Acupuncture point identifying and positioning method and system, moxibustion robot and storage medium

Abstract

The invention discloses an acupoint recognition and positioning method and system, a moxibustion robot and a storage medium. The method comprises the steps of obtaining multi-modal data of a human body to be subjected to physiotherapy; carrying out preprocessing on the multi-modal data; constructing a multi-modal feature fusion model, and generating an enhanced acupoint feature map; constructing a human body meridian-acupoint incidence relation matrix on the basis of the enhanced acupoint feature map in combination with a preset human body meridian-acupoint incidence relation matrix; dynamically compensating the candidate area, calculating the real-time offset of the acupuncture point based on the body surface micro-motion data, and correcting the depth coordinate of the acupuncture point in combination with the body surface elastic modulus data to obtain the accurate three-dimensional coordinate of the acupuncture point to be positioned; on the basis of the precise three-dimensional coordinates, blood vessels and nerve dense areas around the acupuncture points are recognized through an improved Harris corner detection algorithm, and a safe moxibustion applying path is generated; meanwhile, the confidence value of the precise three-dimensional coordinate is calculated. According to the method, high-precision, self-adaptive, all-position compatible, safe and controllable acupuncture point recognition positioning is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of acupoint recognition, and in particular to an acupoint recognition and positioning method and system, a moxibustion robot, and a storage medium. BACKGROUND

[0002] Moxibustion, as a traditional external treatment method of traditional Chinese medicine, highly depends on the accuracy of acupoint positioning. The existing acupoint recognition and positioning technology has the following core problems: Firstly, traditional manual positioning relies on the experience of doctors, and the error is generally more than ±2mm, and is greatly affected by fatigue and subjective judgment. In addition, the existing mainstream machine positioning method, such as single-mode image matching, only compares feature images with reference images, without considering the differences in biological characteristics such as human body elasticity and temperature. The positioning success rate for special body type users such as obese and emaciated users is less than 70%.

[0003] Secondly, dynamic behaviors such as human respiration and muscle micro-movement can cause the acupoint position to shift by ±3mm to ±5mm. For example, the abdominal acupoint positioning method disclosed in CN118141681B and the back acupoint positioning method disclosed in CN118356344A are mostly based on static models and lack real-time compensation mechanisms, which are prone to “off-target” during moxibustion.

[0004] Thirdly, the existing technology mostly designs positioning logic for a single part (such as the face, abdomen, back, etc.), which requires separate algorithm development for different parts, has low hardware compatibility, and does not establish a unified acupoint-meridian correlation model. When cross-part physiotherapy is performed, it needs to be recalibrated, which is inefficient.

[0005] Fourthly, during moxibustion, if there are blood vessels and nerve-intensive areas around the acupoint, most of the existing technology does not combine biomechanical data for path planning, which is prone to safety hazards due to improper moxibustion depth and angle, and lacks acupoint position reliability evaluation, which cannot judge the reliability of the positioning result. SUMMARY

[0006] Therefore, the present application aims to provide an acupoint recognition and positioning method, system, moxibustion robot, and storage medium to solve the technical problems of low positioning accuracy, weak dynamic adaptability, poor multi-part compatibility, and high safety risks in the prior art, and to achieve high-precision, self-adaptive, all-part compatible, and safe and controllable acupoint recognition and positioning.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solutions: The first purpose of the present application is to provide an acupoint recognition and positioning method, comprising the following steps: S1: Obtain multi-modal data of a human body to be treated, the multi-modal data comprising body surface three-dimensional point cloud data collected by a depth camera, body surface temperature distribution data collected by an infrared thermal imager, body surface elastic modulus data collected by a piezoelectric sensor, and body surface micro-motion data collected by a millimeter wave radar; S2: Preprocess the multi-modal data, the preprocessing comprising: registering the body surface three-dimensional point cloud data by an improved ICP algorithm, processing the body surface temperature distribution data by Gaussian filtering and histogram equalization, removing noise in the body surface elastic modulus data by a sliding window filter, and extracting dynamic offsets in the body surface micro-motion data by Kalman filtering; S3: Construct a multi-modal feature fusion model, and perform weight mapping feature level fusion on the preprocessed body surface three-dimensional point cloud data, the body surface temperature distribution data, the body surface elastic modulus data, and the body surface micro-motion data to generate an enhanced acupoint feature map; The weight is dynamically adjusted by a hidden Markov model. S4: Based on the enhanced acupoint feature map, combine a pre-set human body meridian-acupoint correlation matrix to preliminarily determine a candidate region of an acupoint to be positioned, and construct a human body meridian-acupoint correlation matrix; the human body meridian-acupoint correlation matrix comprises a topological relationship of each acupoint and a meridian to which the acupoint belongs, a distance constraint of adjacent acupoints, and a body surface biological feature threshold corresponding to the acupoint; S5: Dynamically compensate the candidate region, calculate a real-time offset of the acupoint based on the body surface micro-motion data, correct a depth coordinate of the acupoint based on the body surface elastic modulus data, and obtain a precise three-dimensional coordinate of the acupoint to be positioned; S6: Based on the precise three-dimensional coordinate, identify a blood vessel and a nerve dense area around the acupoint by an improved Harris corner detection algorithm to generate a safe moxibustion path; and calculate a confidence value of the precise three-dimensional coordinate. If the confidence value is greater than a pre-set threshold, output the precise three-dimensional coordinate and the safe moxibustion path; if the confidence value is less than or equal to the pre-set threshold, return to step S1 to re-collect multi-modal data.

[0008] Further, in step S3, the constructing a multi-modal feature fusion model specifically comprises: S 31 : Standardize the body surface three-dimensional point cloud data, the body surface temperature distribution data, the body surface elastic modulus data, and the body surface micro-motion data; Convert the body surface three-dimensional point cloud data into a body surface curvature feature, the body surface temperature distribution data into a "high temperature zone" area mask, the body surface elastic modulus data into a stiffness distribution matrix, and the body surface micro-motion data into a dynamic offset vector; S32 : constructing an acupoint-meridian association graph through a graph convolution network, nodes of the acupoint-meridian association graph being acupoint candidate points, and edge weights being meridian connectivity and biological feature similarity between nodes; S 33 : adopting an attention mechanism to perform weighted fusion on the body surface three-dimensional point cloud data, the body surface temperature distribution data, the body surface elastic modulus data and the body surface micro-motion data; wherein initial weights of the body surface three-dimensional point cloud data, the body surface temperature distribution data, the body surface elastic modulus data and the body surface micro-motion data are 0.4, 0.25, 0.2 and 0.15 respectively; If the signal-to-noise ratio of a certain modal data is lower than a preset value, the weight of the certain modal data is reduced, and the weights of other modal data are increased.

[0009] Further, in step S4, the constructing a meridian-acupoint association relationship matrix comprises: S 41 : determining reference positions of 14 main meridians and 361 commonly used acupoints of a human body based on Chinese medicine classics, and establishing an acupoint-meridian correspondence table; S 42 : determining reference coordinates and deviation thresholds of each acupoint by collecting and counting coordinate distribution ranges of each acupoint in three-dimensional space through large sample data; S 43 : setting feature thresholds based on body surface biological features of each acupoint to form the meridian-acupoint association relationship matrix; wherein the body surface biological features include a temperature range of 36.5℃-37.8℃ and an elastic modulus range of 10kPa-50kPa.

[0010] Further, in step S5, the dynamic compensation of the candidate region specifically comprises: S 51 : based on the body surface micro-motion data, extracting frequency components corresponding to respiration and muscle micro-motion through Fourier transform, calculating displacement amplitudes of acupoints at different frequencies, and obtaining a dynamic offset Δ(x, y, z); S 52 : according to the body surface elastic modulus data, inversely calculating the subcutaneous tissue thickness at the acupoint through the Hertz contact theory, and correcting the depth coordinate z of the acupoint; The correction formula is: z 修正 = z 初始 + (E0-E 实测 )×k; wherein E0 is a standard elastic modulus, E 实测 is the elastic modulus of the current acupoint, and k is a correction coefficient; S 53The dynamic offset Δ(x,y,z) is superimposed with the corrected depth coordinate z to obtain the precise three-dimensional coordinates (x,y,z) of the acupoint to be located. 终 y 终 , z 终 ).

[0011] Furthermore, in step S6, calculating the confidence value of the precise three-dimensional coordinates specifically includes: S 61 Calculate the feature similarity S1 between the candidate region in the enhanced acupoint feature map and the reference acupoint in the meridian-acupoint association matrix; S 62 Calculate the standard deviation σ of the dynamic offset Δ(x,y,z) to obtain the dynamic stability coefficient S2 = 1 - σ / σ max ; Where: σ max The maximum permissible standard deviation of the offset; S 63 Calculate the deviation rate between the elastic modulus data and the reference elastic modulus, S3 = 1 - |E 实测 - E0| / E0; S 64 The confidence level value is calculated as 0.5 × S1 + 0.3 × S2 + 0.2 × S3, and the preset threshold is 0.8. If the confidence level is > 0.8, the location is considered reliable; if 0.6 ≤ confidence level ≤ 0.8, the location result is output and a message "Manual verification required" is displayed; if the confidence level is < 0.6, the location is considered to have failed, and the process returns to step S1.

[0012] Furthermore, it also includes step S7: according to the location type of the acupoint to be located, call the corresponding moxibustion parameter template, the moxibustion parameter template includes the recommended moxibustion temperature of 40℃~55℃, the moxibustion distance of 3cm~8cm and the moxibustion technique for the acupoint to be located; the location type is automatically identified by the body surface contour features of the three-dimensional point cloud data of the body surface, with an identification accuracy of ≥95%.

[0013] Furthermore, the moxibustion techniques include hovering, pecking, and rotating, with the hovering technique used for acupoints on the head, the rotating technique used for acupoints on the lower limbs, and the pecking technique used for acupoints on the chest, abdomen, and ribs.

[0014] The second objective of this invention is to provide an acupoint identification and positioning system, comprising: Multimodal data acquisition module: used to acquire multimodal data of the human body to be treated. The multimodal data includes three-dimensional point cloud data of the body surface acquired by a depth camera, body surface temperature distribution data acquired by an infrared thermal imager, body surface elastic modulus data acquired by a piezoelectric sensor, and body surface micro-motion data acquired by a millimeter-wave radar. Data preprocessing module: used to register the three-dimensional point cloud data of the body surface by improving the ICP algorithm, process the body surface temperature distribution data by Gaussian filtering and histogram equalization, remove noise from the body surface elastic modulus data by sliding window filtering, and extract the dynamic offset from the body surface micro-motion data by Kalman filtering; Multimodal fusion module: used to construct a multimodal feature fusion model, which performs weighted mapping feature-level fusion of the preprocessed three-dimensional point cloud data of the body surface, the temperature distribution data of the body surface, the elastic modulus data of the body surface, and the micro-motion data of the body surface to generate an enhanced acupoint feature map; Acupoint Candidate Region Determination Module: This module is used to determine acupoint candidate regions based on enhanced acupoint feature maps and a meridian-acupoint correlation matrix, and to construct a human meridian-acupoint correlation matrix. The human meridian-acupoint correlation matrix includes the topological relationship between each acupoint and its corresponding meridian, distance constraints between adjacent acupoints, and the body surface biometric threshold corresponding to the acupoint. Dynamic compensation module: used to dynamically compensate the candidate area, calculate the real-time offset of the acupoint based on the micro-motion data of the body surface, and correct the depth coordinates of the acupoint by combining the elastic modulus data of the body surface, so as to obtain the accurate three-dimensional coordinates of the acupoint to be located. Safety assessment module: Based on the precise three-dimensional coordinates, it identifies densely populated areas of blood vessels and nerves around acupoints using an improved Harris corner detection algorithm, and generates a safe moxibustion path; at the same time, it calculates the confidence value of the precise three-dimensional coordinates. Parameter adaptation module: used to call the corresponding moxibustion parameter template according to the location type of the acupoint to be located. The moxibustion parameter template includes the recommended moxibustion temperature of 40℃~55℃, the moxibustion distance of 3cm~8cm and the moxibustion technique for the acupoint to be located. The location type is automatically identified through the surface contour features of the three-dimensional point cloud data of the body surface.

[0015] A third objective of this invention is to provide a moxibustion robot, comprising: The robot body includes a six-degree-of-freedom robotic arm and an end-effector moxibustion device. The acupoint recognition and positioning system described above is communicatively connected to the robotic arm and is used to send precise three-dimensional coordinates of the acupoints and a safe moxibustion path to the robotic arm. The controller is used to receive the output data of the acupoint recognition and positioning system, control the robotic arm to move along the safe moxibustion path, and adjust the temperature, distance and technique of the end-effector according to the moxibustion parameter template. The human-computer interaction module is used to display the acupoint location results, confidence values ​​and moxibustion parameters, and to receive confirmation commands from the user.

[0016] A fourth objective of this invention is to provide a computer-readable storage medium storing computer instructions for causing the computer to execute the acupoint identification and positioning method as described above.

[0017] Compared with the prior art, the present invention has at least the following beneficial effects: By integrating multimodal data (3D point cloud, temperature, elasticity, and micro-motion) and dynamic compensation, the static positioning error is ≤0.3mm, and the dynamic positioning error is ≤0.5mm. The positioning success rate for users with special body types is improved to over 95%, solving the problem of low accuracy in existing technologies. Based on millimeter-wave radar, the offset is calculated in real time using micro-motion data. Combined with elastic modulus correction of depth coordinates, it can compensate for acupoint offset caused by breathing and muscle micro-movements, reducing the "off-target" rate during moxibustion to below 3%, solving the problem of weak adaptability to dynamic interference. Through a unified meridian-acupoint association matrix and automatic site recognition, it can cover the positioning of all acupoints in the head, chest, abdomen, back, upper limbs, and lower limbs, without the need to develop separate algorithms for different sites. Hardware compatibility is improved to over 90%, solving the problem of poor universality across multiple sites. Through vascular bifurcation point detection and confidence assessment, it can avoid areas with dense blood vessels and nerves, and only output positioning results with a confidence level ≥0.8, reducing the safety risk of moxibustion to below 0.1%, solving the problem of high safety risk. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the acupoint identification and positioning method in an embodiment of the present invention; Figure 2 This is a schematic diagram of the specific process of step S3 in an embodiment of the present invention; Figure 3 This is a schematic diagram of the specific process of step S4 in an embodiment of the present invention; Figure 4 This is a schematic diagram of the specific process of step S5 in an embodiment of the present invention; Figure 5 This is a schematic diagram of the specific process of step S6 in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structural framework of the acupoint identification and positioning system in an embodiment of the present invention; Figure 7 This is a three-dimensional structural diagram of the moxibustion robot in an embodiment of the present invention.

[0019] Explanation of reference numerals in the attached figures: 10-Robot body; 11-Six-DOF robotic arm; 12-End-effector moxibustion device; 20-Acupoint Recognition and Positioning System; 30-Controller; 40 - Human-Computer Interaction Module; 100 - Multimodal data acquisition module; 200 - Data preprocessing module; 300 - Multimodal fusion module; 400 - Acupoint candidate region determination module; 50 - Dynamic compensation module; 600 - Safety assessment module; 700 - Parameter adaptation module. Detailed Implementation

[0020] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In the description of this invention, it should be noted that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can also refer to the internal connection of two components; and they can refer to a wireless connection or a wired connection. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0021] Please see Figure 1 As shown, this embodiment of the invention provides a method for identifying and locating acupoints, which includes the following steps: S1: Acquire multimodal data of the human body to be treated. The multimodal data includes three-dimensional point cloud data of the body surface collected by a depth camera, body surface temperature distribution data collected by an infrared thermal imager, body surface elastic modulus data collected by a piezoelectric sensor, and body surface micro-motion data collected by a millimeter-wave radar.

[0022] In this step, 3D point cloud data of the body surface is used to provide the spatial morphological benchmark for acupoints, solving the problem of "where is it?"; temperature data corresponds to the temperature anomalies at acupoints in Traditional Chinese Medicine; elastic modulus data corresponds to the unique texture of acupoints; and micro-motion data of the body surface is used to provide information on the pulse movement of acupoints. By collecting data from both morphological and functional dimensions, covering the physical and physiological attributes of acupoints, precise input is provided for the next step.

[0023] S2: Preprocess the multimodal data, wherein the preprocessing includes: achieving three-dimensional point cloud registration by improving the ICP algorithm (introducing normal vector constraints and feature descriptor matching), processing the body surface temperature distribution data by Gaussian filtering and histogram equalization, removing noise from the body surface elastic modulus data by sliding window filtering, and extracting the dynamic offset from the body surface micro-motion data by Kalman filtering.

[0024] In this step, the 3D point cloud is solved by using an improved ICP algorithm through normal vector and feature matching to address the point cloud misalignment problem caused by irregularities on the human body surface, thereby improving the accuracy of the spatial reference. The temperature data is processed using Gaussian filtering and histogram equalization to remove ambient temperature interference and enhance the temperature difference between acupoints and surrounding skin, highlighting the features. The elastic modulus data is filtered using a sliding window to remove random noise from the pressure sensor. The micro-motion data of the body surface is processed using Kalman filtering to extract regular micro-motions caused by breathing / pulse, in order to retain effective signals.

[0025] It should be noted that the improved ICP algorithm for 3D point clouds solves the adaptability problem of traditional ICP in human body surface point cloud registration by introducing "normal vector constraints" and "feature descriptor matching". Specifically, it can be broken down into three key steps: preprocessing, feature enhancement and matching, and iterative optimization. Each step has been specifically optimized for the special characteristics of human body surface point clouds (such as non-rigidity, high noise, and complex features). I. Point Cloud Preprocessing: Before performing the ICP iteration, the original 3D point cloud of the body surface acquired by the depth camera (including the target point cloud "human standard template point cloud" and the source point cloud "real-time point cloud of the human body to be treated") needs to be preprocessed to reduce noise and redundant information: II. Feature-Enhanced Matching: Traditional ICP only selects matching point pairs based on "closest spatial distance," which is prone to mismatches due to similar local morphological features on the body surface (such as skin folds in different areas); the improved ICP optimizes matching point pairs through a dual mechanism of feature descriptor pre-matching and normal vector constraint selection: III. Iterative Optimization: After obtaining high-quality matching point pairs, the ICP is improved by optimizing the iterative process to further enhance the registration accuracy (minimizing the spatial error between the real-time point cloud and the standard template point cloud).

[0026] S3: Construct a multimodal feature fusion model, which performs weighted mapping feature-level fusion on preprocessed three-dimensional point cloud data of the body surface, body surface temperature distribution data, body surface elastic modulus data and body surface micro-motion data to generate enhanced acupoint feature maps. The weights are dynamically adjusted using a hidden Markov model, with higher weights assigned to modal data that contribute significantly to acupoint features.

[0027] In this step, for example, when the elasticity difference of certain acupoints (such as Hegu acupoint) is more obvious, the model will automatically increase the weight of the elastic modulus data; while the temperature difference of certain acupoints (such as Neiguan acupoint) is more prominent, the weight will be tilted towards the temperature data, and the final "enhanced acupoint feature map" can more clearly locate the acupoints.

[0028] S4: Based on the enhanced acupoint feature map, combined with the preset human meridian-acupoint association matrix, the candidate region of the acupoint to be located is initially determined; the meridian-acupoint association matrix is ​​constructed based on the theory of meridians in traditional Chinese medicine, and includes the topological relationship between each acupoint and its corresponding meridian, the distance constraint between adjacent acupoints, and the body surface bio-feature threshold corresponding to the acupoint.

[0029] In this step, the meridian-acupoint matrix is ​​used for preliminary localization: based on the constraints of traditional Chinese medicine theory (such as the distance between acupoints and the meridians they belong to), "misjudgment" caused by similar body surface features is avoided, and 1-2 candidate areas are quickly selected.

[0030] S5: Perform dynamic compensation on the candidate region, calculate the real-time offset of the acupoint based on the micro-motion data of the body surface, and correct the depth coordinates of the acupoint by combining the elastic modulus data of the body surface, so as to obtain the accurate three-dimensional coordinates of the acupoint to be located.

[0031] In this step, the micro-movements of the human body (such as skin displacement caused by breathing) and the depth of pressure (the elastic modulus data of the body surface reflects the tissue hardness and corrects the true depth of the acupoint under the skin) are taken into account, and the two-dimensional planar positioning is upgraded to three-dimensional precise coordinates.

[0032] S6: Based on the precise three-dimensional coordinates, the dense areas of blood vessels and nerves around the acupoint are identified by the blood vessel bifurcation point detection algorithm (improved Harris corner detection) to generate a safe moxibustion path; at the same time, the confidence value of the precise three-dimensional coordinates is calculated. If the confidence value is greater than a preset threshold, the precise three-dimensional coordinates and the safe moxibustion path are output; if it is less than or equal to the preset threshold, the process returns to step S1 to re-collect multimodal data.

[0033] In this step, Harris corner detection is improved to identify areas with dense blood vessels and nerves, and a moxibustion path is planned to avoid dangerous areas, thus reducing operational risks. Confidence values ​​(such as feature matching degree and data consistency) are calculated. If the confidence value is lower than the threshold, data is collected again to avoid positioning errors caused by data anomalies, thus forming a closed loop in the process.

[0034] Compared to the traditional single-modal approach relying on visual observation and finger pressure, this embodiment integrates four types of data to verify acupoints from multiple dimensions, reducing misjudgments based on single features. In step S5, micro-motion offset calculation and depth coordinate correction address static positioning errors caused by the non-rigid nature of the human body (such as acupoint position shifts due to breathing and posture changes). The meridian-acupoint matrix in step S4 provides "prior knowledge" for positioning, avoiding the decrease in accuracy caused by unconstrained search. The blood vessel bifurcation point detection in step S6 directly avoids areas with dense blood vessels and nerves, addressing the concern of "burning blood vessels" in traditional moxibustion. A threshold judgment mechanism automatically eliminates low-quality results, preventing "outputting with defects" and ensuring the reliability of each positioning result. Multiple filtering algorithms in step S2 reduce environmental interference (such as temperature fluctuations and sensor noise), ensuring stable input data.

[0035] For more details, please refer to Figure 2 As shown, in one embodiment of the present invention, step S3, the construction of the multimodal feature fusion model specifically includes: Step S 31 The system standardizes the three-dimensional point cloud data, temperature distribution data, elastic modulus data, and micro-motion data of the body surface. It converts the three-dimensional point cloud data into surface curvature features, the temperature distribution data into a "high temperature zone" region mask, the elastic modulus data into a stiffness distribution matrix, and the micro-motion data into a dynamic offset vector.

[0036] Considering the significant differences in the original data format and dimensions (e.g., point clouds are sets of spatial coordinates, while temperature is a numerical distribution), direct fusion is not feasible. By transforming the data (e.g., point clouds to curvature, temperature to a "high-temperature zone" mask), the core information of the data is preserved while allowing all data to enter the same analytical dimension. This allows the original four types of body surface data (3D point cloud, temperature, elastic modulus, and micromotion) to be converted into structured features.

[0037] Step S 32 A graph of acupoint-meridian association is constructed using a graph convolutional network (GCN). The nodes of the association graph are candidate acupoints, and the edge weights are the meridian connectivity and biometric similarity between nodes.

[0038] Because acupoints and meridians are physiologically related, this step does not analyze individual acupoints in isolation. Instead, it uses a "node-edge" graph structure to reconstruct the natural relationships between acupoints. At the same time, the edge weights combine meridians (anatomical relationships) and biological features (data relationships), so that the relationship graph has both physiological basis and fits the actual data characteristics collected. Subsequent fusion can more accurately focus on the acupoint-related areas.

[0039] By using a graph convolutional network (GCN) to treat acupoint candidate points as nodes, and defining edge weights using meridian connectivity and biometric similarity, a correlation graph is formed.

[0040] Step S 33 An attention mechanism is used to perform weighted fusion of three-dimensional point cloud data, body surface temperature distribution data, body surface elastic modulus data, and body surface micro-motion data.

[0041] The initial weights for the 3D point cloud data of the body surface are 0.4, the initial weights for the temperature distribution data of the body surface are 0.25, the initial weights for the elastic modulus data of the body surface are 0.2, and the initial weights for the micro-motion data of the body surface are 0.15. The weights are adjusted in real time using a hidden Markov model. If the signal-to-noise ratio of a certain modality is lower than the preset value, its weight is reduced, while the weights of other modalities are increased to ensure the stability of the fused features.

[0042] Because the three-dimensional point cloud of the body surface can directly reflect the spatial location of acupoints and is the basis for localization, it has the highest weight; micro-motion data is more likely to be disturbed (such as breathing, slight movements), so it has the lowest weight. When the signal-to-noise ratio of a certain type of data is low (such as temperature data being disturbed by the environment), reducing its weight can prevent "bad data" from lowering the fusion effect; at the same time, increasing the weight of other reliable data ensures that the fusion features are always based on high-quality data and maintains stability.

[0043] The four types of data reflect the acupoint's "spatial morphology" (3D point cloud), "metabolic activity" (temperature), "tissue stiffness" (elastic modulus), and "dynamic function" (micro-movement). After fusion, the acupoint's characteristics can be completely depicted from multiple dimensions, avoiding the one-sidedness of single data (such as the inability to judge the metabolic state of the acupoint using only point cloud).

[0044] The acupoint-meridian association map constructed using GCN can filter out interfering data from non-acupoint areas, allowing the fusion process to focus more on the target area and further improving the specificity of features. When a certain type of data is disturbed (such as ambient temperature affecting body surface temperature data), the weight is automatically reduced, preventing the overall fusion result from "failing," thus solving the difficulty of "unreliable single modality leading to overall error" in multimodal data.

[0045] For more details, please refer to Figure 3 As shown, in one embodiment of the present invention, step S4 of constructing the meridian-acupoint correlation matrix specifically includes: Step S 41 Based on classic Chinese medicine texts such as the "Huangdi Neijing" and "Zhenjiu Juying", the baseline locations of 14 major meridians and 361 commonly used acupoints in the human body were determined, and a correspondence table between acupoints and meridians was established.

[0046] Step S 42By collecting large sample data (sample size ≥ 5000 cases, covering different ages, genders and body types), the coordinate distribution range of each acupoint in three-dimensional space was statistically analyzed, and the baseline coordinates and deviation threshold of each acupoint were determined.

[0047] Considering that the acupoint locations described in classical texts are mostly based on "standard human figures," while individual differences (such as height and weight) can lead to objective deviations in acupoint locations, large-sample statistical analysis can obtain acupoint coordinate benchmarks that are closer to those of real people, and also clarify the range of deviations (e.g., fluctuations within ±5mm of a certain acupoint at the standard location are considered reasonable), thus solving the problem of mismatch between theoretical coordinates and actual human figures.

[0048] Step S 43 Based on the surface biological characteristics of each acupoint (such as temperature range of 36.5℃~37.8℃, elastic modulus range of 10kPa~50kPa), feature thresholds are set to form the meridian-acupoint correlation matrix. The matrix also includes distance constraints between adjacent acupoints (such as the longitudinal distance between adjacent acupoints on the Ren meridian being 1 to 2 horizontal inches), which is used to help verify the rationality of the acupoint location.

[0049] In this step, the temperature, elasticity, and other characteristics of acupoints differ from those of surrounding tissues. Quantifying these differences into thresholds allows the matrix to include not only location information but also functional characteristics, which is more in line with the physiological nature of acupoints. Acupoints on meridians have a fixed spatial arrangement pattern (such as the vertical arrangement of acupoints on the Ren meridian). After adding distance constraints, if the coordinates of an acupoint exceed the distance range of adjacent acupoints, it can be directly determined that its location may be incorrect, providing a basis for error correction for subsequent acupoint identification.

[0050] Therefore, this technical solution constructs a meridian-acupoint correlation matrix that combines traditional Chinese medicine theory with practical application adaptability through a three-step process of setting benchmarks in classical texts, compensating for data deviations, and building a matrix based on features. This transforms the traditional meridian and acupoint system from an abstract description into a quantifiable and verifiable digital model.

[0051] For more details, please refer to Figure 4 As shown, in one embodiment of the present invention, dynamic compensation of the candidate region specifically includes: Step S 51 Based on the micro-motion data of the body surface, the frequency components corresponding to breathing and muscle micro-motion are extracted by Fourier transform, and the displacement amplitude of acupoints at different frequencies is calculated to obtain the dynamic offset Δ(x,y,z).

[0052] Because the human body is not static, for example, breathing causes the chest and abdomen to rise and fall, and muscle micro-movements cause minute displacements on the body surface. These factors cause the (x, y, z) coordinates of acupoints to change dynamically over time. Fourier transform can accurately separate the micro-movement frequencies from different sources (breathing, muscles) (e.g., the breathing frequency is about 0.2-0.3 Hz, and the muscle micro-movement frequency is higher), and then calculate the corresponding displacements. This provides a dynamic interference quantification value for subsequent compensation, avoiding the use of static coordinates to locate dynamic acupoints.

[0053] Step S 52 Based on the elastic modulus data, the thickness of the subcutaneous tissue at the acupoint is inferred using Hertz contact theory, and the depth coordinate z of the acupoint is corrected. The corrected formula is: z 修正 = z 初始 +(E0-E 实测 )×k Where: E0 is the standard elastic modulus (taken as 30 kPa); E 实测 is the elastic modulus of the current acupoint; k is the correction coefficient (value 0.02 mm / kPa).

[0054] Because subcutaneous tissue thickness directly affects the depth of acupoints (z-coordinate): people with thicker fat layers may have deeper acupoints, while those with thinner fat layers may have shallower ones; and because the elastic modulus E is related to tissue thickness. The thicker the tissue (e.g., a thicker fat layer), the lower the elastic modulus (E). 实测 Smaller), and vice versa (E) 实测 (Large), which provides a physical basis for inferring the thickness through E.

[0055] In this embodiment, 30 kPa is used as the standard elastic modulus (E0). When E 实测 When <E0, (E0-E 实测 z is a positive value. 修正 It will be more than z 初始 The larger (i.e., deeper acupoints) aligns with the actual situation that thicker tissue corresponds to deeper acupoints; the correction coefficient k (0.02mm / kPa) transforms the difference in elastic modulus into a specific depth correction value, making the correction results quantifiable and repeatable.

[0056] Step S 53 The dynamic offset Δ(x,y,z) is superimposed with the corrected depth coordinate z to obtain the precise three-dimensional coordinates (x,y,z) of the acupoint to be located. 终 y 终 , z 终 ).

[0057] Due to step S 51 This addresses the dynamic real-time offset problem of (x, y, z) (e.g., breathing causes horizontal offset of x and y, and vertical offset of z); step S 52This addresses the static deviation problem of individual tissue differences in the z-coordinate, with both methods covering all localization interferences. By superimposing the two, it is equivalent to simultaneously eliminating two types of core interferences, and the final coordinates can accurately match the "real-time actual location" of the acupoint.

[0058] For more details, please refer to Figure 5 As shown, in one embodiment of the present invention, the calculation of the confidence value in step S6 specifically includes: Step S 61 Calculate the feature similarity S1 (value range 0 to 1) between the candidate region in the enhanced acupoint feature map and the reference acupoint in the meridian-acupoint association matrix.

[0059] The essence of acupoint localization is to find regions that match the characteristics of benchmark acupoints, with feature similarity being the core criterion. The closer S1 is to 1, the more closely the multimodal features (such as temperature, elasticity, and three-dimensional morphology) of the candidate region match the benchmark acupoint, and the more accurate the localization. Using this as the primary evaluation indicator aligns with the core logic of acupoint localization.

[0060] Step S 62 Calculate the standard deviation σ of the dynamic offset Δ(x,y,z) to obtain the dynamic stability coefficient S2 = 1 - σ / σ max , where σ max The maximum permissible standard deviation of offset is 0.5 mm.

[0061] Micro-movements in the human body can cause dynamic changes in the location of acupoints, but the magnitude of these changes must be within a reasonable range to be reliable. σ reflects the degree of fluctuation in the offset; a larger σ indicates a more unstable location (e.g., deep breathing causing large offset fluctuations). max (0.5mm) is a pre-set "credible fluctuation upper limit". When σ≤0.5mm, S2≥0, and the smaller σ is, the closer S2 is to 1, which intuitively reflects the stability of the position. Adding the dynamic stability coefficient S2 can exclude unreliable cases where the feature matches but the position continues to fluctuate (such as candidate region feature images that are always moving, which may be interference areas).

[0062] Step S 63 Calculate the deviation rate between the elastic modulus data and the reference elastic modulus, S3 = 1 - |E 实测 - E0| / E0.

[0063] Because the elastic modulus of acupoints is an important physiological characteristic, and there is a relatively stable baseline range (E0=30kPa) in healthy individuals. If E 实测If the deviation from E0 is too large (e.g., far exceeding the normal range of 10-50 kPa), even if the characteristics and location seem to match, it may be abnormal tissue (such as nodules or inflammatory areas). The closer S3 is to 1, the more consistent the physiological characteristics are with the baseline, the stronger the "physiological rationality" of the location, and the less likely it is to be misidentified as acupoint.

[0064] Step S 64 The confidence value is calculated as 0.5 × S1 + 0.3 × S2 + 0.2 × S3, and the preset threshold is 0.8. If the confidence value is greater than 0.8, the positioning is considered reliable. If the confidence value is less than 0.6, the positioning result is output and a message "Manual verification required" is displayed. If the confidence value is less than 0.6, the positioning is considered to have failed, and the process returns to step S1.

[0065] Traditional positioning often relies on a single indicator (such as whether the location is within the coordinate range), which can easily lead to misjudgments such as "similar features but unstable location" or "correct location but physiological abnormalities." This new solution comprehensively evaluates from three dimensions: "features, location, and physiology." Only when all three indicators meet the criteria can a high confidence level be obtained, which is equivalent to setting up a triple check. This effectively filters out situations where a single dimension meets the criteria but the overall system is unreliable, significantly reducing the misjudgment rate.

[0066] More specifically, in one embodiment of the present invention, the acupoint identification and positioning method further includes step S7: Based on the location type of the acupoint to be located (head, chest, abdomen, back, upper limb, lower limb), the corresponding moxibustion parameter template is called. The moxibustion parameter template includes the recommended moxibustion temperature (40℃~55℃), moxibustion distance (3cm~8cm), and moxibustion technique (suspending, pecking, circling) for that acupoint. The location type is automatically identified through the surface contour features of the three-dimensional point cloud data, with an accuracy rate of ≥95%.

[0067] Specifically, in this embodiment, the surface contours of different parts of the human body vary significantly (e.g., the head is spherical, the back has a spinal protrusion, and the limbs are elongated). Three-dimensional point cloud data of the body surface can completely record these contour features (e.g., curvature, size, and shape). Through contour feature matching (e.g., comparing the collected contours with a preset "head contour model"), the location of acupoints can be accurately determined without manual annotation, providing a basis for subsequent automatic retrieval of parameter templates. Furthermore, an accuracy rate of ≥95% ensures the reliability of the judgment results.

[0068] The skin thickness, sensitivity, and subcutaneous tissue distribution vary greatly in different areas, so the requirements for moxibustion parameters will inevitably differ.

[0069] For example, the skin on the scalp is thin and highly sensitive, requiring a lower temperature (e.g., 40℃~45℃) and a slightly greater distance (e.g., 6cm~8cm), with a gentle (suspending) technique. The skin on the back is thicker and more tolerant, allowing for a slightly higher temperature (e.g., 48℃~55℃) and a shorter distance (e.g., 3cm~5cm), with a circular motion technique to cover a larger area. Customized parameter templates for different body parts perfectly match their physiological characteristics, avoiding the pitfalls of applying the same set of parameters to all areas.

[0070] Please see Figure 6 As shown, another embodiment of the present invention also provides an acupoint identification and positioning system, the acupoint identification and positioning system comprising: Multimodal data acquisition module 100: used to acquire multimodal data of the human body to be treated, including three-dimensional point cloud data of the body surface acquired by a depth camera, body surface temperature distribution data acquired by an infrared thermal imager, body surface elastic modulus data acquired by a piezoelectric sensor, and body surface micro-motion data acquired by a millimeter-wave radar; wherein the scanning resolution of the depth camera is ≥0.05mm, the temperature accuracy of the infrared thermal imager is ±0.1℃, the detection range of the piezoelectric sensor array is 1kPa~10MPa, and the micro-motion detection accuracy of the millimeter-wave radar is ±0.1mm; Data preprocessing module 200: used to register the three-dimensional point cloud data of the body surface by improving the ICP algorithm, process the body surface temperature distribution data by Gaussian filtering and histogram equalization, remove noise in the body surface elastic modulus data by sliding window filtering, and extract the dynamic offset in the body surface micro-motion data by Kalman filtering. Multimodal fusion module 300: used to construct a multimodal feature fusion model, which performs weighted mapping feature-level fusion on the preprocessed three-dimensional point cloud data of the body surface, the temperature distribution data of the body surface, the elastic modulus data of the body surface, and the micro-motion data of the body surface to generate an enhanced acupoint feature map; Acupoint candidate region determination module 400: used to determine acupoint candidate regions based on enhanced acupoint feature maps and meridian-acupoint association matrix, and construct human meridian-acupoint association matrix; the human meridian-acupoint association matrix includes the topological relationship between each acupoint and its corresponding meridian, the distance constraint between adjacent acupoints, and the body surface bio-feature threshold corresponding to the acupoint. Dynamic compensation module 500: used to dynamically compensate the candidate area, calculate the real-time offset of the acupoint based on the micro-motion data of the body surface, and correct the depth coordinates of the acupoint by combining the elastic modulus data of the body surface, so as to obtain the accurate three-dimensional coordinates of the acupoint to be located. Safety assessment module 600: Based on the precise three-dimensional coordinates, it identifies densely populated areas of blood vessels and nerves around acupoints using an improved Harris corner detection algorithm, and generates a safe moxibustion path; at the same time, it calculates the confidence value of the precise three-dimensional coordinates. Parameter adaptation module 700: used to call the corresponding moxibustion parameter template according to the location type of the acupoint to be located. The moxibustion parameter template includes the recommended moxibustion temperature of 40℃~55℃, the moxibustion distance of 3cm~8cm and the moxibustion technique for the acupoint to be located. The location type is automatically identified through the surface contour features of the three-dimensional point cloud data of the body surface.

[0071] Traditional positioning relies on human experience (such as finger measurement and anatomical landmark judgment), resulting in large errors (often reaching millimeter or even centimeter levels). In contrast, this system achieves sub-millimeter positioning accuracy through high-precision hardware acquisition (such as a depth camera with 0.05mm resolution), multimodal fusion (enhancing feature differences), and dynamic compensation (eliminating interference). It can also track the dynamic position of acupoints in real time, completely solving the problems of static, subjective, and error-prone traditional positioning methods.

[0072] Furthermore, in another embodiment of the present invention, the multimodal data acquisition module 100 further includes a calibration unit, which is used to periodically perform accuracy calibration on the depth camera, infrared thermal imager, piezoelectric sensor and millimeter-wave radar, including: acquiring calibration data through a standard phantom (containing acupoint simulation points with known coordinates), calculating the measurement error of each device, generating an error compensation matrix, ensuring the consistency of multimodal data in the same coordinate system, and the overall positioning error after calibration is ≤0.3mm.

[0073] like Figure 7 As shown, another embodiment of the present invention also provides a moxibustion robot, the moxibustion robot comprising: The robot body 10 includes a six-degree-of-freedom robotic arm 11 and an end-effector 12. The acupoint recognition and positioning system 20 is connected to the six-degree-of-freedom robotic arm 11 and is used to send the precise three-dimensional coordinates of the acupoints and the safe moxibustion path to the six-degree-of-freedom robotic arm 11. The controller 30 is used to receive the output data of the acupoint recognition and positioning system 20, control the six-degree-of-freedom robotic arm 11 to move along the safe moxibustion path, and adjust the temperature, distance and technique of the end-effector moxibustion device according to the moxibustion parameter template. The human-computer interaction module 40 is used to display the acupoint location results, confidence values ​​and moxibustion parameters, and to receive confirmation commands from the user.

[0074] In this embodiment, after the moxibustion robot is started, the acupoint recognition and positioning system 20 collects multimodal data (point cloud, temperature, etc.), completes acupoint positioning, and generates precise coordinates, a safe path, and a moxibustion parameter template. The positioning system sends the data to the controller 30, and at the same time displays key information on the human-computer interaction module 40. After the user confirms the information, the controller 30 parses the data, sends path control instructions to the six-degree-of-freedom robotic arm 11, and sends parameter control instructions to the end-effector 12. The robotic arm moves the end-effector to the acupoint coordinates according to the safe path, and the end-effector performs moxibustion according to the parameters. During the moxibustion process, the interaction module displays the status in real time (such as "Moxibustion in progress, temperature 43℃"), and the user can pause or adjust at any time.

[0075] Accordingly, embodiments of the present invention also provide a computer-readable storage medium for storing computer-readable programs or instructions. When the programs or instructions are executed by a processor, they can implement the acupoint identification and positioning method steps or functions provided in the above-described method embodiments.

[0076] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0077] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of this disclosure, and all such changes and modifications will fall within the scope of protection of this invention.

Claims

1. A method for identifying and locating acupoints, characterized in that, Includes the following steps: S1: Acquire multimodal data of the human body to be treated. The multimodal data includes three-dimensional point cloud data of the body surface collected by a depth camera, body surface temperature distribution data collected by an infrared thermal imager, body surface elastic modulus data collected by a piezoelectric sensor, and body surface micro-motion data collected by a millimeter-wave radar. S2: Preprocess the multimodal data, the preprocessing including: registering the three-dimensional point cloud data of the body surface by improving the ICP algorithm, processing the body surface temperature distribution data by Gaussian filtering and histogram equalization, removing noise from the body surface elastic modulus data by sliding window filtering, and extracting the dynamic offset from the body surface micro-motion data by Kalman filtering. S3: Construct a multimodal feature fusion model, and perform weighted mapping feature-level fusion on the preprocessed three-dimensional point cloud data of the body surface, the temperature distribution data of the body surface, the elastic modulus data of the body surface, and the micro-motion data of the body surface to generate an enhanced acupoint feature map; The weights are dynamically adjusted using a hidden Markov model. S4: Based on the enhanced acupoint feature map, combined with the preset human meridian-acupoint association matrix, the candidate region of the acupoint to be located is initially determined, and the human meridian-acupoint association matrix is ​​constructed; the human meridian-acupoint association matrix includes the topological relationship between each acupoint and its corresponding meridian, the distance constraint between adjacent acupoints, and the body surface bio-feature threshold corresponding to the acupoint. S5: Perform dynamic compensation on the candidate region, calculate the real-time offset of the acupoint based on the micro-motion data of the body surface, and correct the depth coordinates of the acupoint by combining the elastic modulus data of the body surface to obtain the accurate three-dimensional coordinates of the acupoint to be located. S6: Based on the precise three-dimensional coordinates, the improved Harris corner detection algorithm is used to identify densely populated areas of blood vessels and nerves around the acupoints to generate a safe moxibustion path; at the same time, the confidence value of the precise three-dimensional coordinates is calculated. If the confidence value is greater than the preset threshold, the accurate three-dimensional coordinates and safe moxibustion path are output; if it is less than or equal to the preset threshold, the process returns to step S1 to re-acquire multimodal data.

2. The acupoint identification and positioning method according to claim 1, characterized in that, In step S3, the construction of the multimodal feature fusion model specifically includes: S 31 The three-dimensional point cloud data of the body surface, the temperature distribution data of the body surface, the elastic modulus data of the body surface, and the micro-motion data of the body surface are standardized. The three-dimensional point cloud data of the body surface is converted into body surface curvature features, the body surface temperature distribution data is converted into a "high temperature zone" region mask, the body surface elastic modulus data is converted into a stiffness distribution matrix, and the body surface micro-motion data is converted into a dynamic offset vector. S 32 A graph convolutional network is used to construct an acupoint-meridian association graph, where the nodes of the acupoint-meridian association graph are candidate acupoints, and the edge weights are the meridian connectivity and biometric similarity between nodes. S 33 An attention mechanism is used to perform weighted fusion of the three-dimensional point cloud data of the body surface, the temperature distribution data of the body surface, the elastic modulus data of the body surface, and the micro-motion data of the body surface. The initial weights of the three-dimensional point cloud data of the body surface, the temperature distribution data of the body surface, the elastic modulus data of the body surface, and the micro-motion data of the body surface are 0.4, 0.25, 0.2, and 0.15, respectively. If the signal-to-noise ratio of a certain modality is lower than the preset value, its weight is reduced, while the weights of other modalities are increased.

3. The acupoint identification and positioning method according to claim 1, characterized in that, In step S4, constructing the meridian-acupoint correlation matrix includes: S 41 Based on traditional Chinese medicine classics, the baseline locations of 14 major meridians and 361 commonly used acupoints in the human body were determined, and a correspondence table between acupoints and meridians was established. S 42 By collecting and statistically analyzing large sample data, the coordinate distribution range of each acupoint in three-dimensional space is determined, and the reference coordinates and deviation threshold of each acupoint are determined. S 43 Based on the surface bio-characteristics of each acupoint, feature thresholds are set to form the meridian-acupoint correlation matrix; The biological characteristics of the body surface include a temperature range of 36.5℃ to 37.8℃ and an elastic modulus range of 10kPa to 50kPa.

4. The acupoint identification and positioning method according to claim 1, characterized in that, In step S5, the dynamic compensation of the candidate region specifically includes: S 51 Based on the micro-motion data of the body surface, the frequency components corresponding to breathing and muscle micro-motion are extracted by Fourier transform, and the displacement amplitude of acupoints at different frequencies is calculated to obtain the dynamic offset Δ(x,y,z). S 52 Based on the surface elastic modulus data, the subcutaneous tissue thickness at the acupoint is inferred using Hertz contact theory, and the depth coordinate z of the acupoint is corrected. The corrected formula is: z 修正 = z 初始 +(E0-E 实测 )×k; Where: E0 is the standard elastic modulus; E 实测 is the elastic modulus of the current acupoint; k is the correction coefficient; S 53 The dynamic offset Δ(x,y,z) is superimposed with the corrected depth coordinate z to obtain the precise three-dimensional coordinates (x,y,z) of the acupoint to be located. 终 y 终 , z 终 ).

5. The acupoint identification and positioning method according to claim 1, characterized in that, In step S6, calculating the confidence value of the precise three-dimensional coordinates specifically includes: S 61 Calculate the feature similarity S1 between the candidate region in the enhanced acupoint feature map and the reference acupoint in the meridian-acupoint association matrix; S 62 Calculate the standard deviation σ of the dynamic offset Δ(x,y,z) to obtain the dynamic stability coefficient S2 = 1 - σ / σ max ; Where: σ max The maximum permissible standard deviation of the offset; S 63 Calculate the deviation rate between the elastic modulus data and the reference elastic modulus, S3 = 1 - |E 实测 - E0| / E0; S 64 The confidence level value is calculated as 0.5 × S1 + 0.3 × S2 + 0.2 × S3, and the preset threshold is 0.

8. If the confidence level is > 0.8, the location is considered reliable; if 0.6 ≤ confidence level ≤ 0.8, the location result is output and a message "Manual verification required" is displayed; if the confidence level is < 0.6, the location is considered to have failed, and the process returns to step S1.

6. The acupoint identification and positioning method according to claim 1, characterized in that, The method also includes step S7: according to the location type of the acupoint to be located, the corresponding moxibustion parameter template is called. The moxibustion parameter template includes the recommended moxibustion temperature of 40℃~55℃, the moxibustion distance of 3cm~8cm and the moxibustion technique for the acupoint to be located. The location type is automatically identified by the surface contour features of the three-dimensional point cloud data of the body surface, with an accuracy rate of ≥95%.

7. The acupoint identification and positioning method according to claim 6, characterized in that, The moxibustion techniques include hovering, pecking, and rotating. The hovering technique is used for acupoints on the head, the rotating technique is used for acupoints on the lower limbs, and the pecking technique is used for acupoints on the chest, abdomen, and ribs.

8. An acupoint identification and positioning system, characterized in that, include: Multimodal data acquisition module: used to acquire multimodal data of the human body to be treated. The multimodal data includes three-dimensional point cloud data of the body surface acquired by a depth camera, body surface temperature distribution data acquired by an infrared thermal imager, body surface elastic modulus data acquired by a piezoelectric sensor, and body surface micro-motion data acquired by a millimeter-wave radar. Data preprocessing module: used to register the three-dimensional point cloud data of the body surface by improving the ICP algorithm, process the body surface temperature distribution data by Gaussian filtering and histogram equalization, remove noise from the body surface elastic modulus data by sliding window filtering, and extract the dynamic offset from the body surface micro-motion data by Kalman filtering; Multimodal fusion module: used to construct a multimodal feature fusion model, which performs weighted mapping feature-level fusion of the preprocessed three-dimensional point cloud data of the body surface, the temperature distribution data of the body surface, the elastic modulus data of the body surface, and the micro-motion data of the body surface to generate an enhanced acupoint feature map; Acupoint Candidate Region Determination Module: This module is used to determine acupoint candidate regions based on enhanced acupoint feature maps and a meridian-acupoint correlation matrix, and to construct a human meridian-acupoint correlation matrix. The human meridian-acupoint correlation matrix includes the topological relationship between each acupoint and its corresponding meridian, distance constraints between adjacent acupoints, and the body surface biometric threshold corresponding to the acupoint. Dynamic compensation module: used to dynamically compensate the candidate area, calculate the real-time offset of the acupoint based on the micro-motion data of the body surface, and correct the depth coordinates of the acupoint by combining the elastic modulus data of the body surface, so as to obtain the accurate three-dimensional coordinates of the acupoint to be located. Safety assessment module: Based on the precise three-dimensional coordinates, it identifies densely populated areas of blood vessels and nerves around acupoints using an improved Harris corner detection algorithm, and generates a safe moxibustion path; at the same time, it calculates the confidence value of the precise three-dimensional coordinates. Parameter adaptation module: used to call the corresponding moxibustion parameter template according to the location type of the acupoint to be located. The moxibustion parameter template includes the recommended moxibustion temperature of 40℃~55℃, the moxibustion distance of 3cm~8cm and the moxibustion technique for the acupoint to be located. The location type is automatically identified through the surface contour features of the three-dimensional point cloud data of the body surface.

9. A moxibustion robot, characterized in that, include: The robot body includes a six-degree-of-freedom robotic arm and an end-effector moxibustion device. The acupoint recognition and positioning system as described in claim 8, wherein the acupoint recognition and positioning system is communicatively connected to the robotic arm and is used to send the precise three-dimensional coordinates of the acupoints and the safe moxibustion path to the robotic arm; The controller is used to receive the output data of the acupoint recognition and positioning system, control the robotic arm to move along the safe moxibustion path, and adjust the temperature, distance and technique of the end-effector according to the moxibustion parameter template. The human-computer interaction module is used to display the acupoint location results, confidence values ​​and moxibustion parameters, and to receive confirmation commands from the user.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the acupoint identification and positioning method as described in any one of claims 1-7.

Images