Posture assessment device and assessment method

Abstract

The invention discloses a posture evaluation device and method. The device comprises an integrated base (1), a guide rail column (2), a sliding support (3), a binocular vision module (4), a gait detection pad (5), a calibration plate (6) and a display module (7). The device is compact in overall structure, convenient to operate and high in evaluation precision, covers static / dynamic / plantar multi-dimensional posture evaluation, provides a standardized and automatic posture evaluation method, realizes rapid recognition, grading judgment and personalized suggestion generation of posture abnormity, and is suitable for scenes such as hospital rehabilitation departments, physical examination centers, fitness organizations and schools; the evaluation report contains personalized improvement suggestions, and can directly guide the user to carry out posture correction.

Description

Technical Field

[0001] This invention relates to the fields of biomedical engineering and sports health technology, specifically to a posture assessment device integrating static posture detection, dynamic posture detection, and plantar pressure detection, and an assessment method based on this device. Background Technology

[0002] Postural abnormalities (such as uneven shoulders, scoliosis, anterior pelvic tilt, and flat feet) not only affect physical appearance but can also lead to a range of health problems, including muscle strain, joint degeneration, and motor dysfunction. Currently, postural assessments primarily rely on manual observation and simple measuring tools (such as protractors and tape measures), which suffer from high subjectivity, low measurement accuracy, and a limited range of assessment dimensions. Some existing assessment equipment focuses only on static posture imaging or single-site detection, lacking a combination of static and dynamic assessments and failing to fully integrate plantar pressure data with musculoskeletal data, resulting in incomplete and inaccurate assessment results. Furthermore, existing assessment methods suffer from insufficient standard database sample size and limited applicable populations, making it difficult to meet the assessment needs of different age groups and body types. Their low level of automation also hinders the rapid generation of intuitive assessment reports and personalized improvement suggestions.

[0003] To address the aforementioned issues, this invention proposes an integrated posture assessment device and method that combines binocular visual detection and plantar pressure detection. Through multi-dimensional data acquisition, deep learning model analysis, and standardized database comparison, it achieves accurate, comprehensive, and automated posture assessment. Summary of the Invention

[0004] I. Purpose of the Invention The purpose of this invention is to overcome the shortcomings of the prior art and provide a body posture assessment device that is compact, easy to operate, highly accurate, and covers multiple dimensions including static, dynamic, and plantar aspects. At the same time, it provides a standardized and automated body posture assessment method to achieve rapid identification, grading, and personalized suggestion generation of body posture abnormalities.

[0005] II. Technical Solution (a) Posture assessment device

[0006] The posture assessment device of the present invention, as described in claims 1-2 and 6, includes an integrated base (1), a guide rail column (2), a sliding bracket (3), a binocular vision module (4), a gait detection pad (5), a calibration plate (6), and a display module (7). The specific structure and connection relationship of each component are as follows:

[0007] Integrated base (1): The base shell (101) is made of ABS engineering plastic injection molding, with dimensions of 60cm×40cm×15cm. It has a hollow interior design and anti-slip rubber pads on the bottom to ensure stable placement of the device. The computing system (102) is fixed to the inner wall of the bottom of the base shell (101). The computing unit module (1021) uses an NVIDIA Jetson Xavier NX embedded processor to support real-time operation of deep learning models. The power supply module (1022) is a 220V to 12V DC power adapter with an output power of 50W and overvoltage and overcurrent protection functions. The communication module (1023) uses a Wi-Fi 6 module (model AX200) and a Bluetooth 5.0 module (model CSR8675) to support wireless data transmission to mobile terminals. It is also equipped with an RJ45 wired communication interface to ensure data transmission stability.

[0008] Guide rail post (2): Made of 304 stainless steel, 30mm in diameter and 1.8m in height (can be customized within the range of 1.5-2.2m), with laser-engraved length scale on the surface and a scale accuracy of 1mm. The guide rail post (2) is vertically fixed to the top end face of the integrated base (1) via a flange, with a verticality deviation ≤0.1mm / m, ensuring smooth sliding of the sliding bracket (3).

[0009] Sliding bracket (3): Made of aluminum alloy, with a U-shaped structure, and a guide sleeve matching the guide column (2) on the inner side. The inner wall of the guide sleeve is provided with a wear-resistant nylon bushing. The locking knob (301) on the sliding bracket (3) adopts a butterfly bolt design, which locks the guide sleeve by pressing the thread. After locking, the axial displacement deviation is ≤0.5mm. The height of the binocular vision module (4) can be adjusted according to the height of the person being assessed (adjustment range 1.2-2.0m).

[0010] Binocular vision module (4): It adopts two IMX307 infrared cameras with a resolution of 1920×1080, a frame rate of 60fps, a lens focal length of 8mm, and a field of view of 60°. The optical axes of the two cameras are parallel and the distance between them is 12cm (adjustable within the range of 8-15cm). They are fixed together by a bracket and can be detachably installed on a sliding bracket (3). The lens faces the center area of ​​the gait detection pad (5) and the shooting distance is 2.5m (the horizontal distance between the center of the preset detection area and the camera).

[0011] Gait detection pad (5): Made of flexible polyurethane material, with dimensions of 100cm×60cm×0.5cm and a non-slip textured surface. 16 weighing sensors (501) (model Loadstar LC201, range 0-200kg, accuracy 0.1kg) are evenly distributed inside, located at the four corners, the midpoint of the four sides, and the center position; the thin-film pressure sensor (502) is made of PVDF material, with a resolution of 1mm×1mm, a pressure detection range of 0-1MPa, and completely covers the effective detection area (90cm×50cm) of the gait detection pad (5), with a sensor data sampling frequency of 150Hz.

[0012] Calibration plate (6): It is made of acrylic rigid plate with a width × height × thickness of 80cm × 200cm × 1cm. The surface is printed with black equally spaced grid scales with a minimum grid unit scale of 2mm (which can be selected within the range of 1-5mm). The edge of the scale is marked with x-axis and y-axis coordinate scales (range 0-80cm, 0-200cm). The calibration plate (6) is vertically fixed to the side of the gait detection pad (5) away from the integrated base (1) by the bracket. The horizontal distance between the calibration plate (6) and the center of the gait detection pad (5) is 2.5m, and the verticality deviation is ≤0.1mm / m.

[0013] Display module (7): The display body (701) adopts a 15.6-inch touch LCD screen with a resolution of 1920×1080 and a brightness of 300cd / m²; the mounting bracket (703) is made of aluminum alloy and is detachably connected to the guide rail column (2) by M8 studs. The display body (701) can rotate ±30° along the horizontal axis of the mounting bracket (702) for easy observation by the person being evaluated and the operator.

[0014] Electrical connection: The computing unit module (1021) is connected to the power module (1022), communication module (1023), binocular vision module (4), gait detection pad (5) and display module (7) respectively through shielded wires (model RVVP 4×0.75). The wires are routed through the wire groove inside the integrated base (1) and are provided with protective sleeves to ensure electrical safety and signal stability.

[0015] (II) Posture Assessment Methods The posture assessment method of the present invention, as described in claims 3-5, includes the following specific implementation steps:

[0016] The calibration plate (6) is fixed at the preset calibration position (2.5m horizontal distance from the binocular vision module (4), and the vertical height is level with the center of the camera). After the device is started, the computing unit module (1021) automatically controls the binocular vision module (4) to capture images of the calibration plate (6), and captures 20 frames continuously. After removing the blurry frames, 15 valid images are retained. The Zhang Zhengyou binocular vision calibration algorithm is adopted, and the intrinsic parameters (focal length, principal point coordinates, distortion coefficient) and extrinsic parameters (rotation matrix, translation vector) of the camera are calculated by the calibrateCamera function in the OpenCV open source library to establish a benchmark parameter model. The vertical distance in the first evaluation data is verified by the laser rangefinder (error ≤ 0.1cm). The detection error values ​​∆x and ∆y are determined by comparing the actual scale of the grid with the algorithm detection value to ensure that they are not greater than 0.1mm.

[0017] Step 2: Collection and Processing of Static Human Posture Data The person being evaluated stands in the area marked with a standing sign in the center of the gait detection pad (5), and completes the standard static evaluation movements (arms hanging naturally at the outside of the legs, feet shoulder-width apart) according to the voice prompts of the display module (7): standing facing forward, standing to the left, standing to the right, standing behind, slowly bending over to 90° and holding still, each movement held for 4 seconds, and the binocular vision module (4) continuously captures 8-10 frames of images at a frame rate of 45fps. The computation unit module (1021) performs Gaussian filtering for noise reduction, Canny edge detection, and SIFT feature extraction on the images, and calculates the static depth point cloud image of the human body by combining the benchmark parameters in the first evaluation data and using the triangulation method. The point cloud image is input into a static three-dimensional reconstruction model based on the PointNet network architecture (input dimension is point cloud coordinates (x,y,z) + color information, output is a three-dimensional human body mesh model), which has been trained with 60,000 frames of labeled images (including static images of the human body of different ages and body postures) and has a recognition accuracy of over 97%. Based on a 3D mesh model, key points of the human skeleton (a total of 25, including the top of the head, tip of the nose, chin, waist, back, acromion, elbow joint, wrist joint, finger joint, anterior superior iliac spine, knee joint, ankle joint, etc.) are extracted. Various parameters in the second evaluation data are calculated. For example, the difference in acromion height between the two sides is calculated by the difference in the z-coordinates of the two points, and the verticality of the spine is determined by the angle between the line connecting the key points of the spinous processes and the vertical line.

[0018] Step 3: Foot pressure data acquisition and processing The subject maintains a forward standing posture. The weighing sensor (501) and the thin-film pressure sensor (502) of the gait detection pad (5) simultaneously collect plantar pressure data for 8 seconds at a sampling frequency of 150Hz. The computation unit module (1021) filters the raw pressure data (moving average filtering, window size 5) and performs noise reduction. It then generates a plantar pressure distribution heatmap (color range from blue (low pressure) to red (high pressure)) using an interpolation algorithm. Based on the pressure distribution data, the peak position of plantar pressure (usually the heel, first metatarsal, and fifth metatarsal regions), the pressure center trajectory (normal trajectory is "S" shaped), the pressure value of the arch region (average pressure in the central area of ​​the arch), and the ratio of heel to forefoot pressure are calculated to establish third assessment data. For example, in patients with flat feet, the arch region pressure value accounts for more than 30%, and the ratio of heel to forefoot pressure is less than 1.2.

[0019] Step 4: Human dynamic posture data acquisition and processing The person being evaluated completes the standard dynamic evaluation movements according to the animation guidance of the display module (7): slowly raise head (45°, lasting 2 seconds), lower head (45°, lasting 2 seconds), turn upper body left and right (90° each, lasting 2 seconds in each direction), squat (bend knees 90°, lasting 3 seconds), extend arms forward to expand chest (repeated 3 times), extend arms to the sides and raise and lower (raise angle 0-90°, lower angle 0-90°, repeated 3 times), and lower arms diagonally backward (60°, repeated 3 times). The binocular vision module (4) captures dynamic images at a frame rate of 50fps. The computing unit module (1021) extracts the trajectory of human motion feature points using optical flow and generates a dynamic depth point cloud sequence in combination with the first evaluation data. The point cloud sequence is input into a dynamic motion analysis model based on the LSTM+CNN fusion architecture (CNN extracts spatial features, LSTM extracts time series features). This model is trained on 70,000 frames of labeled dynamic images and has a recognition accuracy of 95.5%. Based on the dynamic 3D model data output by the model, the range of motion of each joint (such as the range of motion of the neck flexion and extension calculated by the angle change of the key points of the cervical spine) and the deviation value of the motion trajectory (Euclidean distance from the standard trajectory) are calculated to establish the fourth evaluation data.

[0020] Step 5: Establish a database of posture assessment standards Body posture assessment data were collected from over 1000 healthy individuals, covering children (6-12 years old, 200 cases), adolescents (13-18 years old, 200 cases), adults (19-59 years old, 400 cases), and the elderly (60 years and older, 200 cases), with a male-to-female ratio of 1:1 and a ratio of 1:2:1 for different body types (underweight, standard, overweight). Statistical analysis (using SPSS software) was performed on the collected second, third, and fourth assessment data, calculating the mean ± 2 standard deviations of each parameter as the normal range threshold. For example, the normal range for cervical spine physiological curvature in adults is 30-40° (mean 35°, standard deviation 2.5°). The normal range thresholds, standard 3D models, and standard motion trajectories were stored in the storage unit (capacity over 500GB) of the computing unit module (1021) to establish a body posture assessment standard database. The database supports updates and iterations via the communication module (1023), with each update adding 150 new samples, covering different population characteristics (such as special occupational groups and rehabilitation patients).

[0021] Step 6: Generation of Posture Assessment Results The calculation unit module (1021) matches and compares the second, third, and fourth assessment data of the person to be assessed with the corresponding parameters in the standard database, and calculates the degree of deviation: deviation value ≤10% is "normal", 10% < deviation value ≤30% is "mildly abnormal", 30% ≤50% is "moderately abnormal", and deviation value >50% is "severely abnormal". The assessment level of all parameters is integrated to generate an assessment report, which includes: ① basic information (name, age, gender, test time); ② assessment conclusion (overall body posture level, main abnormal parts); ③ abnormal parts marking (abnormal areas are marked in red on the three-dimensional model); ④ improvement suggestions (home stretching training program is provided for mild abnormalities, rehabilitation training plan is provided for moderate abnormalities, and medical examination is recommended for severe abnormalities). The assessment report is displayed on the touch screen of the display module (7) and can also be sent to the mobile phone or computer of the person being assessed through the communication module (1023).

[0022] The order of steps is as follows: When using the body posture assessment device of the present invention for the first time, the body posture assessment standard database must be updated according to the requirements of step 5. The body posture assessment standard database in step 5 can be updated and iterated according to the actual application scenario. When updating, the number of new samples should not be less than 100, and the new samples should cover different age and gender characteristics. The execution order of steps 2, 3 and 4 can be changed according to actual needs. For example, the foot pressure test can be performed first (step 3), then the static body posture test (step 2), and finally the dynamic body posture test (step 4), without affecting the accuracy of the assessment results.

[0023] III. Beneficial Effects Comprehensive assessment: It integrates three dimensions of detection: static posture, dynamic posture, and plantar pressure, covering key parts of the whole body (head, neck, spine, limbs, joints, and feet). Compared with existing single-dimensional devices, the assessment is more comprehensive and can detect potential postural abnormalities (such as abnormal joint movement, body shape deviation, and pelvic tilt caused by flat feet).

[0024] High precision: It adopts a binocular vision module + high-precision sensor, combined with a deep learning model (recognition accuracy ≥95%), with a detection error ≤0.1mm and joint range of motion measurement accuracy ≤1°, which is far superior to manual assessment and traditional equipment.

[0025] High degree of automation: Full voice / animation guidance throughout the process, no professional personnel required, data collection, processing, analysis and report generation are all automated, detection time ≤15 minutes, high efficiency.

[0026] High degree of standardization: A standard database is established based on 1000+ healthy population samples, supporting multi-population adaptation and database updates, ensuring objective and consistent evaluation results and avoiding the subjectivity of manual evaluation.

[0027] Highly practical: The device has a compact structure and is easy to move, making it suitable for scenarios such as hospital rehabilitation departments, physical examination centers, fitness institutions, and schools; the assessment report includes personalized improvement suggestions, which can directly guide users in posture correction. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below: Figure 1: Schematic diagram of the overall three-dimensional structure of the body posture assessment device of the present invention; Figure 2: Electrical block diagram connection of the body posture assessment device of the present invention; Figure 3: Schematic diagram of sensor layout of gait detection pad (5); Figure 4: Schematic diagram of the cross-sectional structure of the sliding bracket (3) and the binocular vision module; Figure 5: Flowchart of the body posture assessment method; (Note: The attached figure is a schematic diagram. In actual implementation, it can be drawn according to the specific dimensions and structure. The component numbers in the attached figure are consistent with the claims: 1 - integrated base, 101 - base shell, 102 - computing processing system, 1021 - computing unit module, 1022 - power module, 1023 - communication module, 2 - guide rail column, 3 - sliding bracket, 301 - locking knob, 4 - binocular vision module, 5 - gait detection pad, 501 - weighing sensor, 502 - thin film pressure sensor, 6 - calibration plate, 601 - xy coordinate scale line, 7 - display module, 701 - display body, 702 - fixed bracket, 703 - fixed base). Detailed Implementation

[0029] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0030] 1. Place the integrated base (1) on a level surface and adjust the bottom anti-slip rubber pad to ensure that the device is stable and does not shake. 2. Fix the guide rail column (2) to the top end face of the integrated base (1) through the flange, and use a level to calibrate the verticality of the guide rail column (2), with a deviation ≤0.1mm / m; 3. Place the sliding bracket (3) onto the guide rail column (2), install the locking knob (301), adjust the sliding bracket (3) to a height of 1.5m and lock it, and check whether it is loose after locking; 4. Fix the binocular vision module (4) on the sliding bracket (3), adjust the camera lens to face the center area of ​​the gait detection pad (5), and calibrate the horizontal axis of the camera using a level. 5. Lay the gait detection pad (5) 2.5m in front of the integrated base (1), ensuring that the detection pad is flat and wrinkle-free, and connect the sensor wire to the terminal block of the integrated base (1); 6. Fix the calibration plate (6) to the side of the gait detection pad (5) away from the integrated base (1) using the bracket, and adjust the verticality and height of the calibration plate (6) to ensure that it is flush with the center of the camera; 7. Connect the mounting bracket (702) of the display module (7) to the guide rail column (2) through studs, adjust the height to 1.2~1.5m, install the display body (701), and connect the power cord and data cable; 8. Connect the shielded wires of each component, tidy up the wire channels, ensure that the wires are not tangled or compressed, connect the power module (1022), and test whether each component is working properly.

[0031] II. Equipment Commissioning 1. Start the computing system (102), enter the system settings interface through the touch screen of the display module (7), and configure the Wi-Fi or Bluetooth settings parameters of the communication module (1023); 2. Perform binocular vision module calibration: Click the “calibrate” button on the display UI interface. The system will automatically take a picture of the calibration board (6) and calculate the reference parameters, display the ∆x and ∆y values. If the difference exceeds 0.1mm, readjust the camera position and calibrate again. 3. Load or update the body posture assessment standard database; 4. Test the gait detection pad (5): Place standard weights (5kg, 10kg, 20kg) in the center area of ​​the detection pad to mark the standing position, and check the deviation between the sensor reading and the weight of the weight to ensure that it is ≤0.1kg; 5. Test the deep learning model: Import the preset standard static / dynamic images, check whether the 3D model output by the model and the evaluation parameters are accurate. If the recognition accuracy is lower than 95%, reload the trained model weights until the recognition accuracy is greater than or equal to 95%.

[0032] III. Actual Usage Process 1. The person being evaluated understands the key points of the evaluation process; 2. Enter the basic information of the person being evaluated (name, age, gender) through the display module (7), and click "Start Evaluation"; 3. Follow the voice prompts to complete each static assessment action; the system will automatically collect and process the data. 4. Maintain a standing posture and complete the plantar pressure data collection; 5. Follow the voice prompts or animation guidance to complete the dynamic evaluation actions; the system will automatically collect and process the data. 6. The system automatically compares the body posture assessment standard database, generates an assessment report, and displays it on the display module (7); 7. The person being evaluated can view the report details, save the report to their mobile phone by scanning the QR code, or print the report; 8. After the assessment is completed, click "End". Depending on the selection of the person being assessed, the system will automatically save / clear the assessment data for the next test.

[0033] IV. Equipment Maintenance 1. The device should be placed in a dry and ventilated environment, avoiding direct sunlight on the calibration plate (6) and binocular vision module (4). 2. The gait detection pad (5) should be protected from scratches by sharp objects. The surface should be cleaned regularly to remove stains. The sensor should be replaced promptly if it is damaged. 3. Regularly (every 3 months) calibrate the binocular vision module (4) to ensure detection accuracy; 4. The standard database should be updated regularly according to the usage scenario, and new samples must be labeled and verified by professionals; 5. If the device malfunctions, the power should be cut off first, and a professional technician should be contacted for repair. Do not disassemble the internal parts yourself.

[0034] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can be modified and varied, with adjustments or combinations of steps. Any modifications, equivalent substitutions, adjustments and combinations of steps, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A body assessment device, characterized by, The device comprises an integrated base (1), a guide rail column (2), a sliding support (3), a binocular vision module (4), a gait detection mat (5), a calibration plate (6) and a display module (7). The guide rail column (2) is vertically fixed to the top end face of the integrated base (1), the sliding support (3) can slide and lock along the guide rail column (2), the binocular vision module (4) is detachably fixed to the sliding support (3), the gait detection mat (5) is laid in the preset detection area in front of the integrated base (1), the calibration plate (6) is vertically arranged on the side of the gait detection mat (5) away from the integrated base (1), and the display module (7) is detachably connected with the integrated base (1).

2. A body assessment device as claimed in claim 1, wherein, The integrated base (1) comprises a base shell (101) and an operation processing system (102). The base shell (101) is a hollow rectangular shell, and the operation processing system (102) is fixed to the bottom inner wall of the base shell (101) by bolts. The operation processing system (102) comprises an operation unit module (1021), a power module (1022) and a communication module (1023). The power module (1022) provides stable power supply for the operation unit module (1021), the communication module (1023) and other electric components of the device. The communication module (1023) supports wired communication and wireless communication. The wireless communication mode includes at least one of Wi-Fi and Bluetooth. The gait detection mat (5) is a rectangular flexible mat. Weight sensors (501) and thin film pressure sensors (502) are uniformly arranged in the gait detection mat (5). The thin film pressure sensors (502) completely cover the effective detection area of the gait detection mat (5). At least four weight sensors (501) are symmetrically arranged in the central area of the gait detection mat (5). The binocular vision module (4) comprises two infrared cameras with the same resolution. The optical axes of the two infrared cameras are parallel, and the distance between them is 8-15 cm. The lenses of the two infrared cameras are directed to the calibration plate (6) corresponding to the central area of the gait detection mat (5). The display module (7) comprises a display body (701), a fixed support (702) and a fixed seat (703). The fixed seat (703) is detachably connected with the guide rail column (2) through a stud. The display body (701) can rotate horizontally by ±30° along the fixed support (702). The calibration plate (6) is a rigid flat plate. An equidistant grid scale (601) is printed on the surface of the calibration plate (6). The minimum unit scale of the grid is 1-5 mm, and the coordinate scale is marked on the edge of the scale. The operation unit module (1021) is electrically connected with the power module (1022), the communication module (1023), the binocular vision module (4), the gait detection mat (5) and the display module (7) through shielding wires to realize data transmission and instruction control.

3. A body condition evaluation method based on the body condition evaluation device according to claim 1, characterized by, The device comprises the following steps: Step 1: Binocular vision module calibration, fix the calibration plate (6) at the preset calibration position, control the binocular vision module (4) to shoot the image of the calibration plate (6), calculate the reference parameter model of the binocular vision camera relative to the calibration plate (6) through the binocular vision calibration algorithm, and establish the first evaluation data; The first evaluation data includes the vertical distance from the binocular vision camera to the calibration plate (6), the actual length x of the smallest grid unit scale of the calibration plate (6) in the x axis direction, the actual length y in the y axis direction, the detection value X of the binocular vision algorithm, the detection value Y, and the detection error value Δx = |X-x|, Δy = |Y-y|, wherein Δx and Δy are both not greater than 0.1mm; Step 2: Human static posture data acquisition and processing, control the binocular vision module (4) to obtain the static image of the human static evaluation standard action, the static evaluation standard action includes standing front, standing left, standing right, standing back and slowly bending to 90° and keeping still, each action keeps for 3-5 seconds, and the binocular vision module (4) continuously shoots 5-10 frames of images for each action; Combine the first evaluation data, process the static image through the image noise reduction and feature extraction image processing algorithm to obtain the human static depth point cloud image; input the static depth point cloud image into the pre-trained neural network deep learning model, output the static three-dimensional model data of the human external contour, construct the human static skeleton model and static posture evaluation model based on the static three-dimensional model data, and establish the second evaluation data; The second evaluation data includes: Front evaluation parameters: the coincidence degree of the body midline and the vertical line, specifically whether the head top midpoint, nose tip, sternum midpoint, umbilical region midpoint, pubic symphysis midpoint, double knee joint midpoint and double ankle joint midpoint are collinear; The height difference of the bilateral shoulder peaks, the symmetry degree of the clavicle, the horizontal degree difference of the bilateral anterior superior iliac spines, the varus / valgus angle of the double knees, and the weight difference of the double feet; Wherein the bilateral shoulder peak height difference ≤5mm is normal, the clavicle symmetry degree is evaluated by the length difference and angle difference of the bilateral clavicle, the bilateral anterior superior iliac spine horizontal degree difference ≤3mm is normal, the varus / valgus angle of the double knees ≤5° is normal, and the weight difference of the double feet ≤10% is normal; Side evaluation parameters: the physiological arc compliance of the side body curve, specifically whether the earlobe, shoulder peak, greater trochanter of femur, knee joint front and lateral malleolus tip are collinear; The physiological curvature value of cervical vertebra, the thoracic vertebra kyphosis angle, the lumbar vertebra lordosis angle, and the knee joint hyperextension angle; Wherein the cervical vertebra physiological curvature value is 30-40°, the thoracic vertebra kyphosis angle is 20-40°, the lumbar vertebra lordosis angle is 30-50°, and the knee joint hyperextension angle ≤5° is normal; Back evaluation parameters: spinal column verticality, specifically whether the occipital protuberance, thoracic spinous process, lumbar spinous process, and midpoint of sacrum are collinear; bilateral scapula symmetry, distance difference between medial border of scapula and spinal column, bilateral trapezius muscle elevation height difference, bilateral gluteal crease horizontal level difference, and bilateral calf muscle thickness difference; wherein the distance difference between medial border of scapula and spinal column is ≤5mm, and the bilateral gluteal crease horizontal level difference is ≤3mm, which are normal; back elevation height difference when bending 90°, with a difference of ≤10mm, which is normal, and is used to assist in judging the direction and degree of spinal column scoliosis; Step 3: foot pressure data acquisition and processing, controlling the gait detection mat (5) to collect foot pressure data of both feet when the human body stands, with a sampling frequency of 100-200Hz and a collection time of 5-10 seconds; processing the collected pressure data through a pressure distribution algorithm to generate a foot pressure distribution heat map and a foot pressure distribution model, and establishing third evaluation data; the third evaluation data includes foot pressure peak position, pressure center trajectory, arch area pressure value, and heel-to-forefoot pressure ratio, and is used to detect foot shape abnormalities such as flat feet, foot valgus and varus, high and low feet, and splayfoot; wherein the arch area pressure value accounts for 20-30% of the total foot pressure, and the heel-to-forefoot pressure ratio is 1.2-1.8, which are normal; Step 4: human body dynamic posture data acquisition and processing, controlling the binocular vision module (4) to obtain dynamic images of dynamic evaluation standard actions, the dynamic evaluation standard actions including slowly lifting the head (head lifting angle 0-45°), lowering the head (head lowering angle 0-45°), turning the upper body (left and right turning each 0-90°), squatting (knee flexion angle 0-90°), stretching the arms forward to expand the chest, stretching the arms sideways to raise and lower (raising angle 0-90°, lowering angle 0-90°), and lowering the arms obliquely to the back (back lowering angle 0-60°), each action being repeated 3-5 times, and the binocular vision module (4) having a shooting frame rate of 30-60fps; combining the first evaluation data, extracting human motion feature points in the dynamic images through a dynamic image processing algorithm to obtain a human dynamic depth point cloud sequence; inputting the dynamic depth point cloud sequence into a pre-trained dynamic neural network model to output dynamic three-dimensional model data of the human body external contour, constructing a human dynamic skeletal model and a dynamic posture evaluation model based on the dynamic three-dimensional model data, and establishing fourth evaluation data; the fourth evaluation data includes joint activity of the neck, shoulder, waist, hip, knee, and ankle, and trajectory deviation values when each joint moves, wherein the neck forward flexion activity is 0-45°, the backward extension activity is 0-80°, and the left and right rotation activities are 0-80° respectively, the shoulder abduction activity is 0-180°, the forward flexion activity is 0-180°, the waist forward flexion activity is 0-90°, the left and right rotation activities are 0-45°, the knee flexion and extension activity is 0-145°, the ankle dorsiflexion activity is 0-20°, and the plantarflexion activity is 0-50°. Step 5: Establish a posture evaluation standard database, repeat steps 2 to 4, collect posture evaluation data of healthy people of different ages, different genders and different body types, the sample size is not less than 1000 cases, and the collected second evaluation data, third evaluation data and fourth evaluation data are statistically analyzed to determine the normal range threshold of each evaluation parameter, and a posture evaluation standard database is established. Step 6: Posture evaluation result generation, compare the second evaluation data, third evaluation data and fourth evaluation data of the person to be evaluated with the corresponding parameters in the posture evaluation standard database, analyze the model matching degree, calculate the deviation degree of each evaluation parameter from the normal range threshold, and divide the evaluation level according to the deviation degree (normal, mild abnormality, moderate abnormality, severe abnormality), integrate the level results of all evaluation parameters, generate a posture evaluation report containing evaluation conclusion, abnormal part labeling and improvement suggestion, and display it through the display module (7).

4. A body shape assessment method as claimed in claim 3, wherein, When using the posture evaluation device of claim 1 for the first time, the posture evaluation standard database must be updated according to the requirements of step 5. The posture evaluation standard database in step 5 can be updated and iterated according to actual application scenarios. When updating, the number of new samples added is not less than 100, and the new samples need to cover different age and gender characteristics of the population. The execution order of steps 2, 3 and 4 can be interchanged.

5. A body shape assessment method as claimed in claim 3, wherein, The neural network deep learning model includes a static three-dimensional reconstruction model and a dynamic motion analysis model. The static three-dimensional reconstruction model is optimized based on the PointNet network architecture, and the dynamic motion analysis model is constructed based on the fusion of the LSTM network architecture and the CNN network architecture. Both models have been trained with not less than 50,000 labeled images, and the model recognition accuracy is not less than 95%.

6. A body assessment device as claimed in claim 2, wherein, The sliding support (3) is provided with a locking knob (301), and the sliding support (3) is locked and slides with the guide rail column (2) through the locking knob (301). The axial displacement deviation of the locked sliding support (3) is ≤0.5mm; the height of the guide rail column (2) is 1.5-2.2m, and the surface is provided with a length scale with a scale accuracy of 1mm.

Images