A posture estimation and standard action comparison based infant physical fitness evaluation and guidance system and method

Abstract

This invention discloses a preschool physical fitness assessment and guidance system based on posture estimation and comparison with standard movements. It includes a hardware module for acquiring panoramic video streams of children performing target movements within a designated area; a data processing module for posture estimation and preprocessing of the video streams; an assessment and feedback module for quantitative assessment of movement standardization and real-time guidance; and a data storage and management module for recording data from each training session and generating individual physical fitness development curves. The method includes steps one, movement acquisition; step two, posture estimation and preprocessing; step three, standard movement comparison and assessment; and step four, real-time feedback and data storage. This invention uses a regular camera to acquire videos of children's basic movements, utilizes an AI posture estimation algorithm to extract the coordinates of key skeletal points, and compares and analyzes these coordinates with a pre-built standard movement model library. This enables real-time quantitative assessment, personalized feedback, and long-term development tracking of children's basic movements, assisting teachers and parents in scientifically guiding children's exercise.

Description

Technical Field

[0001] This invention relates to the field of early childhood education technology, and in particular to a system and method for assessing and guiding children's physical fitness based on posture estimation and comparison with standard movements. Background Technology

[0002] The preschool stage is a critical period for the development of motor skills and the formation of physical fitness. The standardization of basic movements such as running, jumping, and throwing directly affects motor ability and physical coordination. Currently, in kindergarten physical activities, teachers need to pay attention to multiple children at the same time, making it difficult to observe the details of each child's movements in real time. This makes it impossible to accurately judge whether the movements are standard, let alone provide personalized correction guidance. In addition, traditional assessments rely on subjective observation and lack objective data recording, making it impossible to form a continuous tracking file for physical fitness development, which is not conducive to the joint efforts of home and school in cultivating children's exercise habits.

[0003] While existing computer vision technology has been applied to adult motion analysis, it has not established a dedicated standard movement model for the physical development characteristics of young children, resulting in inaccurate analysis data. Furthermore, it lacks a real-time feedback mechanism adapted to young children and does not integrate long-term data tracking and growth curve generation functions, making it impossible to provide targeted guidance and track physical development. Therefore, this invention proposes a system and method for assessing and guiding children's physical fitness based on posture estimation and standard movement comparison to solve the problems existing in the prior art. Summary of the Invention

[0004] To address the aforementioned problems, the present invention aims to propose a system and method for assessing and guiding preschool physical fitness based on posture estimation and comparison with standard movements. This system and method can collect videos of preschool children's basic movements by deploying ordinary cameras, extract the coordinates of key skeletal points using AI posture estimation algorithms, and compare and analyze them with a pre-built standard movement model library to achieve real-time quantitative assessment, personalized feedback, and long-term development tracking of preschool children's basic movements, thereby assisting teachers and parents in scientifically guiding preschool children's exercise.

[0005] To achieve the objectives of this invention, the following technical solution is provided: a preschool physical fitness assessment and guidance system based on posture estimation and standard movement comparison, comprising a hardware module, a data processing module, an assessment feedback module, and a data storage and management module. The hardware module is used to acquire panoramic video streams of preschoolers completing target movements within a designated area. The data processing module is used to perform posture estimation and preprocessing on the video streams. The assessment feedback module is used for quantitative assessment of movement standardization and real-time guidance. The data storage and management module is used to record the preschooler's ID, movement type, skeletal point sequence, standardization score, and feedback suggestions for each training session, generating an individual physical fitness development curve.

[0006] A further improvement is that the hardware module is a common camera, including a tablet camera, a mobile phone camera, or a webcam, which is deployed 1-3 meters in front of the children's activity area at a height of 0.8-1.2 meters.

[0007] A further improvement is that the data processing module includes a pose estimation unit and a preprocessing unit. The pose estimation unit extracts the coordinates of N key skeletal points of the child's body in real time based on a lightweight neural network algorithm and outputs a temporal sequence of skeletal point coordinates P={p1(t1), p2(t2), ..., p n (t) n )}, where n represents the total number of key skeleton points, p i (t) i () indicates that the i-th key skeleton point is at time t i Two-dimensional coordinates, i = 1, 2, ..., n, t i p represents a timestamp. i The coordinates of the i-th key skeleton point are represented by p. i = (x i y i ), x i Let x be the horizontal coordinate of the skeletal point. i The vertical coordinates of the skeletal points are shown below. The key skeletal points extracted include the top of the head, left and right acromions, left and right elbows, left and right wrists, midpoint of the hip, left and right knees, left and right ankles, and left and right toes, totaling 16 points. The preprocessing unit is used to filter, denoise, and normalize the skeletal point coordinate sequence.

[0008] Further improvements are made in that: the evaluation feedback module includes a model library unit, a comparison analysis unit, and a real-time feedback unit. The model library unit is used to construct a standard movement model library to store standard movement models divided by age group. The comparison analysis unit is used to perform dynamic time-normalization comparison between the preprocessed skeletal point coordinate sequence P and the key skeletal point temporal coordinate sequence M of the standard movement model to calculate the standardization score S. The real-time feedback unit is used to generate feedback information based on the standardization score S and the deviation index.

[0009] Further improvements are made in the following ways: the age groups include 3-4 years old, 4-5 years old, and 5-6 years old; the standard movement model includes the key skeletal point temporal coordinate sequence M corresponding to the target movement and the movement standardization index threshold T, M={m1(τ1), m2(τ2), ..., m n (τ) n )}, where n represents the total number of key skeleton points, m i and τ i These represent the i-th key skeleton point at time τ. iThe coordinates of the bone point and the time of the standard action sequence, i=1,2,...,n, the index threshold T includes the joint angle range, the force application timing difference, and the center of gravity displacement amplitude; The normative score S is calculated by the following formula. S = α·(1-D / D) max )+β·(1-Δθ / θ max )+γ·(1-Δt / t) max ) Where α+β+γ=1, Δθ is the joint angle error, Δt is the timing synchronization difference, and D is the dynamic time warping distance. max For the maximum allowable distance, θ max To determine the maximum permissible joint angle error, t max This represents the maximum permissible timing synchronization difference.

[0010] A further improvement lies in the fact that the construction of the standard action model library includes: S1. Collect videos of target movements demonstrated by professional preschool sports coaches, covering all age groups and movement types; S2. Obtain the three-dimensional coordinates of the skeletal points of the coach's movements through high-precision motion capture equipment, and generate an ideal coordinate sequence through time alignment and smoothing. S3. The threshold value T of the indicator is calibrated by the team of early childhood sports experts to determine the range of joint angles, the difference in force application timing, and the amplitude of center of gravity displacement. S4. Store the calibrated models according to age group and movement type to obtain a standard movement model library.

[0011] Further improvements include: the feedback information includes voice feedback and screen animation feedback. The voice feedback is generated using a children's voice synthesis engine, and the feedback content includes a description of specific movement deviations and encouraging statements. The screen animation feedback is a virtual skeleton model overlaid on the screen, including a real-time movement skeleton animation of the child on the left and a standard movement skeleton animation overlaid on the right. Deviations are highlighted in red and flashed, and the text of the voice feedback information is displayed synchronously at the bottom.

[0012] An assessment and guidance method for a preschool physical fitness assessment and guidance system based on posture estimation and comparison with standard movements, comprising the following steps: Step 1: Motion capture. Guide the child to complete the target action in front of the camera on the hardware module according to the system instructions. The camera records a 2-5 second panoramic video stream. Step 2: Pose estimation and preprocessing. The skeleton point coordinate sequence extracted by the pose estimation unit is filtered, denoised, and normalized by the preprocessing unit to generate a temporal coordinate sequence P. Step 3: Standard movement comparison and evaluation. Based on the children's age group and movement type, retrieve the standard movement model M and the index threshold T from the standard movement model library. Then, use the dynamic time warping algorithm to calculate the trajectory coordination distance D between P and M. At the same time, calculate the joint angle error Δθ and the timing synchronization difference Δt. Finally, calculate the standardization score S through the weighting formula. Step 4: Real-time feedback and data storage. Based on the standardization score S, real-time feedback is provided and displayed through voice feedback and screen animation. Finally, the data is stored in the data storage and management module to update the individual fitness development curve.

[0013] Further improvements are made in the following ways: In step two, the filtering and denoising specifically involves averaging the coordinates of skeletal points over five consecutive frames to eliminate instantaneous jitter, and using interpolation of the coordinates from the previous frame to correct abrupt coordinate changes; In step three, the calculation of the joint angle error Δθ includes the knee, elbow, shoulder, and hip joints, and the weight calculation formula is expressed as follows. S = α·(1-D / D) max )+β·(1-Δθ / θ max )+γ·(1-Δt / t) max ) Among them α=0.4, β=0.4, γ=0.2.

[0014] Further improvements are made in the following ways: In step four, the real-time feedback based on the standardization score S is specifically as follows: when S < 60, the core deviation index is fed back first; when S ≤ 60 < 90, optimization suggestions are fed back; and when S ≥ 90, encouraging statements are fed back.

[0015] The beneficial effects of this invention are as follows: By comparing posture estimation with standard movements, this invention transforms the standardization of children's basic movements into a quantifiable objective basis for assessment, greatly improving the accuracy of assessment. It generates exclusive voice and visual animation feedback for children's movement deviations, realizing personalized real-time guidance, effectively compensating for the problem that teachers have limited energy and cannot take care of many children. At the same time, it can automatically record each training data and generate a physical fitness development curve, intuitively presenting the trend of children's movement progress, providing a scientific basis for teachers to adjust teaching strategies and for parents to cooperate in cultivating exercise habits. The system can be deployed with just a regular camera, without the need for professional motion capture equipment. It is low-cost and easy to operate, and is suitable for daily physical activities in kindergartens, helping to scientifically and systematically track children's motor skills and physical development. Attached Figure Description

[0016] Figure 1 This is a system architecture diagram for evaluating and guiding the present invention.

[0017] Figure 2 This is a flowchart of the evaluation and guidance method of the present invention. Detailed Implementation

[0018] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0019] according to Figure 1 and Figure 2 As shown, this embodiment provides a preschool physical fitness assessment and guidance system based on posture estimation and comparison with standard movements, including a hardware module, a data processing module, an assessment feedback module, and a data storage and management module.

[0020] The hardware module is used to collect panoramic video streams of children completing target actions within a designated area. The hardware module is a regular camera, including a tablet camera, a mobile phone camera, or a webcam, which is deployed 1-3 meters in front of the children's activity area at a height of 0.8-1.2 meters.

[0021] The data processing module is used to perform pose estimation and preprocessing on the video stream, including a pose estimation unit and a preprocessing unit. The pose estimation unit extracts the coordinates of N key skeletal points of the child's body in real time from the video based on a lightweight neural network algorithm (such as MoveNet Thunder, OpenPose-Lite, or BlazePose) and outputs a temporal sequence of skeletal point coordinates P={p1(t1), p2(t2), ..., p n (t) n )}, where n represents the total number of key skeleton points, p i (t) i () indicates that the i-th key skeleton point is at time t i Two-dimensional coordinates, i = 1, 2, ..., n, t i p represents a timestamp. i The coordinates of the i-th key skeleton point are represented by p. i = (x i y i ), x i Let x be the horizontal coordinate of the skeletal point. i The vertical coordinates of the skeletal points are N≥10; the key skeletal points extracted include the top of the head, left and right acromions, left and right elbows, left and right wrists, midpoint of the hip, left and right knees, left and right ankles, and left and right toes, a total of 16 points. The preprocessing unit is used to filter, denoise, and normalize the skeletal point coordinate sequence. The filtering and denoising uses Kalman filtering or moving average filtering, and the normalization process establishes a unified coordinate system with the midpoint of the hip as the origin to eliminate the influence of shooting angle and height differences.

[0022] The assessment and feedback module is used for quantitative assessment and real-time guidance of action standardization, including a model library unit, a comparative analysis unit, and a real-time feedback unit. The model library unit is used to construct a standard movement model library to store standard movement models divided by age groups, including 3-4 years old, 4-5 years old, and 5-6 years old. Each standard movement model includes a sequence of key skeletal points (t) M corresponding to the target movement and a threshold value T for movement standardization, where M = {m1(τ1), m2(τ2), ..., m...} n (τ) n )}, where n represents the total number of key skeleton points, m i and τ i These represent the i-th key skeleton point at time τ. i The coordinates of the bone point and the time of the standard action sequence, i=1,2,...,n, the index threshold T includes the joint angle range, the force application timing difference, and the center of gravity displacement amplitude; The construction of the standard motion model library includes: S1. Collect videos of target movements demonstrated by professional preschool sports coaches, covering all age groups and movement types; S2. Obtain the three-dimensional coordinates of the skeletal points of the coach's movements through high-precision motion capture equipment, and generate an ideal coordinate sequence through time alignment and smoothing. S3. The threshold values ​​T of the indicators are calibrated by the team of early childhood sports experts to determine the range of joint angles, the timing difference of force exertion and the magnitude of center of gravity displacement. For example, the knee joint angle is 130°-140° when taking off in the standing long jump, the timing difference between arm swing and leg force exertion is ≤0.2 seconds, and the trunk rotation angle is ≥60° when throwing. S4. Classify and store the calibrated models according to age group and movement type to obtain a standard movement model library; The comparative analysis unit is used to perform dynamic time normalization comparison between the preprocessed skeletal point coordinate sequence P and the key skeletal point temporal coordinate sequence M of the standard motion model, and calculate the standardization score S, which is calculated by the following formula. S = α·(1-D / D) max )+β·(1-Δθ / θ max )+γ·(1-Δt / t) max ) Where α+β+γ=1, Δθ is the joint angle error, Δt is the timing synchronization difference, and D is the dynamic time warping distance. max For the maximum allowable distance, θ max To determine the maximum permissible joint angle error, t max This represents the maximum permissible timing synchronization difference; The real-time feedback unit is used to generate feedback information based on the normative score S and deviation index. The feedback information includes voice feedback and screen animation feedback. The voice feedback is generated using a child voice synthesis engine. The feedback content includes a description of the specific movement deviation and encouraging statements, such as "Bend your knees a little more!" "Swing your arms!" "This landing was very stable, great job!" The on-screen animation feedback is displayed as an overlay of a virtual skeletal model, including a real-time skeletal animation of the child's movements on the left and a standard skeletal animation overlaid on the right. Deviations (such as joint angles deviating from the threshold area) are highlighted and flashed in red, and the text of the voice feedback information is displayed synchronously at the bottom.

[0023] The data storage and management module is used to record the child's ID, movement type, skeletal point sequence, standardization score, and feedback suggestions for each training session, and to generate an individual physical fitness development curve. The individual physical fitness development curve uses time as the horizontal axis and key indicators such as standardization score S and time synchronization difference as the vertical axis. It supports the generation of trend charts by week, month, and quarter, and marks the notes manually entered by the teacher, such as "good condition today" and "needs to strengthen balance practice".

[0024] An assessment and guidance method for a preschool physical fitness assessment and guidance system based on posture estimation and comparison with standard movements, comprising the following steps: Step 1: Motion capture. Guide the child to complete the target action in front of the camera on the hardware module according to the system instructions. The camera records a 2-5 second panoramic video stream. The target movements include running, jumping, throwing, walking on a balance beam, and climbing. Among them, "jumping" is further divided into standing long jump, single-leg hop, and continuous double-leg hop, and "throwing" is further divided into single-handed shoulder throw and two-handed ball toss and catch.

[0025] Step 2: Pose estimation and preprocessing. The skeleton point coordinate sequence extracted by the pose estimation unit is filtered, denoised, and normalized by the preprocessing unit to generate a temporal coordinate sequence P. The filtering and denoising process involves averaging the coordinates of skeletal points over five consecutive frames to eliminate instantaneous jitter, and using the coordinates of the previous frame for correction of abrupt coordinate shifts such as coordinate offsets greater than 20 pixels.

[0026] Step 3: Standard Movement Comparison and Evaluation. Based on the children's age group and movement type, retrieve the standard movement model M and index threshold T from the standard movement model library. Then, use the dynamic time warping algorithm to calculate the trajectory coordination distance D between P and M, and simultaneously calculate the joint angle error Δθ and the timing synchronization difference Δt. Finally, calculate the standardization score S using the weighting formula. Among them, the joint angle error is the deviation between the real-time joint angle and the model threshold range, and the timing synchronization difference is the deviation between the real-time force application timing and the model timing. The calculation objects of joint angle error Δθ include the knee joint, elbow joint, shoulder joint, and hip joint. For example, the knee joint angle error during the standing long jump take-off is |measured angle - 135°|, and the shoulder joint abduction angle error during throwing is |measured angle - 90°|. The weight calculation formula is expressed by the following formula. S = α·(1-D / D) max )+β·(1-Δθ / θ max )+γ·(1-Δt / t) max ) Among them α=0.4, β=0.4, γ=0.2.

[0027] Step 4: Real-time feedback and data storage. Real-time feedback is provided based on the standardization score S. When S < 60, priority is given to feedback on core deviation indicators, such as "insufficient knee flexion"; when S ≤ 60 < 90, optimization suggestions are given, such as "the arm swing amplitude can be increased"; when S ≥ 90, encouraging statements are given, such as "the movement is very standard, keep it up!" The system displays real-time feedback results through voice feedback and on-screen animation feedback, and finally stores the data in the data storage and management module to update the individual's physical fitness development curve.

[0028] The virtual skeletal model used for screen animation feedback is rendered in a cartoon style. The skeleton for real-time movement of young children is represented by blue lines, the skeleton for standard movement is represented by green lines, and the deviation parts are marked with yellow semi-transparent circles.

[0029] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A preschool physical fitness assessment and guidance system based on posture estimation and comparison with standard movements, characterized in that: It includes a hardware module, a data processing module, an assessment and feedback module, and a data storage and management module. The hardware module is used to collect panoramic video streams of children completing target actions within a designated area. The data processing module is used to perform posture estimation and preprocessing on the video streams. The assessment and feedback module is used for quantitative assessment of action standardization and real-time guidance. The data storage and management module is used to record the child's ID, action type, skeletal point sequence, standardization score, and feedback suggestions for each training session, and to generate an individual physical fitness development curve.

2. The preschool physical fitness assessment and guidance system based on posture estimation and standard movement comparison as described in claim 1, characterized in that: The hardware module is a common camera, including a tablet camera, a mobile phone camera, or a webcam, which is deployed 1-3 meters in front of the children's activity area at a height of 0.8-1.2 meters.

3. The preschool physical fitness assessment and guidance system based on posture estimation and standard movement comparison as described in claim 1, characterized in that: The data processing module includes a pose estimation unit and a preprocessing unit. The pose estimation unit extracts the coordinates of N key skeletal points of the child's body in real time based on a lightweight neural network algorithm and outputs a temporal sequence of skeletal point coordinates P={p1(t1), p2(t2), ..., p n (t) n )}, where n represents the total number of key skeleton points, p i (t) i () indicates that the i-th key skeleton point is at time t i Two-dimensional coordinates, i = 1, 2, ..., n, t i p represents a timestamp. i The coordinates of the i-th key skeleton point are represented by p. i = (x i y i ), x i Let x be the horizontal coordinate of the skeletal point. i The vertical coordinates of the skeletal points are shown below. The key skeletal points extracted include the top of the head, left and right acromions, left and right elbows, left and right wrists, midpoint of the hip, left and right knees, left and right ankles, and left and right toes, totaling 16 points. The preprocessing unit is used to filter, denoise, and normalize the skeletal point coordinate sequence.

4. The preschool physical fitness assessment and guidance system based on posture estimation and standard movement comparison as described in claim 1, characterized in that: The evaluation feedback module includes a model library unit, a comparison analysis unit, and a real-time feedback unit. The model library unit is used to construct a standard movement model library to store standard movement models divided by age group. The comparison analysis unit is used to perform dynamic time-normalization comparison between the preprocessed skeletal point coordinate sequence P and the key skeletal point temporal coordinate sequence M of the standard movement model to calculate the standardization score S. The real-time feedback unit is used to generate feedback information based on the standardization score S and the deviation index.

5. The preschool physical fitness assessment and guidance system based on posture estimation and standard movement comparison according to claim 4, characterized in that: The age groups include 3-4 years old, 4-5 years old, and 5-6 years old. The standard movement model includes the key skeletal point temporal coordinate sequence M corresponding to the target movement and the movement standardization index threshold T, M={m1(τ1), m2(τ2), ..., m n (τ) n )}, where n represents the total number of key skeleton points, m i and τ i These represent the i-th key skeleton point at time τ. i The coordinates of the bone point and the time of the standard action sequence, i=1,2,...,n, the index threshold T includes the joint angle range, the force application timing difference, and the center of gravity displacement amplitude; The normative score S is calculated by the following formula. S=α·(1-D / D max )+β·(1-Δθ / θ max )+γ·(1-Δt / t max ) Where α+β+γ=1, Δθ is the joint angle error, Δt is the timing synchronization difference, and D is the dynamic time warping distance. max For the maximum allowable distance, θ max To determine the maximum permissible joint angle error, t max This represents the maximum permissible timing synchronization difference.

6. The preschool physical fitness assessment and guidance system based on posture estimation and standard movement comparison according to claim 4, characterized in that: The construction of the standard action model library includes: S1. Collect videos of target movements demonstrated by professional preschool sports coaches, covering all age groups and movement types; S2. Obtain the three-dimensional coordinates of the skeletal points of the coach's movements through high-precision motion capture equipment, and generate an ideal coordinate sequence through time alignment and smoothing. S3. The threshold value T of the indicator is calibrated by the team of early childhood sports experts to determine the range of joint angles, the difference in force application timing, and the amplitude of center of gravity displacement. S4. Store the calibrated models according to age group and movement type to obtain a standard movement model library.

7. The preschool physical fitness assessment and guidance system based on posture estimation and standard movement comparison according to claim 4, characterized in that: The feedback information includes voice feedback and screen animation feedback. The voice feedback is generated using a children's voice synthesis engine, and the feedback content includes a description of specific movement deviations and encouraging statements. The screen animation feedback is a virtual skeleton model overlaid on the screen, including a real-time movement skeleton animation of the child on the left and a standard movement skeleton animation overlaid on the right. Deviations are highlighted in red and flashing, and the text of the voice feedback information is displayed synchronously at the bottom.

8. The evaluation and guidance method for the system according to any one of claims 1-7, characterized in that, Includes the following steps: Step 1: Motion capture. Guide the child to complete the target action in front of the camera on the hardware module according to the system instructions. The camera records a 2-5 second panoramic video stream. Step 2: Pose estimation and preprocessing. The skeleton point coordinate sequence extracted by the pose estimation unit is filtered, denoised, and normalized by the preprocessing unit to generate a temporal coordinate sequence P. Step 3: Standard movement comparison and evaluation. Based on the children's age group and movement type, retrieve the standard movement model M and the index threshold T from the standard movement model library. Then, use the dynamic time warping algorithm to calculate the trajectory coordination distance D between P and M. At the same time, calculate the joint angle error Δθ and the timing synchronization difference Δt. Finally, calculate the standardization score S through the weighting formula. Step 4: Real-time feedback and data storage. Based on the standardization score S, real-time feedback is provided and displayed through voice feedback and screen animation. Finally, the data is stored in the data storage and management module to update the individual fitness development curve.

9. The assessment and guidance method according to claim 8, characterized in that: In step two, the filtering and noise reduction specifically involves averaging the coordinates of skeletal points over five consecutive frames to eliminate instantaneous jitter, and correcting abrupt coordinate changes by interpolating the coordinates from the previous frame. In step three, the calculation of the joint angle error Δθ includes the knee, elbow, shoulder, and hip joints, and the weighting formula is expressed as follows: S=α·(1-D / D max )+β·(1-Δθ / θ max )+γ·(1-Δt / t max ) Among them α=0.4, β=0.4, γ=0.

2.

10. The assessment and guidance method according to claim 8, characterized in that: In step four, the real-time feedback based on the normative score S is as follows: when S < 60, the core deviation index is fed back first; when S ≤ 60 < 90, optimization suggestions are fed back; when S ≥ 90, encouraging statements are fed back.

Images