates a realistic process of "rapid angle changes initially, followed by a decrease in rate due to muscle fatigue," rather than simply linearly achieving the target angle. This personalized simulation not only allows users to more intuitively see the possible trajectory of their own posture adjustment but also helps rehabilitation instructors adjust the intensity of movements based on the simulation results, making the correction plan more targeted and operable.

[0045] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. An AI posture analysis system, characterized in that: It includes a data acquisition module (1), a posture analysis module (2), a dynamic posture correction simulation module (3), a personalized solution customization module (4), and a user interaction module (5); The posture analysis module (2) is internally equipped with an image preprocessing module (6), a posture feature extraction module (7), and a problem diagnosis module (8); The dynamic posture correction simulation module (3) is internally equipped with an action library construction module (9) and a dynamic simulation engine module (10).

2. The AI ​​posture analysis system according to claim 1, characterized in that: The data acquisition module (1) consists of an image acquisition unit, a basic information input unit, a health target setting unit, and a data transmission unit.

3. The AI ​​posture analysis system according to claim 1, characterized in that: The image preprocessing module (6) is internally equipped with a multi-source image alignment unit, a noise suppression unit, a key region enhancement unit, and a coordinate normalization unit.

4. The AI ​​posture analysis system according to claim 1, characterized in that: The dynamic curvature compensation formula of the body feature extraction module (7) is as follows: in: K represents the actual curvature of the spine in the sagittal plane (unit: mm). -1 The larger the value, the more abnormal the curvature; This represents the second derivative of the spinal curve at the arc length parameter s; This represents the first derivative of the spinal curve at the arc length parameter s; α represents the age compensation coefficient; δ age This represents the difference between a user's age and the average age of healthy individuals of the same sex, used to correct for physiological changes in spinal curvature caused by aging.

5. The AI ​​posture analysis system according to claim 1, characterized in that: The formula for the multi-indicator synergistic deviation index (SCBI) of the problem diagnosis module (8) is: In the formula: w k The weights of the k-th postural indicators are: thoracic kyphosis angle w1 = 0.4, pelvic tilt angle w2 = 0.3, and spinal curvature w3 = 0.3; v k This represents the actual measured value of the user's k-th individual physical metric; μ k This represents the health mean of the k-th indicator in the user's age-gender group; σ k This represents the health standard deviation of the k-th indicator in the user's age-gender group; γ represents the synergy coefficient, used to quantify the cumulative effect of deviations from multiple indicators.

6. The AI ​​posture analysis system according to claim 1, characterized in that: The action library construction module (9) is internally configured with an action classification unit, a parameter annotation unit, an action verification unit, and an update and maintenance unit.

7. The AI ​​posture analysis system according to claim 1, characterized in that: The dynamic simulation engine module (10) is internally equipped with a data docking unit, a simulation calculation unit, and a video generation unit.

8. The AI ​​posture analysis system according to claim 1, characterized in that: The joint angle-time nonlinear adjustment equation formula of the dynamic simulation engine module is as follows: In the formula: θ(t) represents the joint target angle at time t; θ0 represents the user's initial abnormal angle; θ target Indicates the target angle of the action; k represents the muscle elasticity coefficient; t represents the execution time of the action; c represents the muscle fatigue coefficient; m represents the user's muscle mass.

9. The AI ​​posture analysis system according to claim 1, characterized in that: The personalized plan customization module (4) consists of an exercise plan generation unit, a diet plan formulation unit, a rest suggestion planning unit, and a plan adjustment feedback unit.

10. The AI ​​posture analysis system according to claim 1, characterized in that: The user interaction module (5) consists of an output display unit, an input feedback unit, and a progress tracking unit.

Images