Three-dimensional human body surface data generation method and system based on point cloud

Abstract

The invention relates to the technical field of human body three-dimensional point cloud, in particular to a three-dimensional human body surface data generation method and system based on point cloud. The method comprises the following steps: guiding a subject to stand according to a standard posture, and scanning the whole body of the subject through a three-dimensional human body scanning device to obtain a three-dimensional human body point cloud model containing human body surface information; according to a preset key point detection rule, carrying out geometric feature-based human body key point position identification on the three-dimensional human body point cloud model so as to form an initial key point position set; and performing anatomical key point marking on the three-dimensional human body point cloud model based on the initial key point position set so as to obtain an initial key point marking model. According to the method, 47 accurate anatomical key point positions are introduced for identification, so that the problem that a template matching method is insufficient in individual feature capture is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of human body three-dimensional point cloud, and in particular to a method and system for generating three-dimensional human body surface data based on point cloud. Background Art

[0002] 3D human surface data is digital data generated by 3D scanning technology, point cloud processing, shape modeling and other methods based on the geometric and topological information of the external structure of the human body. It mainly includes: geometric shape, the precise 3D shape of the human surface (such as the shape of the head, limbs, and torso); topological structure, the mesh representation of the human surface (such as triangular mesh or quadrilateral mesh); human feature points, markings of key anatomical positions (such as joints, nose tips, eye corners, shoulders, etc.); surface properties, which may include surface texture (such as skin color) or physical properties (such as dynamic changes in muscles and soft tissues). Generally, 3D scanning equipment (such as laser scanning, structured light scanning or depth camera) is used to collect human surface data and generate point cloud models. The point cloud-based 3D human surface data generation method is a technical method that collects point cloud data on the human surface, processes, optimizes and models it, and finally generates a high-precision 3D human surface model. Its core lies in the use of point cloud data processing technology, combined with anatomical key point marking and shape modeling, to achieve complete modeling and optimization of human surface data.

[0003] However, traditional human 3D reconstruction technology usually uses a template matching-based method, that is, using a predefined universal human template for deformation fitting. Although this method is easy to operate, it is difficult to accurately capture the anatomical features of an individual, especially in the joint area and complex surface parts. Most traditional key point marking methods rely on manual intervention or two-dimensional image recognition algorithms, which are difficult to ensure stability and consistency under complex postures or occlusion conditions. Summary of the invention

[0004] Based on this, it is necessary for the present invention to provide a method and system for generating three-dimensional human body surface data based on point cloud to solve at least one of the above technical problems.

[0005] To achieve the above purpose, a method for generating three-dimensional human body surface data based on point cloud comprises the following steps: Step S1: guiding the subject to stand in a standard posture, and performing a full body scan of the subject using a three-dimensional human body scanning device to obtain a three-dimensional human body point cloud model containing human body surface information; Step S2: According to the preset key point detection rules, the positions of the human body key points based on the geometric features are identified on the three-dimensional human body point cloud model, so as to form an initial key point position set; based on the initial key point position set, anatomical key points are marked on the three-dimensional human body point cloud model, so as to obtain an initial key point marked model; Step S3: Optimize the correction of key point positions in the initial key point marking model based on the correction rules of human anatomical proportion, symmetry, and neighborhood structure, and perform spatial position matching with the preset reference anatomical structure to correct the key point positions, thereby obtaining the key point correction position data; Step S4: Construct a human bone topology structure based on the key point correction position data, map the key points into the human shape parameter space through a statistical shape model to form human shape space parameters; generate a human mesh model according to the human shape space parameters, combine the simulation of physical-based muscle and fat layers, and apply surface optimization processing to obtain three-dimensional human body surface data.

[0006] The present invention also provides a three-dimensional human body surface data generation system based on point cloud for performing the above-mentioned three-dimensional human body surface data generation method based on point cloud. The three-dimensional human body surface data generation system based on point cloud includes: A three-dimensional scanning module, configured to guide the subject to stand in a standard posture and perform a full-body scan on the subject through a three-dimensional human body scanning device to obtain a three-dimensional human body point cloud model containing human body surface information; A key point detection module, configured to identify the positions of human key points based on geometric features of the three-dimensional human body point cloud model according to the preset key point detection rules, thereby forming an initial set of key point positions; perform anatomical key point marking on the three-dimensional human body point cloud model based on the initial set of key point positions to obtain an initial key point marking model; A key point correction module, configured to optimize the correction of key point positions in the initial key point marking model based on the correction rules of human anatomical proportion, symmetry, and neighborhood structure, and perform spatial position matching with the preset reference anatomical structure to correct the key point positions, thereby obtaining the key point correction position data; A bone topology construction module, configured to construct a human bone topology structure based on the key point correction position data, map the key points into the human shape parameter space through a statistical shape model to form human shape space parameters; generate a human mesh model according to the human shape space parameters, combine the simulation of physical-based muscle and fat layers, and apply surface optimization processing to obtain three-dimensional human body surface data.

[0007] Through the 3D human body surface data generation technology based on point cloud, the present invention can efficiently extract accurate 3D human body models from the point cloud data of the human body, perform detailed anatomical marking, skeletal structure construction, and soft tissue distribution modeling on them, and finally generate 3D human body surface data with high realism and high precision. This process ensures the high accuracy and operability of the human body surface model through a series of steps such as precise positioning of key points on each part of the human body, statistical modeling of shape features, and mesh optimization, and can be widely applied in multiple fields. Based on human anatomical features and point cloud data, the method can accurately extract the key point information of the human body and establish precise connection relationships according to the anatomical bone topology structure, thereby constructing a complete human skeletal structure. This process avoids the problems of local bone loss or structural incoherence caused by manual errors or simplified processing in traditional modeling methods, making the generated model have high anatomical consistency. By accurately calibrating and correcting the positions of key points, subsequent modeling and soft tissue distribution analysis become more realistic and reliable. The method analyzes the shape features of the human body through a statistical shape model and converts them into shape space parameters, which can comprehensively reflect various physiological characteristics of the human body, such as height, body shape, joint angles, etc. Using these space parameters, a standard topological structure mesh model can be constructed to ensure that each part of the bone and soft tissue area can be represented in a way that conforms to human anatomical features. Through this technology, highly personalized 3D modeling can be carried out for different physiological characteristics of different individuals, making the generated model more in line with actual needs and widely applied in multiple industries such as medical treatment, sports analysis, and virtual simulation. Using the characteristics of soft tissue distribution to adjust the positions of mesh vertices enables the model to not only contain an accurate expression of the skeletal structure but also simulate the distribution of muscle and fat layers and their effects on the human body shape. Through these processes, the generated model can not only reflect the 3D structure of the bone but also realistically represent the characteristics of human soft tissues in terms of shape, enhancing the biological rationality of the model. This is very important for the development of virtual human models and human body simulation scenarios in sports analysis. The deformation balance calculation under physical constraints further optimizes the realism of the human body model. Simulating the natural shape of soft tissues under the action of gravity can better reflect the posture and movement changes of the human body in the real world, which has important application value in fields such as animation production based on human body models, personalized sports training, and even robot operation. Through the simulation of the physical engine, the deformation of muscle and fat layers in different postures can be simulated, making the performance of the 3D human body model more flexible and realistic in a changing environment. Finally, surface optimization processing further improves the smoothness and precision of the mesh, ensuring the smoothness and detail restoration of the 3D human body surface data.This optimization not only improves the aesthetics of the model but also ensures stability and accuracy in applications. Especially in high-resolution application scenarios, the optimized mesh can clearly display every detail, avoiding the rough or irregular patches that may occur in traditional 3D modeling processes. In summary, through multiple technical means, this method solves common problems in traditional 3D modeling processes, such as low accuracy, cumbersome modeling, and insufficient personalization, and has broad application prospects. Whether in the auxiliary analysis of medical images, the modeling and simulation of sports biomechanics, the human interaction experience in virtual reality, or in the fields of personalized health management and high-precision body measurement, this method can provide high-quality 3D human models to support more accurate analysis and decision-making. Through this technology, users can obtain accurate 3D human data in a shorter time, greatly improving the work efficiency and application effects in related fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Other features, objects, and advantages of the present invention will become more apparent by reading the detailed description of the non-limiting embodiments with reference to the following drawings: Figure 1 It is a schematic flowchart of the steps of a method for generating 3D human surface data based on point cloud according to the present invention; Figure 2 For Figure 1 a detailed flowchart of the steps in step S1; Figure 3 For Figure 1 a detailed flowchart of the steps in step S2. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0009] The technical method of the present invention will be clearly and completely described below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0010] In addition, the drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus their repeated description will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0011] It should be understood that although terms such as "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly, the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0012] To achieve the above object, please refer to Figures 1 to 3 , the present invention provides a method for generating three-dimensional human body surface data based on point cloud, and the method includes the following steps: Step S1: Guide the subject to stand in a standard posture, and perform a full-body scan on the subject through a three-dimensional human body scanning device to obtain a three-dimensional human body point cloud model containing human body surface information; In the embodiment of the present invention, a standard standing posture guiding system is set up in a controlled environment, including posture guiding voice prompts, auxiliary ground positioning marks and reference standing posture schematic diagrams, to guide the subject to stand with feet parallel and shoulder-width apart, arms hanging naturally, palms facing the thighs, facing the three-dimensional scanning device, and keep still for about 5 seconds; then a structured light three-dimensional scanning device is used to collect full-body three-dimensional data. Such a device as Artec Eva or a similar device can project structured light and synchronously collect reflected data by a binocular camera at a distance of 0.5 meters to 1.2 meters from the subject to generate complete human body point cloud data. During the collection, the device performs a 360-degree circumferential scan around the subject, and the collection frequency is controlled at 16 frames per second to ensure that the point cloud density reaches no less than 1000 points per square centimeter, so as to obtain a complete three-dimensional point cloud model including regions such as the face, torso, and limbs. The point cloud data is saved in the.ply or.pcd format, including the three-dimensional coordinates and reflection intensity values of each point, as the basic data for subsequent key point detection and anatomical structure marking.

[0013] Step S2: Identify the positions of human body key points based on geometric features of the three-dimensional human body point cloud model according to preset key point detection rules, so as to form an initial set of key point positions; perform anatomical key point marking on the three-dimensional human body point cloud model based on the initial set of key point positions, so as to obtain an initial key point marking model; After obtaining a complete three-dimensional point cloud model, the embodiment of the present invention uses a human key point detection rule based on geometric features to perform initial key point recognition. The rule includes curvature change detection, normal vector continuity analysis, and point cloud density mutation and other feature calculations. Through these geometric attributes, human feature points with geometric recognition such as the corner of the eye, nose tip, acromion, elbow tip, wrist bone, hip bone, knee tip, ankle bone, etc. are identified; for example, in the identification of the acromion point, the curvature mutation point of the neck and upper arm connection area is extracted, and the center of the area where the local high curvature points are concentrated is selected in combination with left-right symmetry as the candidate key point position; an initial key point position set containing more than 20 key points is formed. Subsequently, the initial key point set is mapped back to the original point cloud model, and each key point is marked with anatomical significance by comparison with a preset anatomical reference database to generate an initial key point marking model. The model not only contains the three-dimensional spatial coordinates of each key point, but also comes with key point type annotation information to form a standardized initial marking set, which is convenient for subsequent optimization and topology construction.

[0014] Step S3: optimizing and adjusting the key point position correction in the initial key point marking model based on the correction rules of human anatomical proportion, symmetry and neighborhood structure, and matching the spatial position with the preset reference anatomical structure to correct the key point position, thereby obtaining the key point correction position data; The embodiment of the present invention further optimizes and corrects the above-mentioned initial key point marking model. First, the distance ratio, angle relationship and left-right symmetry error between adjacent key points are calculated, and compared with the average anatomical proportion in the standard human body model. For example, the ratio of shoulder width to height is usually between 0.25 and 0.28. When it is detected that the shoulder width in the current point cloud model exceeds this range, the position of the acromion point is slightly adjusted in the local normal direction to make it meet the ratio range. In addition, for left-right symmetry, by establishing a three-dimensional point cloud symmetry plane, the mirror error of the left-right symmetrical key points (such as the corners of the eyes, elbows, and knees) is calculated. When the error exceeds a set threshold (for example, 10 mm), a weighted average strategy is used to find the optimal symmetrical matching position between the points on both sides for correction. The corrected key point data is then spatially matched with the reference standard anatomical point model, and the iterative closest point (ICP) method is used to perform position correction in a rigid transformation framework to generate calibrated key point corrected position data, providing an accurate positioning basis for skeletal topology and shape space mapping.

[0015] Step S4: construct the human skeleton topology structure based on the corrected position data of the key points, and map the key points to the human body shape parameter space through the statistical shape model to form the human body shape space parameters; generate a human body mesh model according to the human body shape space parameters, combine the physics-based muscle and fat layer simulation, and apply surface optimization processing to obtain three-dimensional human body surface data.

[0016] In the embodiment of the present invention, the corrected key point position data is input into the human body bone topology generation module, and the mapping from the point set to the topological structure is realized by constructing a skeleton connection relationship graph. The skeleton includes the spinal main line, the limb branch connection lines, the neck and the pelvic support structure. The human skeleton structure formed based on these connection relationships has hierarchy and anatomical consistency. Subsequently, based on the Statistical Shape Model (SSM), these key points are projected into the shape parameter space. The shape space is obtained by training a large number of human body samples and includes the principal component dimension and the deformation weight parameter. This projection process is completed through shape alignment and the principal component regression model, and finally a set of shape space parameters is generated for human body shape reconstruction. According to this parameter, an initial human body mesh model is reconstructed. The model uses about 10,000 patch triangular meshes for body expression. On this basis, a physical-based simulation method is used to simulate the variable thickness of the muscle layer and the fat layer. The simulation is derived based on the length between key points and the body shape characteristics reflected in the shape space. For example, in areas such as the thighs and upper arms, a dynamic layered thickness is introduced, and the muscle layer thickness range is set to 5 to 15 millimeters, and the fat layer thickness range is set to 3 to 25 millimeters. After the simulation, the Laplace surface smoothing process is combined to optimize the mesh continuity, and high-fidelity three-dimensional human body surface data is generated, which can be used in application scenarios such as medical modeling, virtual clothing fitting, and posture evaluation.

[0017] Preferably, step S1 includes the following steps: Step S11: Set the scanning resolution, lighting conditions, and scanning range of the three-dimensional human body scanning device to obtain the scanning device initialization data; In the embodiment of the present invention, parameter settings and initialization are performed on the human body scanning device for three-dimensional scanning. The device used is, for example, a three-dimensional scanner based on structured light (such as Artec Eva, Shining 3D Einscan Pro HD, etc.). In terms of scanning resolution, the longitudinal and transverse scanning accuracies are set to 0.5 millimeters to ensure sufficient point cloud density to restore the surface details of the subject. In terms of lighting conditions, a constant-brightness LED ring light source is used for illumination, and the illuminance is maintained in the range of 2000 to 2500 lux to avoid data deviation caused by changes in external light. The scanning range is reserved with a margin based on the maximum height and arm span of an adult subject, and the scanning space is set as a closed area with a height of 2.5 meters, a width of 2.0 meters, and a depth of 1.5 meters to ensure full-body coverage scanning without moving the scanning object. After the device parameter settings are completed, the scanning software is started to enter the preheating and calibration process, including camera calibration (corrected by a whiteboard), spatial depth alignment, and motion interference detection. Finally, the scanning device initialization data is generated and input as the basic condition for subsequent precise scanning operations.

[0018] Step S12: Guide the subject to stand at the center of the scanning area in a standard posture, with the arms hanging naturally and at a certain angle to the body, the feet separated by a shoulder-width, the head facing straight ahead, and initialize the device according to the scanning device to start the three-dimensional human body scanning device, and perform a full-body scan of the subject from multiple angles and directions to obtain the original three-dimensional point cloud data; In the embodiment of the present invention, according to the scanning device that has completed the initialization settings, arrange the subject to enter the scanning area and stand at the center of the area under the guidance of the auxiliary positioning marking line; the posture setting refers to the ISO 20685 standard human body scanning posture specification. Specifically: stand with the feet naturally separated by a shoulder-width, the soles of the feet facing forward, the arms hanging naturally, the palms gently attached to both sides of the thighs, the fingers naturally separated and kept a gap of 5 to 10 cm from the body, and the head kept horizontal and looking at the target point directly ahead; after the posture is stable, the operator starts the three-dimensional scanning program through the scanning control terminal, controls the scanner to take pictures from multiple angles and perspectives. The scanning path usually adopts an up-and-down reciprocating and circular trajectory. Each perspective covers a range of about 60 degrees during acquisition, and a total of 6 groups of point cloud segments in different orientations are obtained; the device internally synchronously executes timestamp recording and posture alignment processing, and outputs the original three-dimensional point cloud data set. Each segment contains the three-dimensional coordinate information and gray intensity of the point positions, and the output format is uniformly the.pcd data structure for subsequent processing.

[0019] Step S13: Perform preliminary denoising processing on the original three-dimensional point cloud data, and perform spatial registration to unify the point cloud data obtained from multiple scans into the same coordinate system to obtain the fused and registered point cloud data; In the embodiment of the present invention, after obtaining multiple groups of original point cloud data, first perform preliminary denoising processing on the point cloud segments at each scanning angle, and adopt the statistical outlier removal method, that is, analyze the distance distribution of the neighboring points of each point and remove the outliers with local density lower than the set threshold. Usually, set the neighborhood of each point to 20 neighboring points and remove the points with an average distance exceeding the mean plus twice the standard deviation; after denoising, perform spatial registration processing on the point cloud data. Use registration algorithms such as global registration combined with fine registration. Global registration extracts key points through FPFH features (Fast Point Feature Histogram) for initial rough alignment of the initial position, and then uses the ICP (Iterative Closest Point) algorithm to align the scanning segments at each angle into the point cloud coordinate system formed by the first scanning perspective under the rigid transformation model. Finally, fuse the 6 groups of point clouds into a unified three-dimensional data set, that is, the fused and registered point cloud data; retain the high-density area and delete the overlapping and redundant points during the fusion process to ensure the integrity and structural consistency of the model.

[0020] Step S14: Optimize the point cloud density of the fused and registered point cloud data, and perform sparse adjustment between points and normal vector reconstruction based on voxel grid filtering, so as to obtain a three-dimensional human body point cloud model containing human surface information.

[0021] In the embodiment of the present invention, after the point cloud fusion registration is completed, density optimization processing is performed on the obtained three-dimensional point cloud data. This process first evaluates the overall density distribution of the point cloud, calculates the average number of points per cubic centimeter distribution, and performs sparse adjustment if there is a region with a sudden increase in point density; the specific method is voxel grid filtering, which divides the point cloud space into equal-volume cube voxel units. For example, the voxel side length is set to 1 mm, and only the center point coordinates or mean points are retained in each voxel, and the rest of the points are deleted to achieve sparse adjustment between points and eliminate the computational overhead brought by redundant dense points; then normal vector reconstruction is performed, using K-nearest neighbor search to construct a local point set (the value of K is generally taken as 10 to 20), calculate the tangent plane in the neighborhood of each point and determine its surface normal direction, and complete the missing direction information in the original point cloud, so as to form a three-dimensional human body point cloud model containing spatial geometric structure and surface normal description; this model has good uniformity between points and complete surface information, and can be used as the basis for subsequent key point extraction and three-dimensional body modeling, and has high practical value in actual scenarios such as human pose recognition, body shape reconstruction, and customized clothing modeling.

[0022] Preferably, the specific human key points in step S2 include 47 points: the top of the head point, the left and right tragus points, the left and right infraorbital points, the glabella point, the nasal bridge point, the submental point, the occipital point, the cervical vertebra point, the left and right acromion points, the mid-thoracic point, the left and right nipple points, the left and right iliac crests, the left and right anterior superior iliac spines, the left and right ulnar styloid processes, the left and right ulnar styloid process points, the left and right tibia points, the left and right lateral malleolus points, the left and right upper patella margin points, the left and right lateral root of the neck points, the suprasternal notch point, the left and right anterior axillary points, the left and right posterior axillary points, the umbilicus point, the left and right greater trochanter points, the left and right gluteal peak points, the left and right radius points, the left and right mid-patella points, the left and right radial styloid processes.

[0023] Preferably, step S2 includes the following steps: Step S21: Perform spatial partitioning processing on the three-dimensional human body point cloud model, divide it into main anatomical regions of the head region, trunk region, upper limb region and lower limb region, and record the spatial boundaries and point cloud density distributions of each region to obtain regional point cloud density distribution data; After obtaining the three-dimensional human body point cloud model containing normal vector information in the embodiments of the present invention, for the convenience of subsequent human body key point extraction and three-dimensional feature analysis, it is necessary to first perform spatial partitioning on the point cloud model. According to the human anatomical structure, the model is divided into four main regions: the head region, the torso region, the upper limb region, and the lower limb region. This process is achieved through human body symmetry axis estimation and initial segmentation of feature positions. First, based on the center of the torso with the highest point density in the point cloud model, the human body center of gravity axis is constructed (i.e., the main axis projection of the body center point in the Y-axis direction). Combining the Z-axis height threshold and the X-Y plane distribution pattern, the region growing and crop box methods are used to perform region segmentation on the point cloud: the head region is defined as the region about 20 cm above the center of gravity and with a rapidly decreasing point density upwards; the torso region is the region within about 30 cm above and below the center of gravity and with the largest volume; the upper limb region is the extended part on the left and right of the torso, usually determined by detecting symmetric point groups with protrusions on both the left and right sides; the lower limb region is the entire region below the torso. The corresponding sub-point clouds are extracted from each region and their boundary information is recorded, including the minimum / maximum X, Y, and Z coordinate values. At the same time, the point density inside the point cloud of each region is statistically analyzed. The point density is defined as the number of points per unit volume (per cubic centimeter), forming the region point cloud density distribution data, which provides a structural basis for subsequent curvature and key point extraction.

[0024] Step S22: Calculate the local curvature value and normal vector information for each point in the region point cloud density distribution data, and generate a full-body curvature distribution map and normal vector field data; After obtaining the region point cloud density distribution data in the embodiments of the present invention, in order to further extract the local deformation characteristics of the human body surface, it is necessary to calculate the local curvature and normal vector information for each point, and generate a full-body curvature distribution map and normal vector field data. The specific processing flow is as follows: taking each point as the center, select its neighboring point set in the 3D space (K-nearest neighbor search, K = 20), and use this neighborhood to construct a local fitting plane and calculate the principal curvature. The local curvature value can be estimated by the change rate of the normal vector in the principal direction. Larger curvature values are commonly found at positions such as joints, fingertips, the tip of the nose, and knees. The normal vector calculation uses the PCA principal component analysis method to estimate the normal direction of the best fitting plane within the neighborhood, and the results are uniformly stored as a three-dimensional component vector of the normal vector, with the direction uniformly pointing outside the point cloud. The curvature values of all points are visualized in pseudo-color to form a three-dimensional human body curvature distribution map, and the normal vector information forms a vector field, representing the local orientation characteristics of the human body surface. The curvature and normal vector data obtained in this stage provide a quantitative index basis for subsequent key point geometric feature recognition, and can significantly improve the recognition accuracy, especially with high discrimination in regions with morphological mutations.

[0025] Step S23: According to the preset key point detection rules, the 47 human key points are respectively subjected to geometric feature recognition and position calculation based on the whole body curvature distribution map and the normal vector field data to obtain human key point data, including point extraction of the head area, point extraction of the torso area, point extraction of the upper limb area, and point extraction of the lower limb area; After constructing the full body curvature distribution map and normal vector field, the embodiment of the present invention performs geometric feature recognition and spatial position calculation on 47 three-dimensional human standard key points according to the pre-set human key point detection rules. The key point detection rules combine the prior knowledge of anatomical structure with local geometric indicators, specifically including: in the head area, positioning by detecting high curvature protrusions and symmetry such as the nose tip, ears, chin (chin); in the trunk area, extracting the local curvature minimum points and their relative height relationships of the suprasternal fossa, umbilicus, acromion and other points; in the upper limb area, identifying the elbow joint, wrist joint, fingertip and other points through sudden changes in curvature and jumps in the direction of the normal vector; in the lower limb area, positioning the key points by detecting the normal change rate of the knee convexity, ankle joint position, toe end point and other areas. During the key point extraction process, a search window (for example, a 5-cm radius neighborhood) is set for each target point, and the local extrema, symmetry of the axis of symmetry, and spatial position relationship are calculated. At the same time, the left and right key points are ensured to meet the physiological symmetry conditions, thereby extracting the three-dimensional coordinate information of all 47 standard key points of the human body and storing them as a structured key point dataset. The data format is usually JSON or CSV, and each record contains the point name, the area to which it belongs, and its spatial three-dimensional coordinates.

[0026] Step S24: performing spatial position extraction processing on the key point data of the human body, thereby forming an initial key point position set; After acquiring the key point data of the human body, the embodiment of the present invention further independently extracts the coordinates of all key points from the original point cloud to form an initial key point position set. This set not only contains the three-dimensional coordinates of each key point, but also the anatomical region, curvature value and local normal vector information of the point. The extraction method is to traverse the three-dimensional human point cloud model, compare the coordinates of each point with the identified key point set, and extract matching points within the error threshold (such as 1 mm); at the same time, in order to ensure that the corresponding relationship between the key point and the model is consistent, the index number of each key point in the point cloud model is also recorded, so that subsequent operations can be directly mapped to the original model structure. This initial key point position set serves as the key input for subsequent posture analysis, surface feature fitting and individual model adjustment, and has integrity and spatial consistency.

[0027] Step S25: marking anatomical key points on the three-dimensional human body point cloud model based on the initial key point position set, thereby obtaining an initial key point marked model.

[0028] After obtaining the set of initial key point positions in the embodiments of the present invention, it is necessary to mark them back into the three-dimensional human point cloud model to complete the intuitive identification of anatomical key points and form an initial key point marking model. The specific implementation method is as follows: Load the three-dimensional point cloud model and the key point set in a visualization tool (such as MeshLab or PCL Viewer), generate marker points at the corresponding positions on the point cloud surface through spatial coordinate mapping. The marking method can use small spheres to represent or highlight the key points with different colors. At the same time, attach a label identifier to each point, such as "Left_Shoulder", "Right_Knee", etc. At the data level, the key point information is bound as a label attribute to the attribute table of the corresponding point in the point cloud to construct a marked point cloud structure. This initial key point marking model can be used not only for human body measurement and pose analysis, but also for subsequent tasks such as surface mesh reconstruction and dynamic tracking modeling. Its marking structure has high consistency and anatomical interpretability, and provides a key basis for practical application scenarios such as medical rehabilitation, virtual fitting, and sports analysis.

[0029] Preferably, the extraction of key points in the head region in step S23 includes: Calculate the highest point of the head region based on the whole body curvature distribution map and the normal vector field data, and select the point with the largest vertical coordinate value and the angle between the normal vector and the vertical direction less than 15° as the vertex of the head; Analyze the shape index and depth information of the auricle region, and locate the feature points at the junction of the depression and protrusion in the middle of the auricle as the tragus points on the left and right; Analyze the curvature change in the front region of the face, identify the obvious curvature change at the lower edge of the orbit, and select the point corresponding to the midpoint of the lower edge of the orbit as the infraorbital points on the left and right; Determine the saddle point between the left and right superciliary arches as the glabella point according to the symmetry analysis of the facial contour and the surface fitting of the forehead region; Analyze the vertical curvature profile of the nasal region and determine the highest point of the nasal bridge as the nasal bridge point; Calculate the curvature change in the mandibular region, and select the point with the largest curvature change in the mandible and located at the bottom of the mandible as the submental point; Calculate the curvature distribution in the posterior region of the head, and select the point with the largest curvature of the occipital protuberance as the posterior occipital point.

[0030] Based on the initial key point marking model generated in step S25 of the embodiments of the present invention, for the sub-point cloud model of the head region, it is sorted according to the Z-axis coordinate (usually representing the vertical direction) value, and several points with the largest Z values (such as the top 1%) are preferentially selected. Further, points with the included angle between the normal vector and the Z-axis less than 15 degrees are screened out, that is, the normal vector direction of the selected points is basically perpendicular to the ground and points upward. The screening of this included angle can be obtained by calculating the dot product of the normal vector and the unit vector of the Z-axis and performing an inverse cosine transformation to obtain the included angle value. Finally, the point with the largest Z value is selected from the points that meet the conditions as the "top of the head point". This point is usually located slightly behind the center of the top of the skull and is an important reference point in many scenarios such as hat fitting and head circumference measurement. Based on the curvature distribution data of the head region, a point cloud subset of the auricle region is extracted, and the left and right ear regions can be extracted by setting a Y-axis offset range and an X-axis coordinate constraint window on both sides of the head (for example, offset 60 to 80 millimeters from the center of the head on both sides). Inside each ear region, the "shape index" and "depth information" of each point are calculated. The shape index is an index characterizing the local curvature type (such as concave, convex or saddle-shaped), usually derived from the ratio of the principal curvatures and the mean curvature, and the depth information is obtained by fitting the ear surface to form a local surface and measuring the distance from each point to the lowest point of the surface groove. In the middle of the auricle, there is usually a junction of a depression and a protrusion, which is the position of the tragus. The transition point can be determined by detecting the position where the shape index changes from negative (concave) to positive (convex) and combining the position of the largest change in the depth value. Finally, the point corresponding to this position is selected as the left tragus point and the right tragus point. This point is located at the front edge of the ear canal and is an important position reference for glasses fitting and hearing instrument modeling. In the front region of the face, by reading the sub-point cloud within the range where the Z-axis height in the curvature distribution map is approximately in the eye region (30 - 60 millimeters below the top of the head), curvature analysis is performed on each point in the X-axis horizontal direction to find the region of sudden change in the horizontal curvature, which is exactly the orbital margin. The suborbital point is the central depression point at the lower edge of the orbit. Therefore, the change trend of the first derivative of the curvature can be calculated separately in the left and right eye regions, and the curvature section with the largest change rate is selected as the orbital margin. Further, the corresponding coordinates of the midpoint in the X-axis direction in this section are selected as the left and right suborbital points. This point plays an important role in face recognition and expression fitting and is often used to establish an eye reference frame in medical plastic surgery and virtual expression capture. After determining the suborbital points, the region above the superciliary arch is further analyzed. It is located within about 20 millimeters above the upper edge of the eyes. In this region, sub-point clouds corresponding to the left and right superciliary arches are extracted and three-dimensional surface fitting is performed respectively (such as bilateral fitting of a quadratic surface). Then, based on the symmetry of the facial midline, the saddle point position in the X-axis direction between the two fitted surfaces is found, that is, the lowest point at the midpoint between the two superciliary arches. The normal vector of this point is usually approximately perpendicular, and the curvature shows a convexity in the front-back direction and a concavity in the left-right direction, with an obvious saddle-shaped structure. Therefore, it is also called the "glabella point" or "nasion point".This point is the central reference point for facial symmetry and is commonly used in face normalization, standard facial feature modeling, and glasses fitting analysis. In the point cloud of the nasal region (within a range of approximately 40 mm below the glabella), the anterior lateral points are extracted, and a vertical curvature profile of the nasal bridge is constructed along the Y-axis direction (the anteroposterior axis of the human body). The principal curvature value of each point in the profile along the vertical direction is calculated. Usually, the highest point of the nasal bridge corresponds to the point with a local curvature minimum and a normal vector approximately perpendicular and forward. By traversing this profile and finding the highest point before the position where the Z-axis coordinate gradually decreases and the curvature changes abruptly, the nasal bridge point can be determined. This point is usually slightly lower than the glabella and is a key reference in glasses nose pad modeling, breathing mask fitting, and facial symmetry adjustment. The mandibular region is the lowermost region of the head point cloud, located approximately 80 to 100 mm below the tip of the nose. By extracting the point cloud data of this region, calculating the curvature change within its local neighborhood (e.g., a radius of 5 mm), finding the point with the largest overall curvature value, and screening it in combination with its geometric position at the bottom of the mandible. This point is usually located at the lowest depression where the mandibular margin transitions to the neck, and the normal vector usually tilts forward and downward, making it one of the important reference points for chin contour modeling and head orientation pose recognition. The posterior head point cloud is the region with a relatively large Z-axis height and a relatively backward Y-axis depth, specifically within a range of approximately 60 to 120 mm below the vertex of the head and in the region where the Y-axis is negatively offset. After extracting the point cloud of this region, the principal curvature value of the points in the local neighborhood is calculated. The location with the largest curvature usually corresponds to the occipital protuberance. The shape of this part is convex, slightly below the middle of the back of the human head, and the curvature shows a very large rate of change along the Y direction. By screening the point with the largest principal curvature and a normal vector pointing backward in this region as the posterior occipital point. This point plays a key role in helmet modeling, the design of posterior head protection devices, and sleeping posture assessment.

[0031] Of particular importance is that the extraction of the point positions in the torso region in step S23 includes: According to the whole-body curvature distribution map, the normal vector field data, combined with the extraction of bone protrusion features and the analysis of the central axis, select the high-curvature point corresponding to the seventh cervical vertebra protrusion on the midline of the posterior neck as the cervical vertebra point; Calculate the curvature peak distribution in the shoulder region, and select the highest point on the outer contour of the shoulder with an obvious curvature change as the left and right acromion points; Analyze the normal vector distribution in the anterior chest region, and select the point at the center of the sternum with the smallest angle between the normal vector and the forward direction as the mid-chest point; Calculate the curvature distribution in the chest region, and select the curvature peak points in the left and right breast regions as the left and right nipple points; Analyze the curvature change in the lateral waist region, and select the curvature change point at the junction of the lateral waist contour and the upper edge of the ilium as the left and right iliac crests; Calculate the curvature distribution above the inguinal region, and select the high-curvature point at the most prominent part of the anterior superior iliac spine as the left and right anterior superior iliac spines; Analyze the curvature changes in the neck-shoulder connection area, and select the curvature change points at the connection of the side of the neck and the shoulder as the left and right outer lateral neck root points; Calculate the normal vector distribution in the bottom area of the front side of the neck, and select the point with the smallest angle between the normal vector and the forward direction at the center point of the suprasternal fossa as the neck pit point; Analyze the anterior curvature changes in the armpit area, and select the point at the deepest depression in the anterior side of the armpit as the left and right anterior armpit points; Calculate the posterior curvature changes in the armpit area, and locate the point at the deepest depression in the posterior side of the armpit as the left and right posterior armpit points; Analyze the curvature distribution in the abdominal area, and select the point at the center of the umbilical depression in the center of the abdomen as the umbilicus point.

[0032] In the embodiments of the present invention, the collected three-dimensional whole-body point cloud data is subjected to normal vector field calculation and Gaussian curvature calculation. The bony protrusions are identified by constructing the central axis of the human back and searching for the curvature peak points on it. The seventh cervical vertebra is the most prominent bone point at the back of the human neck, which appears as a high-curvature point near the shoulder on the neck midline in the point cloud. To improve the recognition accuracy, the local spherical fitting method is used to obtain the principal curvature value of the point. Combined with the central axis filtering, only the points with the maximum vertical curvature and within the neck region on the central axis are retained as candidate cervical vertebra points. Then, a lightweight classifier trained with manually labeled data is used for secondary screening, and finally, the high-curvature point corresponding to the seventh cervical vertebra protrusion is determined as the cervical vertebra point. This processing flow is particularly suitable for the stable extraction of neck reference points in medical human body modeling and pose analysis. Based on the position of the known cervical vertebra points in the aforementioned point cloud data, the shoulder region point cloud is intercepted by horizontally extending from the cervical vertebra points, the principal curvature value of each point in this region is calculated, and the curvature peak distribution map is extracted. The acromion is the most prominent bony structure on the outer side of the shoulder, which appears as the outermost point with a significant curvature change and far from the cervical vertebra points in the horizontal section in the point cloud. To enhance the stability, a neighborhood with a radius of 20 mm is used to perform weighted averaging on the local curvature, and the curvature change threshold is set to 0.02 to identify the outer contour mutation region. Finally, the points with the maximum curvature and on the outer edge of the contour are selected on the left and right sides respectively as the left and right acromion points, which is applicable to shoulder motion modeling applications such as virtual fitting and rehabilitation monitoring. According to the cervical vertebra points and acromion points, a point cloud subset of the upper chest region is constructed. After calculating the normal vector of each point, by analyzing the angle between the normal vector and the forward direction of the human body (usually the Z-axis direction in the point cloud space), the point set facing directly forward is screened out. The center of the sternum is usually located on the body midline, and its normal vector direction is almost parallel to the forward direction. Therefore, the angle threshold can be set within 15 degrees. After extracting the qualified points and then using the median value of the chest height as a constraint, finally, the point with the middle vertical coordinate value and the smallest angle on the midline is selected as the chest midpoint. This point has important positioning significance in scenarios such as medical image alignment and chest stent wearing. Based on the position of the aforementioned chest midpoint, the breast region range is determined, which is usually the region within 70 - 100 mm horizontally extending from the chest midpoint and 80 mm vertically downward. The curvature distribution map of the point cloud in this region is calculated, and the nipple position characteristics are identified by detecting the local maximum points of the principal curvature or Gaussian curvature. In actual operation, to eliminate the interference of skin folds and surface noise, the multi-scale curvature analysis method is adopted to compare the curvature change stability under different neighborhood sizes, and the points with stable curvature peaks and symmetric distributions are extracted on the left and right sides respectively as the left and right nipple points. This processing is applicable to application scenarios such as breast reconstruction and prosthesis design that require precise breast positioning. Based on the already located nipple points and chest midpoint, the point cloud subsets of the abdomen and waist are constructed downward, and the point cloud is sliced and analyzed for curvature changes along the side direction of the human body (positive and negative X-axis directions).The upper edge of the ilium is located on the outer side of the waist, which is basically consistent with the point of sudden change in the curvature of the body side. The junction area is located by calculating the contour curvature and detecting the point of maximum change in its first derivative. Points with a curvature change rate greater than 0.04 and located in the lower 1 / 3 area of the trunk are selected as candidate points, and the left and right iliac crest points are screened by combining left-right symmetry. This feature point plays an important role in pelvic modeling and the fitting analysis of wearable devices. Further analyze the point cloud data in the groin area downward, calculate the curvature distribution map in the area and locate the most prominent bony structure at the anterior superior iliac spine. The anterior superior iliac spine often appears as a point with a high curvature peak. Set the local neighborhood radius to 10 mm, screen points with a curvature greater than 0.05, and then determine the left and right anterior superior iliac spine points through spatial position constraints (located obliquely below the iliac crest point and with a horizontal distance from the iliac crest point less than 80 mm). This point is often used for the extraction of bony constraint points in lower limb motion analysis and sports orthopedic design. Extract local point clouds from the area extending from the cervical vertebra point to the acromion point on both sides. By calculating the curvature gradient, identify the connection transition zone between the neck and the shoulder. The characteristic of this area is that the curvature changes from high to low and then turns high, showing a typical curvature mutation point. Use the sliding window method to extract the one-dimensional curvature profile and detect the point with the largest change in the first derivative as the candidate point. To ensure left-right symmetry, only retain points that are equidistant from the cervical vertebra point and have a curvature mutation value greater than 0.03 as the left and right lateral cervical root points. This point helps to realize the modeling of the neck-shoulder junction and is crucial in intelligent wear and pose tracking. Analyze the normal vector distribution in the bottom area of the front side of the neck and locate the suprasternal fossa depression area. The suprasternal fossa is a depression point located on the upper edge of the manubrium sterni and usually appears as the lowest depression point on the midline in the three-dimensional point cloud. After calculating the normal vectors, screen out the point set with an angle less than 10 degrees with the forward direction of the body, sort them using the Z coordinate value and select the lowest point as the center point of the suprasternal fossa, and then determine the jugular fossa point. This point is often used for the calibration of the head and neck model and the neck reference marker in medical navigation. Construct a subset of the axillary area point cloud based on the extended area of the mid-thoracic point and the nipple point, and analyze the curvature profile of its anterior contour. The anterior axillary depression is usually located at the junction of the upper arm and the chest, and it appears as the lowest curvature point in the curvature profile. Obtain the contour line of the axillary area by slicing the point cloud, detect the point with the minimum curvature on the contour line as the axillary depression point, and then match the symmetric positions on the left and right sides respectively, and select the point with the minimum depth value as the left and right anterior axillary points. This point is of guiding significance for upper limb dynamic modeling and the design of axillary structure supports. Use a method similar to the anterior side to analyze the curvature profile of the subscapular area and screen out the lowest point of the posterior depression. The posterior axillary point is generally located in front of the lower edge of the scapula and appears as the point with the lowest vertical curvature value in the point cloud. By combining the normal vector direction to distinguish the back-facing point set and extract the point with the minimum curvature value, further identify the left and right posterior axillary points through left-right symmetry constraints to ensure its stable positioning. This point is widely used in the design of back braces and the modeling of wearable interaction feedback systems. Based on the point cloud data in the central abdominal area, extract the deepest point of the central depression by analyzing the normal vector and curvature changes in the Z direction.The umbilical depression is usually located on the midline of the human body and has a relatively low vertical height. First, the abdominal boundary is determined by the aforementioned mid-thoracic point and iliac crest point, and the point with the minimum curvature and the maximum depth on the midline is extracted as the umbilical point. This point plays a key role in abdominal center positioning, aesthetic medical modeling, and clothing body-fitting design.

[0033] Especially importantly, the point extraction in the upper limb region in step S23 includes: Based on the whole-body curvature distribution map and the normal vector field data analysis of the curvature change in the forearm and wrist junction region, the high-curvature points corresponding to the medial protrusion of the distal ulna are selected as the left and right ulnar styloid processes; Calculate the curvature change in the lateral wrist region, and select the high-curvature point corresponding to the lateral protrusion of the distal radius as the left and right radial styloid processes; Calculate the curvature change in the lateral forearm region, and select the high-curvature point at the most prominent part of the lateral radius as the left and right radial points; Calculate the curvature change in the lateral wrist region, and select the high-curvature point corresponding to the lateral protrusion of the distal radius as the left and right radial styloid processes.

[0034] In the embodiments of the present invention, when extracting the styloid processes of the left and right ulnas, first, the area from the acromion point to the anterior and posterior axillary points is used as the initial upper limb region. The point cloud of the forearm region is screened by combining the region growing algorithm with the normal vector direction, and a local curvature distribution map is constructed in this region. Then, combined with the normal vector field information, the junction region between the forearm and the wrist is located, and the detail curvature change points are extracted by means of high-pass filtering. The curvature peak point near the medial side of the distal forearm is identified, and this point usually corresponds to the styloid process of the ulna at the distal end of the ulna, which is a bony structure protruding near the little finger side of the forearm in human anatomy. To enhance stability, in this embodiment, the lower limit of the curvature threshold is set to 0.15 for the curvature points extracted from both the left and right sides, and the curvature change within a 5-mm neighborhood needs to be higher than twice the average value to improve the accuracy. When extracting the styloid processes of the left and right radii, based on the normal vector field distribution of the forearm-wrist connection area in the previous processing step, the point cloud data on the lateral side of the wrist is further refined. A slice curvature profile is established for the cross-section of the wrist, and the position of the styloid process of the radius on the lateral side of the distal radius can be obtained by finding the local most prominent curvature peak point in the radius side region. The styloid process of the radius is located on the wrist near the thumb side in medicine and is an important positioning point clinically. In this step, multi-scale Gaussian curvature analysis is used, and the radius of the curvature window is set to 3 mm to ensure high-resolution ability for the extraction of the structures of the superficial skin protrusions. When extracting the left and right radii, following the processing results of the forearm region point cloud, the overall curvature trend on the lateral side of the radius is mainly analyzed, and the bony protrusions are screened by the stability of the angle between the normal vector and the forearm axis. In the specific operation, the local principal curvature analysis method is used to extract the curvature line on the lateral side of the radius, and the point with the largest curvature amplitude in this line is identified as the radius point. This point is usually located on the lateral side of the middle and lower segments of the radius and is an obvious bony landmark anatomically, which is used for hand gesture analysis and exoskeleton matching modeling. In this embodiment, an adaptive threshold is set for the curvature change intensity to adapt to the shape differences of individuals with different body types.

[0035] Particularly importantly, the extraction of the point positions in the lower limb region in step S23 includes: Calculating the curvature change in the lateral hip region based on the whole-body curvature distribution map and the normal vector field data, and determining the high-curvature point at the corresponding lateral protrusion of the greater trochanter of the femur as the left and right greater trochanter points; Analyzing the curvature distribution in the buttock region and selecting the high-curvature point at the most prominent part of the buttock as the left and right hip peak points; Analyzing the curvature distribution in the knee joint region and selecting the point corresponding to the center of the patella as the left and right mid-patella points; Calculating the curvature distribution in the anterior lower leg region and identifying the high-curvature point at the most prominent part of the anterior edge of the tibia as the left and right tibia points; Analyzing the curvature change in the ankle joint region and selecting the high-curvature point corresponding to the lateral malleolus protrusion as the left and right lateral malleolus points; Calculating the curvature distribution in the knee joint region and determining the curvature change point corresponding to the upper edge of the patella as the left and right upper patella points.

[0036] In the embodiments of the present invention, when extracting the left and right greater trochanter points, based on the hip bounding box constructed from the aforementioned anterior superior iliac spine points and iliac crest points, a range extended 20 mm downward is selected as the hip lateral point cloud extraction area; the curvature distribution of this area is calculated, and by screening local curvature maxima and combining the characteristic of the normal vector facing outward, the protruding part corresponding to the femoral greater trochanter is identified. This part protrudes spherically in the human body and is an important bony point for the connection of the lower limbs. Since the tissue around this point changes greatly, in the extraction process of this embodiment, a morphological filtering algorithm based on difference interpolation is used for background removal, and the curvature gradient change threshold is set to 0.2 to avoid misjudgment due to muscle undulation. When extracting the left and right gluteal peak points, based on the greater trochanter points extracted in the previous step, a curvature map of the buttock area is constructed in a range 30 mm behind it, and a local maximum detection algorithm is used to identify the protruding area, and the outermost protruding part of the buttock is determined as the gluteal peak point. The gluteal peak refers to the highest point on the outermost side of the buttock when a person stands, and its position is of great significance for personalized body shape modeling and posture recognition. To enhance the robustness of the extraction, the curvature peak width control parameter is set to 10 mm to ensure accurate positioning under different postures. When extracting the midpoints of the left and right patellas, first, a central axis framework from the lower limb pelvis to the knee is constructed according to the left and right anterior superior iliac spine points and iliac crest points, and the point cloud in the front of this area is sliced to obtain the point cloud in the knee joint area; then, the point set in the positive front direction is screened according to the normal vector direction, and the central point of the knee convex area in this area is analyzed using a morphological edge detection method. This point corresponds to the position of the patella center point. In this step, the width of the high curvature band is set not to exceed 12 mm in combination with the body surface morphology and anatomical structure characteristics, and the normal vector included angle less than 15 degrees is used as the positive front screening criterion. When extracting the left and right tibia points, 30 cm of point cloud is screened downward from the knee joint area as the anterior lower leg area, a curvature distribution curve of the anterior tibia edge is established, and a one-dimensional curvature change detection algorithm is used to find the point with the maximum curvature. This point corresponds to the protruding part of the anterior tibia edge and is a relatively obvious bony high point on the anterior side of the lower leg. During implementation, curvature statistics are performed along the lower leg axis, and the recognition accuracy is enhanced by combining the characteristic of the normal vector facing forward. The lower limit of the curvature change rate is set to 0.18 to avoid misselection. When extracting the left and right lateral malleolus points, the ankle joint area is screened based on the distance from the tibia point to the ground, and the curvature change of the lateral point cloud is mainly analyzed to extract the protruding part of the lateral malleolus bone. The lateral malleolus is the bony high point closest to the outer side of the foot in the human anatomical structure and is stably visible in the static standing posture. In this embodiment, spherical fitting and normal vector clustering methods are used to enhance the processing of the lateral protruding area, and the minimum point cloud density requirement is set to 100 points per square centimeter during extraction to ensure the extraction accuracy. When extracting the upper edge points of the left and right patellas, return to the point cloud area in the front of the knee joint, establish an upper edge cross-section contour by aligning with the midpoint of the patella, and use the curvature cross-section method to analyze the curvature change of this contour, and find the edge point where the curvature transitions from high to low as the upper edge point of the patella. This point can be used for the bony constraint of supporting knee joint posture estimation and gait capture.To enhance stability, this step performs five-point smoothing on the cross-sectional profile and sets the change threshold to more than 1.5 times the local average curvature to ensure the anatomical rationality of the selected point positions.

[0037] Preferably, step S25 includes the following steps: Step S251: Structure the set of initial key point positions into a key point data table containing coordinate information and point identifiers. In the embodiment of the present invention, based on the three-dimensional human point cloud data for which point extraction has been completed, the three-dimensional coordinate information of each key point is extracted, that is, the spatial position values in the X, Y, and Z directions. Then, a structured key point data table is constructed in combination with the anatomical identifiers of each point (such as "left acromion point", "right greater trochanter point", etc.). The key point data table is stored in a standard table structure, where each row represents a key point, and the fields include point number, point name, spatial coordinates (X, Y, Z), the region to which it belongs (such as upper limb, lower limb, torso), and the bone to which it belongs (such as "humerus", "femur", etc.). This table can be saved in CSV format or database format for subsequent processing. In practical applications, to improve data accuracy and subsequent processing efficiency, floating-point numbers with three decimal places are used as the coordinate accuracy. For example, the coordinates of the left acromion point are recorded as (152.678, 304.923, 180.114), thus establishing a standard data interface for subsequent point visualization and skeleton generation.

[0038] Step S252: Establish a key point spatial positioning system on the three-dimensional human point cloud model, determine the exact position of the points on the model according to the three-dimensional coordinate information in the key point data table, and perform visual marking processing to generate marking style data. In the embodiment of the present invention, based on the key point data table constructed in step S251, it is imported into a three-dimensional point cloud analysis system (such as MeshLab or a self-developed visualization platform). First, a local human coordinate system is established, with the center of gravity point or the top of the head of the human body as the reference origin, and the axes are set according to the standard anatomical posture of the human body (such as the Z-axis upward, the Y-axis backward, and the X-axis to the right). Then, the specific positions are located in the point cloud model according to the spatial coordinate information of the key points. The positioning method includes the nearest neighbor search algorithm accelerated by k-d tree indexing, which matches the coordinates in the data table with the nearest points in the point cloud, and further confirms the positioning accuracy through the consistency verification of the point normal vector direction (ensuring that the normal direction is consistent with the target structure). After positioning, a visualization marking tool is used to mark each point on the point cloud with a colored sphere or crosshair. The marking style data includes the color of the point (for example, red represents joint points, and blue represents bony prominences), size (such as the sphere radius is set to 2 mm), and identification text. The final output format is in the form of JSON or PLY with an attached attribute file, serving as the basis for subsequent visual marking.

[0039] Step S253: Perform anatomical key point marking on the three-dimensional human body point cloud model according to the marking style data to form three-dimensional point cloud data with markings; In the embodiment of the present invention, the marking style data generated in step S252 is used to perform key point annotation operations in the original three-dimensional human body point cloud model. Specifically, a point cloud editing engine is used to fuse the coordinates of each point with the point position attributes, identify the key points as special type points in the model, and automatically load their point position names as labels for display in the visualization window. For example, the label "RP_Patella_Center" is displayed at the "right patella midpoint". During implementation, the key point information is embedded into the vertex additional attributes in the PLY or OBJ format file. A "point_type" field is added to the vertex row of the point cloud file to indicate whether it is a key point, and a "label" field is added to mark its name, so that the marked data is integrated with the main body data of the point cloud. The three-dimensional point cloud generated by this method not only has visual recognition but also facilitates subsequent algorithms for programmatic recognition and analysis.

[0040] Step S254: Determine the connection relationship between adjacent key points based on the anatomical structure characteristics of the human body according to the three-dimensional point cloud data, so as to draw bone connection lines; In the embodiment of the present invention, based on the three-dimensional point cloud with key point markings, the connection relationship between adjacent key points is calculated, and the bone connection lines corresponding to the anatomical structure are drawn. The determination of the connection relationship is based on the joint structure relationship defined in the anatomical knowledge graph. For example, the "left acromion point" and the "left elbow point" belong to the "humerus" structure connection. On this basis, a list of connection pairs is constructed. In specific implementation, the key point combinations in the marked point data table are traversed, their spatial coordinates are read, and three-dimensional line segments are used for connection (i.e., drawing spatial broken lines). At the same time, the start and end point numbers, lengths, and spatial directions of each bone line segment are recorded. Visually, the connection lines are represented by thin line segments, with a unified light gray color, a line width set to 1 mm. If it is the main skeleton line (such as the spine, the central axis of the lower limbs), it is thickened to 2 mm and different color encodings are used. For example, the "left femur connection line" is dark blue and the "right radius connection line" is green. The result of this step generates a set of structured connection relationship data and visual line segment objects, laying a geometric foundation for the final skeleton visualization model.

[0041] Step S255: Combine the bone connection lines with the three-dimensional point cloud data to generate an initial key point marking model with key point markings and bone connection lines for visualization.

[0042] In the embodiment of the present invention, the bone connection lines generated in step S254 are fused with the point cloud data with key point markings to construct an initial key point marking model. This model superimposes two types of information on the basis of the three-dimensional structure of the point cloud: one is visual marking points, including their coordinates, point names, identification colors, and normal information, and the other is the connection lines of the skeleton structure, including their connection start and end points, geometric forms, and attribute fields. The fusion process is realized by rendering using a visualization engine (such as VTK or Unity3D). The point cloud, marking points, and skeleton line segments are loaded as different layers respectively, and interactive switching, zooming, and rotation are realized. The final output model supports multiple three-dimensional model formats such as PLY, OBJ, and GLTF, and is accompanied by key point and bone connection metadata in JSON format, which is convenient for direct calling in subsequent fields such as motion analysis, virtual human modeling, and medical education. For example, in the clinical rehabilitation scenario, this model can be used to evaluate the deviation of key points or the change of limb connection relationship after the patient's operation, so as to provide an objective basis for pose reconstruction.

[0043] Preferably, step S3 includes the following steps: Step S31: Obtain a human anatomy structure rule library, where the human anatomy structure rule library contains the relative position relationship between key points of the human body, the bone length ratio relationship, and the joint range of motion constraint conditions; In the embodiment of the present invention, a set of human anatomy structure rule libraries is established to guide the rationality analysis and correction of the subsequent key point model. This rule library mainly comes from standard medical atlases, clinical human model databases (such as VisibleHuman Project or Zygote Body data), and anatomical textbook materials, and combines statistical methods to summarize the average spatial relationship between key points under the normal body type of adults. The rule library contains three types of key data: one is the "relative position relationship of key points", such as "the elbow point is usually about 30-35 cm directly below the shoulder point and slightly offset inward"; the second is the "bone length ratio relationship", such as "the ratio of the humerus length to the height is about 0.18-0.21"; the third is the "joint range of motion constraint", such as "the angle range when the knee joint is bent is between 0 and 140 degrees". Each rule includes part identification, reference range value, data source level (such as CT reconstruction or expert annotation), and applicable population characteristics (gender, age, height, etc.). This rule library is saved in the form of a database, and the structured fields support efficient query and matching, and support selecting an appropriate rule subset according to the basic information of the input model.

[0044] Step S32: Perform a rationality analysis on the initial key point marking model according to the human anatomy structure rule library, calculate the deviation value between each key point and the expected anatomical position, so as to generate key point position deviation data; In the embodiment of the present invention, for the human anatomical structure rule library established in step S31, the initial key point marking model constructed in step S25 is loaded. By traversing the key point data and comparing it one by one with the corresponding expected anatomical position relationship in the rule library, the spatial deviation of each key point is calculated. The specific processing method is as follows: taking each key point as the target, determining its actual coordinate value in the model, and querying the spatial relationship of this point relative to its superior bone point or adjacent point in the rule library (for example, the hip point should be located about 15 cm below the umbilical point and be symmetrically distributed left and right). Then, the distance between its actual position and the expected position is calculated through three-dimensional Euclidean distance to obtain the deviation value. All deviation value data is organized into a deviation data table, and the recorded content includes the key point number, name, actual coordinates, expected coordinates, deviation magnitude, deviation direction, and over-limit level (such as mild, moderate, severe). In practical applications, for example, if it is found that the "right knee point" deviates by more than 30 mm, it is recorded as a severe deviation, marked with a red warning, and a prompt is given for subsequent correction processing.

[0045] Step S33: Based on the principle of human symmetry, analyze the position relationship between the left and right symmetric key points of the initial key point marking model, calculate the position difference and deviation direction between the symmetric point pairs, so as to form symmetry deviation data; In the embodiment of the present invention, based on the principle of left-right symmetry of the human body structure (such as the left and right shoulder points, left and right hip points, left and right knee points, etc. should be symmetrically distributed on both sides of the human body central axis), a symmetry analysis is performed on the initial key point marking model. During the operation process, first, the human body central axis is defined as the spatial straight line passing through the nasal tip point and the umbilical point. Taking this central axis as the symmetry reference plane, all point pairs with a symmetric relationship are analyzed one by one. The vertical distance deviation value of the symmetric points relative to the central axis and the spatial errors in the X, Y, and Z directions between the two points are calculated. For example, when analyzing the "left ankle point" and the "right ankle point", if its offset value in the X-axis direction exceeds 10 mm and the height difference in the Z-axis is more than 5 mm, it is determined as a symmetry deviation. Finally, the error data of all symmetric point pairs is organized into a symmetry deviation data table, and the fields include the point pair number, left / right point name, corresponding coordinates, position deviation value, deviation direction, symmetric plane distance difference, and the abnormal symmetric point pairs are visually highlighted in the three-dimensional model by combining the color marking method, so as to facilitate the operator to quickly locate the inconsistent structure area.

[0046] Step S34: According to the connection relationship of the human bone chain in the human anatomical structure rule library, analyze the distance and angle relationship between adjacent key points of the initial key point marking model, judge whether it meets the anatomical structure requirements, and generate adjacent point position relationship evaluation data; In an embodiment of the present invention, in combination with the "skeletal chain connection relationship" in step S31, a spatial relationship analysis is performed on adjacent key point pairs (such as the shoulder point and the elbow point, the hip point and the knee point) in the initial key point marking model to determine whether their distances and angles conform to the normal human anatomical structure. The specific analysis method is as follows: for each pair of connection points, calculate their spatial straight-line distance, and combine the included angle formed by the front and rear bone segments (such as the knee angle formed by the thigh and the calf), and then compare it with the reference interval values provided in the rule library. For example, the length of the humerus connection segment should be between 28 and 35 centimeters, and the angle of the elbow joint when hanging naturally should be close to 180 degrees. The angle calculation method uses the three-point vector included angle calculation, and constructs a bone segment vector by combining the spatial coordinates of adjacent key points in the point cloud data. The evaluation data finally forms an "adjacent point position relationship evaluation table", recording the actual length, expected length, deviation value, included angle value and structural compliance degree (marked as compliant, mildly abnormal, severely abnormal, etc.) between each pair of key points. In specific scenarios such as pose reconstruction or orthopedic postoperative rehabilitation evaluation, this data can be used to determine whether there are abnormal flexion and extension or bone segment dislocation phenomena.

[0047] Step S35: Perform key point position correction processing on the initial key point marking model according to the key point position deviation data, symmetry deviation data, and adjacent point position relationship evaluation data to obtain structure consistency model data; In an embodiment of the present invention, based on the key point position deviation data generated in step S32, the symmetry deviation data generated in step S33, and the adjacent point position relationship evaluation data generated in step S34, a unified key point correction process is performed on the initial key point marking model. This correction process adopts a hierarchical strategy. First, priority is given to correcting points with serious structural errors, such as points with a deviation greater than 30 millimeters or an angle deviation exceeding 20 degrees. Then, an equilibrium adjustment of the left and right point pairs is performed with the principle of minimizing symmetry deviation. Finally, global scale smoothing and constraint repair are performed on the overall structure. The correction algorithm uses a point position adjustment method based on constraint optimization to construct an energy function including the target point, reference point, and rule constraints, and updates the key point position coordinates through iterative optimization. For example, the minimum deviation energy method is used to converge the key points towards the expected position while keeping the bone segment length constraint unchanged. Finally, the corrected three-dimensional key point coordinate data is output, structure consistency model data is generated, and the correction range and point position change amount are highlighted and displayed in comparison with the initial model on the three-dimensional visualization interface, providing key point inputs with good consistency for subsequent human model reconstruction.

[0048] Step S36: Perform spatial position matching between the structure consistency model data and a preset reference anatomical structure to obtain key point correction position data.

[0049] Based on the structural consistency model data obtained in step S35, the embodiments of the present invention perform three-dimensional spatial position matching with a preset standard reference anatomical structure to further accurately locate the final positions of each key point. This matching process uses a method combining rigid registration and local non-rigid adjustment. First, the Iterative Closest Point (ICP) algorithm is used to align the current structural consistency model with the standard template, adjusting the position, rotation, and scaling parameters of the overall model to achieve rigid registration. Subsequently, local non-rigid fine-tuning is performed in key bone segment regions (such as the spine, pelvis, and long bones of the limbs) to ensure that the model fits the individual's morphology. During the matching process, points with a position deviation less than 10 millimeters are stably retained, and the key points with a deviation greater than the threshold are reconstructed according to the reference structure. Finally, a set of key point corrected position data is generated, and the data format includes point number, corrected coordinates, deviation from the standard position, and registration confidence value (e.g., 0.98 indicates high matching accuracy). This data can be directly used for subsequent construction of a three-dimensional human body mesh model or driving a digital human motion simulation system, providing a high-precision human body structure basis in applications such as virtual reality and motion capture.

[0050] Preferably, step S36 includes the following steps: Step S361: Perform spatial registration on the structural consistency model data and the preset reference anatomical structure, and maximize the spatial matching of the two sets of key points through rigid transformation and scale adjustment to obtain the registration transformation parameters; In the embodiments of the present invention, for the spatial registration of the structural consistency model data obtained in step S35 and the preset reference anatomical structure, a three-dimensional point set registration method based on rigid transformation and scale unification is adopted. First, the point pairs that can be directly corresponding in the two sets of key points are determined, such as anatomical landmark points like the top of the head point, shoulder point, hip point, knee point, and ankle point, and a one-to-one mapping relationship is established. Then, the classic Iterative Closest Point (ICP) algorithm is used to calculate the rigid transformation matrix required for the source model (structural consistency model data) relative to the target model (reference anatomical structure model). This matrix includes a three-dimensional rotation vector, a translation vector, and an overall scale factor, which is used to align the two sets of point sets with the minimum error in space. The minimization objective of the ICP algorithm is to minimize the sum of the squares of the Euclidean distances between corresponding point pairs. During the iteration process, the point pair relationship is dynamically updated in the nearest neighbor manner, and the optimal rigid transformation parameters are obtained through the singular value decomposition method. For example, when applied to a set of male point cloud data with a height of about 170 cm, the scale factor is adjusted to 1.02 to match the standard anatomical model, and the rotation vector makes the shoulder connection line consistent with the horizontal direction of the template. The finally output registration transformation parameters include a rotation matrix, a translation vector, and a scaling factor triple parameter set, which can be used for coordinate transformation in subsequent steps.

[0051] Step S362: Calculate the positions of each key point based on the registration transformation parameters and compare them with the initial positions to generate key point correction vector data; In the embodiment of the present invention, the registration transformation parameters obtained in step S361 are used to perform a transformation operation on the coordinates of each key point in the structural consistency model to obtain the expected key point positions in the standard reference structure, and these positions are compared one by one with the actual coordinates of the key points in the original structural consistency model, thereby generating key point correction vector data. The specific operation is as follows: for the coordinates of each key point, apply the rotation and translation transformations obtained by registration to map it to the standard anatomical structure coordinate system; then calculate the three-dimensional vector difference between the transformed position and the original position, and this vector difference is the key point correction vector, which describes the adjustment direction and amplitude required for this key point to meet the standard anatomical structure. Taking the "right shoulder point" as an example, the original position is (130, 230, 520) millimeters, and the standard position is (128, 228, 518) millimeters, then its correction vector is (-2, -2, -2) millimeters. This correction vector data is indexed by the point number and records the original position, reference position, and correction vector of each key point, which can be used for the next step of continuously adjusting and optimizing the key points.

[0052] Step S363: Smooth the key point correction vector data to ensure the continuity and consistency of the correction process of adjacent key points, and generate optimized correction vector data; In the embodiment of the present invention, to avoid problems such as discontinuity or anatomical structure distortion between key points caused by directly adjusting according to the correction vector data generated in step S362, a spatial smoothing algorithm based on topological constraints is used to globally optimize the correction vector data. The specific method is to construct a key point adjacency graph, establish the relationship between adjacent points connected to the current key point (such as the elbow point and the shoulder point, the knee point and the hip point) through the skeletal topology structure, and use the Laplacian smoothing method to adjust the correction vector, that is, the correction vector of each point will be adjusted to the weighted average of its own and the correction vectors of adjacent points to reduce local mutations. For example, if the correction vector of the "right elbow point" is (4, -2, 1) millimeters, and the correction vectors of the connected "right shoulder point" and "right wrist point" are (2, -1, 0) millimeters and (5, -3, 2) millimeters respectively, then the optimized correction vector of the "right elbow point" may be adjusted to (3.7, -2.2, 1.1) millimeters. During the processing, the smoothing weight parameter is set to 0.6 to control the retention degree of the difference in correction vectors between this point and adjacent points. The optimized correction vector data output by this step better meets the continuity requirements of the physiological structure and provides a structurally stable basis for the subsequent actual adjustment of the key point positions.

[0053] Step S364: Adjust the positions of the initial key points according to the optimized correction vector data and perform anatomical validity verification to obtain position verification data; In the embodiment of the present invention, based on the optimized correction vector data obtained in step S363, the position of the key point coordinates in the structure consistency model is updated, and at the same time, anatomical validity verification is performed on the adjusted coordinates. The operation of updating the key point coordinates is to add the corresponding optimized correction vector to the original coordinates of each key point to obtain the corrected coordinates. After completing the coordinate update, the structural validity verification process is started. The verification rules include two aspects: one is to check the bone segment length between adjacent points to ensure that the updated length is still within the normal anatomical range; the other is to detect the angles of key bone segments (such as the torso, upper and lower limbs) to confirm that no non-physiological flexion or torsion occurs at each joint. For example, it is judged whether the bone segment length between the "left knee point" and the "left hip point" is still within the thigh length range of a standard adult (between 350 and 420 mm), and it is verified that the angle between it and the "left ankle point" is between 100 and 170 degrees. If it is found during the verification that the corrected coordinates of some points cause the local structure to exceed the reasonable range, it is marked as "position non-compliant", and the point number, corrected coordinates and failure reason (such as "excessive torsion") are recorded in the position verification data table, providing a basis for the next secondary correction.

[0054] Step S365: Perform secondary correction on the positions of the key points in the structure consistency model data that do not meet the anatomical requirements according to the position verification data, so as to obtain the key point correction position data.

[0055] In the embodiment of the present invention, for the key points where the position verification fails in step S364, a secondary correction operation is performed to ensure that the model meets the complete anatomical structure constraints. In this step, first, according to the list of abnormal points in the position verification data table, the corresponding optimized correction vector and adjacent point information are extracted, and the position of this point is readjusted using the structure recovery algorithm based on local iterative optimization. The operation method is to establish a minimum structure deviation function while retaining the corrected coordinates of its adjacent points. The goal of this function is to simultaneously minimize the distance deviation, angle deviation and left-right symmetry difference between this point and its adjacent points; by introducing a soft constraint mechanism, restrictions are imposed on the joint range of motion, for example, "the maximum bending angle of the knee joint shall not exceed 150 degrees", to prevent non-physiological structures from appearing due to secondary correction. The correction step size is dynamically adjusted during the iterative optimization process, and anatomical verification is performed again after each iteration until the error converges or meets the accuracy requirements. For example, after the initial correction, the "right hip point" deviates from the normal hip width range, and the system adjusts its position 8 mm towards the midline according to the standard value of the pelvic width (about 260 to 300 mm) to restore the structural consistency. The finally output key point correction position data includes the final coordinates of each key point, the correction source identifier (initial correction or secondary correction), the verification pass mark and the anatomical matching degree score, which serve as the basic control points for generating the point cloud-driven three-dimensional human body surface mesh, ensuring that the finally generated human body surface data has high biological rationality and geometric continuity.

[0056] Preferably, step S4 includes the following steps: Step S41: Establish the connection relationship between key points according to the corrected position data of key points to form human body bone topology structure data, where the human body bone topology structure data includes bone nodes, bone connections, and joint range of motion information; In the embodiment of the present invention, according to the corrected position data of key points obtained in step S364, the topology structure of the human body bone is first established according to the anatomical structure and physiological characteristics of the human body. Each key point corresponds to a node of the human body bone, and these nodes are associated through bone connections. For example, the shoulder node is connected to the elbow node, the elbow is connected to the wrist, the hip is connected to the knee, etc. The bone connection relationship clarifies the order and interaction between the bones of each part of the human body. Then, based on the range of motion of the joint and the physiological movement law, the range of motion corresponding to each connection relationship is further determined. For example, the range of motion of the knee joint is limited between 0 and 150 degrees, and the range of motion of the shoulder joint can reach 180 degrees. At this time, the generated bone topology structure data includes the range of motion information of each joint, providing structured bone information for the subsequent three-dimensional human body mesh modeling.

[0057] Step S42: Calculate the human body shape feature vector through the mapping function of the statistical shape model based on the human body bone topology structure data, so as to generate the human body shape space parameters; In the embodiment of the present invention, based on the human body bone topology structure data generated in step S41, the mapping function of the statistical shape model (SSM) is used to calculate the human body shape feature vector. The statistical shape model is a model obtained by analyzing a large number of human body shape sample data. It analyzes the geometric shape features of each part of the human body (such as bone shape, joint spacing, etc.) to establish a statistical model that can describe the overall shape variation of the human body. In actual operation, first calculate the mean value and covariance matrix of the shape features through the training set data. Then, use the mapping function to map the human body bone topology structure data into the shape space to obtain a parameter vector representing the human body shape. This vector can effectively capture the shape differences of different human bodies, such as height, weight, body type and other information. Finally, use these feature vectors as the human body shape space parameters to provide data support for the next step of mesh modeling.

[0058] Step S43: Use the human body shape space parameters to construct an initial human body mesh model with a standard topology structure, determine the initial positions and connection relationships of the mesh vertices, and generate basic mesh data; Based on the human body shape space parameters calculated in step S42, an initial human body mesh model with a standard topological structure is constructed using these parameters. Specifically, first, according to the connection relationship of the human body bone topology structure, the vertex positions of the mesh and their connection methods are determined. Generally, the vertices of the human body mesh represent the three-dimensional coordinates of key points, and the edges of the mesh represent the relationship of bone connections. By mapping the position of each key point to the corresponding vertex of the initial mesh model, a basic mesh data structure can be established. At this time, the basic mesh data includes the connection relationship of each joint of the human body, the vertex coordinates of the mesh, and the topological structure, providing a skeleton basis for the subsequent calculation of soft tissue distribution characteristics.

[0059] Step S44: Calculate the soft tissue distribution characteristics of different regions of the human body based on the human body shape space parameters, including muscle thickness and fat layer distribution, to generate human soft tissue distribution data; Based on the basic mesh data and human body shape space parameters obtained in step S43, the embodiments of the present invention calculate the soft tissue distribution characteristics of different regions of the human body. Specifically, first, soft tissue models of different regions are established according to the muscle thickness and fat layer distribution data of different parts. For example, the muscle layer in the upper body of the human body is thicker, while the fat layer distribution in the abdomen and buttocks is more significant. By combining anatomical knowledge and the soft tissue distribution model, the soft tissue characteristic parameters of each mesh region are generated. These soft tissue data are obtained through a statistical model and include the spatial distribution characteristics of muscle thickness and fat layer, which will affect the adjustment of the mesh and subsequent deformation simulation.

[0060] Step S45: Apply the human soft tissue distribution data to the basic mesh data and adjust the vertex positions of the mesh based on the soft tissue characteristics to generate adjusted mesh data including the influence of soft tissues; Based on the human soft tissue distribution data obtained in step S44, the embodiments of the present invention are applied to the basic mesh data. Specifically, for each mesh vertex, its position is adjusted according to the distribution characteristics of the soft tissues. In particular, the influence of muscle and fat layers on the mesh vertices varies with different parts. A thicker muscle layer will cause the vertices of the mesh to shift outwards, while a thicker fat layer area will cause the vertices to expand accordingly. By making corresponding adjustments to the positions of the mesh vertices, adjusted mesh data including the influence of soft tissues is formed, making the mesh model closer to the true shape of the human body.

[0061] Step S46: Perform a deformation balance calculation under physical constraints on the adjusted mesh data to simulate the natural shape of soft tissues under the action of gravity and generate balanced state mesh data; In the embodiment of the present invention, deformation balance calculation under physical constraints is performed on the adjusted grid data generated in step S45. At this time, by introducing a gravity model, the deformation simulation of the human soft tissue is carried out, and the natural morphological changes of each grid vertex under the action of gravity are calculated. The specific operation is to input the adjusted grid data into a physical engine and apply the finite element analysis method to calculate the deformation of the grid. The deformation amount of each grid is affected by the soft tissue characteristics, gravity, and local constraint conditions. For example, the abdominal area of the human body may sag due to thick fat, while the muscles in the chest area are supported and not easily deformed. Through physical calculation, the equilibrium state of the grid in the natural state is obtained, and the equilibrium state grid data is generated to ensure the biological rationality of the simulation results.

[0062] Step S47: Perform surface optimization on the equilibrium state grid data to generate three-dimensional human body surface data.

[0063] In the embodiment of the present invention, based on the equilibrium state grid data obtained in step S46, surface optimization is performed on it to generate three-dimensional human body surface data. Specifically, the surface optimization is mainly to remove unnecessary redundant patches in the grid and improve the smoothness of the grid. The optimization methods include surface smoothing algorithms and subdivision algorithms. By iterative calculation, the number of grid patches is reduced, while ensuring the smoothness and continuity of the grid surface. For example, the Laplacian smoothing algorithm or the Catmull-Clark subdivision algorithm can be used to optimize the grid surface to make it closer to the real human body shape. Finally, the obtained three-dimensional human body surface data will have high accuracy and biological rationality, and is applicable to fields such as medicine, virtual reality, and animation production.

[0064] The present invention also provides a three-dimensional human body surface data generation system based on point cloud for performing the above-mentioned three-dimensional human body surface data generation method based on point cloud. The three-dimensional human body surface data generation system based on point cloud includes: A three-dimensional scanning module, which is used to guide the subject to stand in a standard posture and perform a full-body scan on the subject through a three-dimensional human body scanning device to obtain a three-dimensional human body point cloud model containing human surface information; A key point detection module, which is used to identify the positions of human key points based on geometric features of the three-dimensional human body point cloud model according to preset key point detection rules, so as to form an initial set of key point positions; anatomical key points are marked on the three-dimensional human body point cloud model based on the initial set of key point positions to obtain an initial key point marked model; A key point correction module, which is used to optimize and adjust the correction of the key point positions in the initial key point marked model based on the correction rules of human anatomical proportion, symmetry, and neighborhood structure, and perform spatial position matching with a preset reference anatomical structure to correct the key point positions, so as to obtain key point corrected position data; A skeletal topology construction module, which is used to construct a human body skeletal topology based on the corrected position data of key points, map the key points into the human body shape parameter space through a statistical shape model to form human body shape space parameters; generate a human body mesh model according to the human body shape space parameters, combine the simulation of physics-based muscles and fat layers, and apply surface optimization processing to obtain three-dimensional human body surface data.

[0065] Therefore, from any perspective, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the application document are intended to be encompassed within the present invention.

[0066] The above description is only a specific implementation manner of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features invented herein.

Claims

1. A method for generating three-dimensional human body surface data based on point cloud, characterized in that, Including the following steps: Step S1: Guide the subject to stand in a standard posture, and conduct a full-body scan of the subject through a three-dimensional human body scanning device to obtain a three-dimensional human body point cloud model containing human surface information; Step S2: Identify the positions of human key points based on geometric features of the three-dimensional human body point cloud model according to preset key point detection rules, so as to form an initial set of key point positions; Perform anatomical key point marking on the three-dimensional human body point cloud model based on the initial set of key point positions to obtain an initial key point marking model; Step S3: Optimize and adjust the correction of the key point positions in the initial key point marking model based on the correction rules of human anatomical proportion, symmetry and neighborhood structure, and perform spatial position matching with the preset reference anatomical structure to correct the key point positions, so as to obtain key point correction position data; Step S4: Construct a human bone topology structure based on the key point correction position data, and map the key points into the human shape parameter space through a statistical shape model to form human shape space parameters; Generate a human body mesh model according to the human shape space parameters, combine with a physics-based simulation of muscle and fat layers, and apply surface optimization processing to obtain three-dimensional human body surface data.

2. The method for generating three-dimensional human body surface data based on point cloud according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Set the scanning resolution, lighting conditions and scanning range of the three-dimensional human body scanning device to obtain scanning device initialization data; Step S12: Guide the subject to stand in the center of the scanning area in a standard posture, with the arms hanging naturally and at a certain angle to the body, the feet separated shoulder-width apart, the head facing straight ahead, and start the three-dimensional human body scanning device according to the scanning device initialization to conduct a full-body scan of the subject from multiple angles and directions to obtain the original three-dimensional point cloud data; Step S13: Perform preliminary denoising processing on the original three-dimensional point cloud data, and perform spatial registration to unify the point cloud data obtained from multiple scans into the same coordinate system to obtain fused and registered point cloud data; Step S14: Optimize the point cloud density of the fused and registered point cloud data, and perform inter-point sparse adjustment and normal vector reconstruction based on voxel grid filtering to obtain a three-dimensional human body point cloud model containing human surface information.

3. The method for generating three-dimensional human body surface data based on point cloud according to claim 2, characterized in that, The specific human key points in Step S2 include 47 points: vertex of head, bilateral tragus points, bilateral suborbital points, glabella point, nasal bridge point, submental point, posterior occipital point, cervical vertebra point, bilateral acromion points, midsternal point, bilateral nipple points, bilateral iliac crest points, bilateral anterior superior iliac spine points, bilateral ulnar styloid processes, bilateral ulnar styloid points, bilateral tibia points, bilateral lateral malleolus points, bilateral upper margin of patella points, bilateral lateral root of neck points, suprasternal notch point, bilateral anterior axillary fold points, bilateral posterior axillary fold points, umbilicus point, bilateral greater trochanter points, bilateral gluteal peak points, bilateral radius points, bilateral midpoint of patella, bilateral radial styloid processes.

4. The method for generating three-dimensional human body surface data based on point cloud according to claim 3, characterized in that, Step S2 includes the following steps: Step S21: Perform spatial partitioning on the three-dimensional human body point cloud model, divide it into main anatomical regions of the head region, trunk region, upper limb region and lower limb region, and record the spatial boundaries and point cloud density distributions of each region to obtain regional point cloud density distribution data; Step S22: Calculate the local curvature values and normal vector information for each point in the regional point cloud density distribution data, and generate a whole-body curvature distribution map and normal vector field data; Step S23: Based on the preset key point detection rules, perform geometric feature recognition and position calculation on 47 human key positions respectively according to the whole-body curvature distribution map and the normal vector field data to obtain human key position data, including point extraction in the head region, point extraction in the torso region, point extraction in the upper limb region, and point extraction in the lower limb region; Step S24: Perform spatial position extraction processing on the human key position data to form an initial key position set; Step S25: Perform anatomical key point marking on the three-dimensional human point cloud model based on the initial key position set to obtain an initial key point marking model.

5. The method for generating three-dimensional human body surface data based on point cloud according to claim 4, characterized in that, The point extraction in the head region in Step S23 includes: Calculate the highest point in the head region based on the whole-body curvature distribution map and the normal vector field data, and select the point with the largest vertical coordinate value and the angle between the normal vector and the vertical direction less than 15° as the top of the head point; Analyze the shape index and depth information of the auricle region, and locate the feature point at the junction of the depression and protrusion in the middle of the auricle as the left and right tragus points; Analyze the curvature change in the front region of the face, identify the obvious curvature change at the lower edge of the orbit, and select the point corresponding to the midpoint of the lower edge of the orbit as the left and right infraorbital points; Determine the saddle point between the left and right superciliary arches as the glabella point according to the symmetry analysis of the facial contour and the surface fitting of the forehead region; Analyze the vertical curvature profile of the nose region and determine the highest point of the nasal bridge as the nasal bridge point; Calculate the curvature change in the mandibular region, and select the point with the largest curvature change in the mandible and located at the bottom of the mandible as the submental point; Calculate the curvature distribution in the posterior region of the head, and select the point with the largest curvature of the occipital protuberance as the posterior occipital point.

6. The method for generating three-dimensional human body surface data based on point cloud according to claim 5, characterized in that, Step S25 includes the following steps: Step S251: Structure the initial key position set into a key point data table containing coordinate information and point position identifiers; Step S252: Establish a key position spatial positioning system on the three-dimensional human point cloud model, determine the precise position of the point on the model according to the three-dimensional coordinate information in the key point data table, and perform visual marking processing to generate marking style data; Step S253: Perform anatomical key point marking on the three-dimensional human point cloud model according to the marking style data to form three-dimensional point cloud data with markings; Step S254: Determine the connection relationship between adjacent key points based on the human anatomical structure characteristics according to the three-dimensional point cloud data, and draw bone connection lines; Step S255: Combine the bone connection lines with the three-dimensional point cloud data to generate a visual initial key point marking model containing key point markings and bone connection lines.

7. The method for generating three-dimensional human body surface data based on point cloud according to claim 6, wherein Step S3 includes the following steps: Step S31: Obtain a human anatomical structure rule library, where the human anatomical structure rule library contains the relative position relationship between key points of the human body, the bone length ratio relationship, and the joint movement range constraint conditions; Step S32: Analyze the rationality of the initial key point marking model according to the human anatomy structure rule base, calculate the deviation values of each key point from the expected anatomical positions, and thus generate key point position deviation data; Step S33: Analyze the positional relationship between the left and right symmetric key points of the initial key point marking model based on the human symmetry principle, calculate the positional differences and deviation directions between the symmetric point pairs, and thus form symmetry deviation data; Step S34: Analyze the distance and angular relationships between adjacent key points of the initial key point marking model according to the human bone chain connection relationships in the human anatomy structure rule base, judge whether they meet the anatomical structure requirements, and generate adjacent point position relationship evaluation data; Step S35: Perform key point position correction processing on the initial key point marking model according to the key point position deviation data, symmetry deviation data, and adjacent point position relationship evaluation data to obtain structure consistency model data; Step S36: Perform spatial position matching between the structure consistency model data and a preset reference anatomical structure to obtain key point corrected position data.

8. The method for generating three-dimensional human body surface data based on point cloud according to claim 7, wherein Step S36 includes the following steps: Step S361: Perform spatial registration between the structure consistency model data and a preset reference anatomical structure, and make the two sets of key point sets match to the greatest extent in space through rigid transformation and scale adjustment, so as to obtain registration transformation parameters; Step S362: Calculate the positions of each key point based on the registration transformation parameters and compare them with the initial positions to generate key point correction vector data; Step S363: Smooth the key point correction vector data to ensure the continuity and consistency of the correction process of adjacent key points, and generate optimized correction vector data; Step S364: Adjust the positions of the initial key points according to the optimized correction vector data and perform anatomical validity verification to obtain position verification data; Step S365: Perform secondary correction on the key point positions in the structure consistency model data that do not meet the anatomical requirements according to the position verification data to obtain key point corrected position data.

9. The method for generating three-dimensional human body surface data based on point cloud according to claim 8, wherein Step S4 includes the following steps: Step S41: Establish the connection relationships between the key points according to the key point corrected position data to form human bone topology structure data, where the human bone topology structure data includes bone nodes, bone connections, and joint range of motion information; Step S42: Calculate the human shape feature vectors based on the human bone topology structure data through the mapping function of the statistical shape model, and thus generate human shape space parameters; Step S43: Use the human shape space parameters to construct an initial human mesh model with a standard topology, determine the initial positions and connection relationships of the mesh vertices, and generate basic mesh data; Step S44: Calculate the soft tissue distribution characteristics of different regions of the human body according to the human shape space parameters, including muscle thickness and fat layer distribution, and generate human soft tissue distribution data; Step S45: Apply the human soft tissue distribution data to the basic mesh data and adjust the positions of the mesh vertices based on the soft tissue characteristics to generate adjusted mesh data including the influence of soft tissues; Step S46: Perform deformation equilibrium calculation under physical constraints on the adjusted grid data to simulate the natural shape of soft tissues under the action of gravity and generate grid data in the equilibrium state; Step S47: Perform surface optimization processing on the grid data in the equilibrium state to generate three-dimensional human body surface data.

10. A system for generating three-dimensional human body surface data based on point cloud, wherein For implementing the method for generating three-dimensional human body surface data based on point cloud as claimed in claim 1, the system for generating three-dimensional human body surface data based on point cloud comprises: A three-dimensional scanning module, configured to guide a subject to stand in a standard posture and perform a full-body scan on the subject through a three-dimensional human body scanning device to obtain a three-dimensional human body point cloud model containing human body surface information; A key point detection module, configured to perform position recognition of human body key points based on geometric features on the three-dimensional human body point cloud model according to preset key point detection rules, so as to form an initial set of key point positions; perform anatomical key point marking on the three-dimensional human body point cloud model based on the initial set of key point positions, so as to obtain an initial key point marking model; A key point correction module, configured to optimize and adjust the correction of key point positions in the initial key point marking model based on correction rules of human body anatomical proportion, symmetry and neighborhood structure, and perform spatial position matching with a preset reference anatomical structure to correct the key point positions, so as to obtain key point correction position data; A bone topology construction module, configured to construct a human body bone topology structure based on the key point correction position data, map the key points into a human body shape parameter space through a statistical shape model to form human body shape space parameters; generate a human body grid model according to the human body shape space parameters, combine with a physics-based muscle and fat layer simulation, and apply surface optimization processing, so as to obtain three-dimensional human body surface data.

Images