An acupoint query and positioning system

Abstract

The application discloses an acupoint query and positioning system, and relates to the field of intelligent devices for traditional Chinese medicine and Mongolian medicine, and comprises an interaction module, a query module, a data acquisition module and an AI positioning module. The interaction module receives a query instruction of a user through voice or touch interaction. The query module queries a positioning method of a corresponding acupoint according to the query instruction and displays the positioning method through the interaction module. The data acquisition module acquires a human body depth image and reconstructs a human body 3D model. The AI positioning module identifies the position of the query acupoint on the human body 3D model through an AI model and displays the human body 3D model and the position of the query acupoint through the interaction module. The application realizes intelligent, accurate and visual acupoint query and positioning, and improves the efficiency, accuracy and ease of use of acupoint positioning for traditional Chinese medicine and Mongolian medicine.

Description

Technical Field

[0001] This invention relates to the field of intelligent equipment technology for traditional Chinese medicine and Mongolian medicine, and more specifically to an acupoint query and positioning system. Background Technology

[0002] The location of acupoints in Traditional Chinese Medicine and Mongolian Medicine is an important foundation for acupuncture, massage, rehabilitation therapy, and home health preservation. Accurate acupoint location directly affects the treatment effect and health preservation safety, and has important clinical and practical value.

[0003] However, traditional Chinese medicine and Mongolian medicine rely heavily on methods such as bone measurement, finger measurement, and surface landmarks for acupoint location. These methods often use textual descriptions, two-dimensional diagrams, or simple rulers to assist in finding and locating acupoints. This approach is highly subjective, prone to errors, and heavily dependent on the operator's experience. It is also susceptible to location deviations due to changes in body proportions, posture, and unclear surface landmarks. Existing digital tools are mostly static images and text, unable to adapt to individual differences in height, body type, and posture, and struggle to achieve real-time, visualized, and accurate location. Furthermore, current technologies are largely two-dimensional, lacking depth information and three-dimensional spatial modeling, making it impossible to dynamically match real-world human morphology. Their interaction methods are limited, location efficiency is low, and visualization is poor, failing to meet the needs of clinical practice, teaching, and family healthcare for rapid, accurate, and intuitive acupoint location.

[0004] Therefore, providing an intelligent and visual automatic acupoint query and location system is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] In view of this, the present invention provides an acupoint query and location system to solve the above problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: This invention discloses an acupoint query and positioning system, characterized in that it includes: an interaction module, a query module, a data acquisition module, and an AI positioning module; The interactive module receives user query commands via voice or touch interaction; The query module queries the location method of the corresponding acupoint according to the query command and displays it through the interaction module; The data acquisition module acquires human depth images and reconstructs a 3D model of the human body. The AI ​​positioning module identifies the location of the acupoints on the 3D human body model using an AI model, and displays the 3D human body model and the location of the acupoints through the interaction module.

[0007] Furthermore, the interaction module includes: a voice recognition unit, a touch display unit, and a voice broadcast unit; The voice recognition unit receives and parses the user's voice query command, converts the voice signal into a text command, and transmits it to the query module; the touch display unit receives the user's touch operation command and displays acupoint location information, the human body 3D model, and acupoint location markings; the voice broadcast unit broadcasts the acupoint location method and acupoint location prompts to the user in voice form.

[0008] Furthermore, the query module includes: an acupoint database unit, an instruction parsing unit, and an information matching unit; The acupoint database unit pre-stores the names of standard acupoints, textual descriptions of acupoint locations, two-dimensional diagrams of acupoints, and information on acupoint efficacy and location methods. The instruction parsing unit parses the query instructions transmitted by the interaction module and extracts keywords of the target acupoint name. The information matching unit matches the pre-stored information of the corresponding acupoint in the acupoint database unit based on the keywords and transmits it to the interaction module for display.

[0009] Furthermore, the acquisition of human depth images and reconstruction of a 3D human model specifically includes: With the human body in a standard standing posture, a depth camera is used to collect depth images of the human body from different perspectives from the front, side, and back, thereby obtaining multi-view depth image data. The multi-view depth image data is preprocessed; Viewpoint stabilization features are extracted and semantic matching is performed on the preprocessed multi-view depth images, and the preprocessed multi-view depth images are aligned based on the feature matching results. 3D model reconstruction based on aligned multi-view depth images.

[0010] Furthermore, the preprocessing specifically includes: The multi-view depth image is denoised to obtain a preliminary denoised multi-view depth image. Color correction is performed on the preliminary denoised multi-view depth image; Histogram analysis is performed on the color-corrected multi-view depth images to obtain the pixel value distribution characteristics of each image, thereby adjusting the pixel values ​​to obtain the multi-view depth images with equal exposure. Distortion correction is performed on the multi-view depth image after exposure equalization, and edge sharpening and detail enhancement are performed on the distortion-corrected multi-view depth image to obtain a preprocessed multi-view depth image.

[0011] Furthermore, the viewpoint stabilization feature extraction and semantic matching specifically include: Viewpoint-stabilized feature points are extracted from the preprocessed multi-view depth image to obtain a preliminary viewpoint-stabilized feature point set. Global semantic features are extracted from the initial stable feature point set to obtain a global semantic enhancement feature set, and a fast feature index tree is constructed based on the global semantic enhancement feature set. Based on the feature fast index tree, view-stable feature points are matched between multi-view depth images to obtain a set of matching pairs; The alignment of the preprocessed multi-view depth image based on feature matching results specifically includes: Based on the set of matching pairs, a multi-view geometric relationship graph is constructed to obtain the view topology; Initial camera pose parameters are generated based on the view topology, and the initial camera pose parameters are iteratively optimized to obtain the target camera pose parameters. Based on the target camera pose parameters, a geometric transformation is performed on the preprocessed multi-view depth image to obtain an aligned multi-view depth image.

[0012] Furthermore, the 3D model reconstruction based on the aligned multi-view depth images specifically includes: The aligned multi-view depth images are voxelized to obtain an initial low-resolution voxel model. The initial low-resolution voxel model is subdivided to obtain a high-resolution voxel model; The high-resolution voxel models corresponding to the depth images from each viewpoint are stitched together into a whole, and the distance from each voxel to the surface of the whole is calculated to obtain the SDF model. The SDF model is adaptively subdivided to obtain a high-precision surface mesh model; The high-precision surface mesh model is smoothed to finally generate a 3D human body model.

[0013] Furthermore, the step of identifying and querying the location of acupoints on the 3D human body model using an AI model specifically includes: The reconstructed 3D human body model is subjected to key feature point detection on the body surface to obtain a feature point set and construct a feature point vector. Determine the body shape parameter vector based on the aforementioned 3D human body model; The encoding of the acupoint to be queried is determined and concatenated with the feature point vector and the body shape parameter vector to obtain a sample feature vector. The sample feature vector is then input into a pre-trained acupoint positioning AI model to obtain the three-dimensional coordinate range of the acupoint to be queried. The intersection area between the three-dimensional coordinate range and the surface of the human body 3D model is used as the location of the acupoint to be queried.

[0014] Furthermore, the detection of key feature points on the body surface specifically includes: The human body 3D model is processed into point cloud, converting the high-precision surface mesh model into a disordered three-dimensional point cloud. The kd-tree algorithm is used to construct a neighborhood index for 3D point clouds. Neighborhood search is performed for each point cloud, and the point cloud normal vector is calculated. Construct a covariance matrix based on the consistency weighting of normal vectors, perform eigenvalue decomposition on this covariance matrix to obtain three eigenvalues, and calculate the point cloud curvature; Background curvature is calculated by selecting a large range of neighboring points, local curvature is calculated by selecting a small range of neighboring points, and relative curvature is calculated. Candidate feature points with curvature abrupt changes are filtered by thresholding. High curvature feature points, low curvature feature points and edge points are distinguished to obtain a set of key feature points on the human body surface. Semantic annotation is performed on the key feature point set to label the head, neck, shoulder, elbow, wrist, hip, knee, and ankle, thus completing the detection of key feature points on the body surface.

[0015] Furthermore, the acupoint positioning AI model includes: an input layer, a hidden layer, and an output layer; The input layer inputs the concatenated sample feature vector to each neuron of the fully connected layer; The hidden layer contains at least three fully connected layers, each of which performs linear transformation and activation on the input features to extract high-dimensional features; The output layer outputs a three-dimensional coordinate range based on the high-dimensional features, including the three-dimensional coordinates and the range radius.

[0016] As can be seen from the above technical solution, compared with the prior art, the present invention provides an acupoint query and positioning system, which has the following beneficial effects: This invention, through depth image acquisition and 3D human body model reconstruction, can realistically restore the surface features of the human body of different body types and postures, eliminating reliance on two-dimensional images and text and human experience, and significantly reducing acupoint location errors. Combined with an AI model, it automatically identifies and labels acupoint locations, achieving visualized, precise, and real-time positioning, improving the efficiency and accuracy of acupoint selection. The system supports both voice and touch interaction modes, making operation simple and quick. The acupoint database provides complete location and efficacy information, catering to clinical diagnosis and treatment, teaching and training, and home health maintenance scenarios. In summary, this invention achieves intelligent, precise, and visualized acupoint query and location, comprehensively improving the efficiency, accuracy, and ease of use of acupoint location in Traditional Chinese Medicine (Mongolian Medicine). Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the overall structure of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] This invention discloses an acupoint query and location system, such as... Figure 1 As shown, it includes: an interaction module, a query module, a data collection module, and an AI positioning module; The interaction module receives user query commands via voice or touch interaction; The query module retrieves the location method of the corresponding acupoint based on the query command and displays it through the interactive module; The data acquisition module acquires depth images of the human body and reconstructs a 3D model of the human body; The AI ​​positioning module uses an AI model to identify the location of acupoints on a 3D human body model, and displays the 3D human body model and the location of the acupoints through an interactive module.

[0021] In one specific embodiment, the interaction module includes: a voice recognition unit, a touch display unit, and a voice broadcast unit; The voice recognition unit receives and parses the user's voice query command, converts the voice signal into a text command, and transmits it to the query module; the touch display unit receives the user's touch operation command and displays acupoint location information, a 3D human body model, and acupoint location labels; the voice broadcast unit broadcasts the acupoint location method and acupoint location prompts to the user in voice form.

[0022] In one specific embodiment, the query module includes: an acupoint database unit, an instruction parsing unit, and an information matching unit; The acupoint database unit pre-stores the names of standard acupoints, textual descriptions of acupoint locations, two-dimensional diagrams of acupoints, and information on acupoint efficacy and location methods. The instruction parsing unit parses the query instructions transmitted by the interactive module and extracts keywords for the target acupoint name. The information matching unit matches the pre-stored information of the corresponding acupoint in the acupoint database unit based on the keywords and transmits it to the interactive module for display.

[0023] In one specific embodiment, acquiring human depth images and reconstructing a 3D human model specifically includes: With the human body in a standard standing posture, a depth camera is used to collect depth images of the human body from different perspectives from the front, side, and back, resulting in multi-view depth image data containing human body contours, surface features, and spatial coordinate information. Preprocessing of multi-view depth image data yields noise-reduced, color-consistent, exposure-balanced, and distortion-free multi-view depth images. Viewpoint stabilization features are extracted and semantic matching is performed on the preprocessed multi-view depth images, and the preprocessed multi-view depth images are aligned based on the feature matching results. 3D model reconstruction based on aligned multi-view depth images.

[0024] In one specific embodiment, the preprocessing includes: Denoising is performed on the multi-view depth image to obtain a preliminary denoised multi-view depth image; A standard color chart is placed in the acquisition scene. The multi-view depth image with preliminary noise reduction is analyzed based on the standard color chart to obtain a global color map. Based on the global color map, color correction is performed on the multi-view depth image with preliminary noise reduction. Histogram analysis is performed on the color-corrected multi-view depth images to obtain the pixel value distribution characteristics of each image, thereby adjusting the pixel values ​​to obtain the multi-view depth images with equal exposure. Distortion correction is performed on the multi-view depth image after exposure equalization, and edge sharpening and detail enhancement are then performed on the distortion-corrected multi-view depth image to obtain the preprocessed multi-view depth image.

[0025] In a specific embodiment, viewpoint stabilization feature extraction and semantic matching specifically include: Viewpoint-stable feature points are extracted from the preprocessed multi-view depth images to obtain a preliminary viewpoint-stable feature point set. Viewpoint-stable feature points refer to the contour inflection points of the human body surface that are stable and unaffected by slight changes in shooting angle and posture. They can serve as a reliable benchmark for multi-view image alignment and matching, ensuring the stability of subsequent feature matching.

[0026] Global semantic features are extracted from the initial stable feature point set to obtain a global semantically enhanced feature set, and a fast feature index tree is constructed based on the global semantically enhanced feature set. Global semantic feature extraction combines the semantic information of human body parts to strengthen the location and spatial association attributes of feature points and improve feature discrimination. The fast feature index tree is a tree structure for efficient feature retrieval, which can quickly locate similar features and significantly shorten the retrieval time for feature matching.

[0027] Based on the feature fast index tree, the viewpoint stable feature points between multi-view depth images are matched to obtain a set of matching pairs. Here, matching refers to associating and pairing viewpoint stable feature points corresponding to the same human body position in different viewpoint images, and filtering out the precise corresponding feature point pairs, so as to provide a reliable feature correspondence for subsequent multi-view image geometric alignment and 3D model reconstruction.

[0028] Alignment is performed on the preprocessed multi-view depth images based on feature matching results, specifically including: A multi-view geometric relationship graph is constructed based on the set of matching pairs to obtain the view topology. The view topology is a graph structure that describes the spatial relationship between images from different viewpoints. It is used to clearly express the relative position, connection relationship and feature correspondence links of each viewpoint image, and provides a stable geometric constraint framework for subsequent camera pose calculation and image alignment.

[0029] A hierarchical sparse bundle adjustment model is established based on the view topology to generate initial camera pose parameters. The initial camera pose parameters are iteratively optimized to minimize the reprojection error, thereby obtaining the target camera pose parameters. Among them, the hierarchical sparse bundle adjustment is a hierarchical optimization model of camera pose and feature point spatial position, which can improve computational efficiency and accuracy. The camera pose parameters represent the position and angle of the depth camera during shooting. Iterative optimization and minimization of reprojection error is achieved by repeatedly adjusting the parameters so that the position of the 3D feature points projected onto the 2D image coincides as much as possible with the actual observation position, thereby obtaining accurate and reliable final camera pose parameters.

[0030] Based on the target camera pose parameters, geometric transformations are performed on the preprocessed multi-view depth images to obtain aligned multi-view depth images.

[0031] In one specific embodiment, 3D model reconstruction based on aligned multi-view depth images specifically includes: Voxelization is performed on the aligned multi-view depth images. Voxelization is a process of dividing three-dimensional space into small cubic units to obtain an initial low-resolution voxel model. The initial low-resolution voxel model is subdivided, for example, by using a bounding box hierarchical tree AABB-Tree / OBB-Tree, to obtain a high-resolution voxel model. The high-resolution voxel models corresponding to the depth images from each viewpoint are stitched together into a whole, and the distance from each voxel to the surface of the whole is calculated to obtain the SDF model. Adaptive subdivision of the SDF model yields a high-precision surface mesh model; The high-precision surface mesh model is smoothed to generate a 3D human body model.

[0032] In one specific embodiment, identifying the location of acupoints on a 3D human body model using an AI model specifically includes: The reconstructed 3D human body model is subjected to key feature point detection on the body surface to obtain a feature point set and construct a feature point vector. Determine body shape parameter vectors based on 3D human models; The unique code of the acupoint to be queried is determined and concatenated with the feature point vector and body shape parameter vector to obtain the sample feature vector. The sample feature vector is then input into the pre-trained acupoint positioning AI model to obtain the three-dimensional coordinate range of the acupoint to be queried. The location of the acupoint is determined by the intersection of the three-dimensional coordinate range and the surface of the human body 3D model, and is marked and displayed on the human body 3D model.

[0033] In a specific embodiment, the detection of key feature points on the body surface includes: The human body 3D model is processed into point cloud, and the high-precision surface mesh model is converted into a disordered three-dimensional point cloud, while preserving geometric features such as body surface contour, bone protrusion, and muscle bulge. The kd-tree algorithm is used to construct a neighborhood index for 3D point clouds. Neighborhood search is performed on each point cloud to calculate the point cloud normal vector, where the normal vector is normalized to become a standard normal vector. A covariance matrix based on normal vector consistency weighting is constructed. Local surface geometric feature representation is optimized through weighted centroid and weighted covariance calculations. Eigenvalue decomposition is then performed on this covariance matrix to obtain three eigenvalues ​​in descending order. And calculate the curvature of the point cloud. The covariance matrix based on the consistency weighting of normal vectors is expressed as: ; ; ; in, For the normal vector weights, Let be the normal vector of this point cloud. For the range of point cloud normal vectors, This is the weighting index parameter, with a value of 1.5; for The weighted centroid of a point cloud within a given range, For the first A range of point clouds; This is the weighted covariance matrix.

[0034] Calculate background curvature by selecting a large range of neighboring points. Calculate the local curvature by selecting a small range of neighboring points. And calculate the relative curvature. Candidate feature points with curvature abrupt changes are selected by thresholding; the normal vector consistency coefficient is calculated for the candidate feature points. , normal vector angle variance Anisotropy of normal vector distribution The stability and directionality of local normal vector changes are evaluated; high curvature feature points, low curvature feature points and edge points are distinguished to obtain the set of key feature points on the human body surface; Semantic annotation is performed on the key feature point set to mark the bony landmarks and surface landmarks related to acupoint location in the head, neck, shoulder, elbow, wrist, hip, knee, and ankle, thus completing the detection of key feature points on the body surface.

[0035] In one specific embodiment, the acupoint positioning AI model includes: an input layer, a hidden layer, and an output layer; The concatenated sample feature vector is input into each neuron of the fully connected layer; The hidden layer contains at least three fully connected layers, each of which performs linear transformation and activation on the input features to extract high-dimensional features related to acupoint location. The output layer outputs a three-dimensional coordinate range based on the high-dimensional features, including the three-dimensional coordinates and the range radius.

[0036] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0037] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An acupoint query and location system, characterized in that, include: Interaction module, query module, data collection module, AI positioning module; The interactive module receives user query commands via voice or touch interaction; The query module queries the location method of the corresponding acupoint according to the query command and displays it through the interaction module; The data acquisition module acquires human depth images and reconstructs a 3D model of the human body. The AI ​​positioning module identifies the location of the acupoints on the 3D human body model using an AI model, and displays the 3D human body model and the location of the acupoints through the interaction module.

2. The acupoint query and positioning system according to claim 1, characterized in that, The interactive module includes: a voice recognition unit, a touch display unit, and a voice broadcast unit; The voice recognition unit receives and parses the user's voice query command, converts the voice signal into a text command, and transmits it to the query module; the touch display unit receives the user's touch operation command and displays acupoint location information, the human body 3D model, and acupoint location markings; the voice broadcast unit broadcasts the acupoint location method and acupoint location prompts to the user in voice form.

3. The acupoint query and positioning system according to claim 1, characterized in that, The query module includes: an acupoint database unit, an instruction parsing unit, and an information matching unit; The acupoint database unit pre-stores the names of standard acupoints, textual descriptions of acupoint locations, two-dimensional diagrams of acupoints, and information on acupoint efficacy and location methods. The instruction parsing unit parses the query instructions transmitted by the interaction module and extracts keywords of the target acupoint name. The information matching unit matches the pre-stored information of the corresponding acupoint in the acupoint database unit based on the keywords and transmits it to the interaction module for display.

4. The acupoint query and positioning system according to claim 1, characterized in that, The acquisition of human depth images and reconstruction of human 3D models specifically includes: With the human body in a standard standing posture, a depth camera is used to collect depth images of the human body from different perspectives from the front, side, and back, thereby obtaining multi-view depth image data. The multi-view depth image data is preprocessed; Viewpoint stabilization features are extracted and semantic matching is performed on the preprocessed multi-view depth images, and the preprocessed multi-view depth images are aligned based on the feature matching results. 3D model reconstruction based on aligned multi-view depth images.

5. The acupoint query and positioning system according to claim 4, characterized in that, The preprocessing specifically includes: The multi-view depth image is denoised to obtain a preliminary denoised multi-view depth image. Color correction is performed on the preliminary denoised multi-view depth image; Histogram analysis is performed on the color-corrected multi-view depth images to obtain the pixel value distribution characteristics of each image, thereby adjusting the pixel values ​​to obtain the multi-view depth images with equal exposure. Distortion correction is performed on the multi-view depth image after exposure equalization, and edge sharpening and detail enhancement are performed on the distortion-corrected multi-view depth image to obtain a preprocessed multi-view depth image.

6. The acupoint query and positioning system according to claim 4, characterized in that, The viewpoint stabilization feature extraction and semantic matching specifically include: Viewpoint-stabilized feature points are extracted from the preprocessed multi-view depth image to obtain a preliminary viewpoint-stabilized feature point set. Global semantic features are extracted from the initial viewpoint stable feature point set to obtain a global semantic enhancement feature set, and a feature fast index tree is constructed based on the global semantic enhancement feature set; Based on the aforementioned feature fast index tree, view-stable feature points are matched among multi-view depth images to obtain a set of matching pairs; The alignment of the preprocessed multi-view depth image based on feature matching results specifically includes: Based on the set of matching pairs, a multi-view geometric relationship graph is constructed to obtain the view topology; Initial camera pose parameters are generated based on the view topology, and the initial camera pose parameters are iteratively optimized to obtain the target camera pose parameters. Based on the target camera pose parameters, a geometric transformation is performed on the preprocessed multi-view depth image to obtain an aligned multi-view depth image.

7. The acupoint query and positioning system according to claim 4, characterized in that, The 3D model reconstruction based on aligned multi-view depth images specifically includes: The aligned multi-view depth images are voxelized to obtain an initial low-resolution voxel model. The initial low-resolution voxel model is subdivided to obtain a high-resolution voxel model; The high-resolution voxel models corresponding to the depth images from each viewpoint are stitched together into a whole, and the distance from each voxel to the surface of the whole is calculated to obtain the SDF model. The SDF model is adaptively subdivided to obtain a high-precision surface mesh model; The high-precision surface mesh model is smoothed to finally generate a 3D human body model.

8. The acupoint query and positioning system according to claim 1, characterized in that, The process of identifying and querying the location of acupoints on the 3D human body model using an AI model specifically includes: The reconstructed 3D human body model is subjected to key feature point detection on the body surface to obtain a feature point set and construct a feature point vector. Determine the body shape parameter vector based on the aforementioned 3D human body model; The encoding of the acupoint to be queried is determined and concatenated with the feature point vector and the body shape parameter vector to obtain a sample feature vector. The sample feature vector is then input into a pre-trained acupoint positioning AI model to obtain the three-dimensional coordinate range of the acupoint to be queried. The intersection area between the three-dimensional coordinate range and the surface of the human body 3D model is used as the location of the acupoint to be queried.

9. The acupoint query and positioning system according to claim 8, characterized in that, The detection of key feature points on the body surface specifically includes: The human body 3D model is processed into point cloud, converting the high-precision surface mesh model into a disordered three-dimensional point cloud. The kd-tree algorithm is used to construct a neighborhood index for 3D point clouds. Neighborhood search is performed for each point cloud, and the point cloud normal vector is calculated. Construct a covariance matrix based on the consistency weighting of normal vectors, perform eigenvalue decomposition on this covariance matrix to obtain three eigenvalues, and calculate the point cloud curvature; Background curvature is calculated by selecting a large range of neighboring points, local curvature is calculated by selecting a small range of neighboring points, and relative curvature is calculated. Candidate feature points with curvature abrupt changes are filtered by thresholding. High curvature feature points, low curvature feature points and edge points are distinguished to obtain a set of key feature points on the human body surface. Semantic annotation is performed on the key feature point set to label the head, neck, shoulder, elbow, wrist, hip, knee, and ankle, thus completing the detection of key feature points on the body surface.

10. The acupoint query and positioning system according to claim 8, characterized in that, The acupoint positioning AI model includes: an input layer, a hidden layer, and an output layer; The input layer inputs the concatenated sample feature vector to each neuron of the fully connected layer; The hidden layer contains at least three fully connected layers, each of which performs linear transformation and activation on the input features to extract high-dimensional features; The output layer outputs a three-dimensional coordinate range based on the high-dimensional features, including the three-dimensional coordinates and the range radius.

Images