Acupuncture point identifying and positioning method based on human head point cloud
By constructing a standard human head point cloud model and performing non-rigid transformation, the problems of accuracy and efficiency in human head acupoint recognition and localization were solved, achieving high-precision acupoint recognition and localization.
Patent Information
- Application Number
- CN202511019344.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-23
- Publication Date
- 2025-11-18
AI Technical Summary
Existing technologies for identifying and locating acupoints on the human head suffer from accuracy issues due to variations in head circumference, head shape, and hairstyle. In particular, deep learning-based methods are fast but have poor accuracy, while non-deep learning-based methods lack both accuracy and real-time performance.
A standard human head point cloud model and its three-dimensional coordinates are constructed. Transformation parameters are calculated without relying on human reference points. A non-rigid transformation model is used to transform the target human head point cloud to obtain the three-dimensional coordinates of acupoints.
It improves the accuracy and efficiency of acupoint identification and positioning, reduces computational load and errors, adapts to different occlusions and changes in head shape, and achieves high-precision acupoint identification and positioning.
Smart Images
Figure CN120976306A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of point cloud, in particular, a point cloud based on human head acupoint recognition and positioning method is provided. BACKGROUND
[0002] In the field of traditional Chinese medicine teaching and training, physiotherapy, etc., the recognition and positioning of human head acupoints are needed. With the development of computer technology, computer devices are also widely used to assist in acupoint recognition and positioning. However, due to the different head circumferences / head shapes of different people, and the hair style which may block some acupoints on the head, the accuracy of computer-aided acupoint recognition and positioning is always affected in reality.
[0003] Currently, in the process of applying point cloud data to human head point cloud acupoint recognition and positioning under different shielding conditions, different head shapes and different face shapes, there are mainly two types of methods. One type is based on deep learning, such as iterative closest point (ICP), etc. The other type is based on non-deep learning, such as random sample consensus (RANSAC), normal distribution transform (NDT), etc.
[0004] Among these two types of methods, the method based on deep learning is fast, but the accuracy requires a large amount of data and has poor generalization. The method based on non-deep learning is mostly for rigid positioning, but it can meet the demand of non-rigid positioning by improving the mathematical model and positioning method. However, the accuracy and real-time performance are poor, and it is difficult to meet the accuracy and real-time performance requirements of acupoint recognition and positioning. SUMMARY
[0005] In order to solve the problems in the background art, the present application provides a point cloud based on human head acupoint recognition and positioning method.
[0006] A point cloud based on human head acupoint recognition and positioning method, comprising the steps of:
[0007] S100, obtaining a standard human head point cloud model of a human head including head acupoints, and obtaining a standard three-dimensional coordinate of the standard human head point cloud model;
[0008] S200, obtaining a target human head point cloud model of a target human head, and obtaining a human head reference point on the target human head point cloud model;
[0009] S300, obtaining a three-dimensional coordinate of the reference point;
[0010] S400, constructing a non-rigid transformation model, and constructing a transformation parameter of the transformation model;
[0011] S500, taking the reference point three-dimensional coordinates as an input of the transformation model and solving to obtain target transformation parameters;
[0012] S600, transforming the standard human head point cloud model by combining the target transformation parameters with the standard three-dimensional coordinates, and obtaining a transformed model and its transformed three-dimensional coordinates;
[0013] S700, taking the transformed model as a point cloud model of the target human head again, and taking the three-dimensional coordinates of the acupoints in the transformed model as the three-dimensional coordinates of the acupoints of the target human head, thereby completing the identification and positioning of the acupoints of the target human head point cloud.
[0014] Based on the above, the selected human head reference points are respectively the left and right outer canthi, the tip of the nose, the left and right corners of the mouth, and the chin.
[0015] Based on the above, the standard three-dimensional coordinates and the three-dimensional coordinates of the reference points are respectively world coordinates.
[0016] Based on the above, in step S400, the expression of the transformation model is:
[0017] Q i = s·R·P i +t
[0018] where P i is a point cloud of a standard human head point cloud model, Q i is a point cloud of a target human head point cloud model, s is a scaling parameter, R and t are respectively a rotation matrix and a translation matrix; the transformation parameters include the scaling parameter, the rotation matrix, and the translation matrix.
[0019] Based on the above, the scaling parameter is a scaling matrix, and the expression of the scaling matrix s is:
[0020]
[0021] where s x ′, s′ y , and s z ′ are 1x3 matrices, and the expressions are respectively:
[0022]
[0023] where s x , s y , and s z are located on the main diagonal of the scaling matrix and are respectively for the transformation of the model point cloud in the x, y, and z directions; s 12 , s 13 , s 21 , s 23 , s31 32 The six coefficients are used to describe the external contour of the standard human head point cloud model.
[0024] Based on the above, the rotation matrix R is:
[0025] R (a, b, g) = R z (g) * R y (b) * R x (a)
[0026] Wherein, the rotation angles a, b, g relative to the x, y, z axes, and the corresponding rotation matrix R x (a), R y (b) and R z (g) formulas are as follows:
[0027]
[0028] Based on the above, the translation matrix t is:
[0029] t = (t x , t y , t z )
[0030] Wherein, t x , t y and t z represent the displacement along the x, y and z axis direction respectively.
[0031] Based on the above, the data information of the rotation matrix is calculated and obtained by the singular value decomposition algorithm.
[0032] Based on the above, after calculating the rotation matrix and rotating the standard human head point cloud model, according to the point cloud of the target human head point cloud model and the rotated model, the data information of the scaling matrix and the corresponding translation matrix is calculated simultaneously by the least square method.
[0033] The present application has outstanding substantial features and significant progress compared with the prior art. Specifically, the present application can be used multiple times as a standard human head point cloud model by pre-constructing a standard human head point cloud model and world three-dimensional coordinates of the point cloud, and labeling the head acupoints, greatly reducing the subsequent workload. After constructing a non-rigid transformation model and the transformation parameters required by the model, the actual transformation parameters corresponding to the target human head point cloud model can be solved according to the target human head point cloud model and the three-dimensional coordinates of the reference points thereon, and then the standard human head point cloud model is transformed through the actual transformation parameters. The transformed human head point cloud model can be used as the final target human head point cloud model, and the point cloud three-dimensional coordinates and acupoint three-dimensional coordinates corresponding to the transformed model can be used as the coordinates of the target human head point cloud and the three-dimensional coordinates of the target human head acupoints, thereby facilitating the identification and positioning of the target human head acupoints. Since the standard human head point cloud model and its corresponding acupoints and three-dimensional coordinates can be reused once constructed, and only a small amount of reference points need to have their three-dimensional coordinates obtained after the target human head point cloud is obtained, the workload and data volume of the present application are greatly reduced, thereby greatly improving the efficiency. In addition, since the standard human head point cloud model is transformed to obtain the final target human head point cloud model, it is not necessary to consider whether the original target human head point cloud model has errors caused by hairstyle blocking and the like, thereby greatly improving the accuracy of acupoint identification and positioning. BRIEF DESCRIPTION OF DRAWINGS
[0034] Figure 1 is a flow structure schematic diagram of the present application.
[0035] Figure 2 is a comparison table of errors of different positioning methods of the present application.
[0036] Figure 3 is an error analysis table of part of the acupoints in the acupoint identification experiment of the present application.
[0037] Figure 4 is an error analysis table of part of the acupoints in the mechanical arm indication experiment of the present application.
[0038] Figure 5 is an error analysis table of head acupoint positioning of the present application. DETAILED DESCRIPTION
[0039] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.
[0040] like Figure 1 As shown, a method for acupoint recognition and localization based on human head point cloud includes the following steps: S100, obtaining a standard human head point cloud model including acupoints, and obtaining the standard three-dimensional coordinates of the standard human head point cloud model; S200, obtaining a target human head point cloud model, and obtaining a reference point on the target human head point cloud model; S300, obtaining the three-dimensional coordinates of the reference point; S400, constructing a non-rigid transformation model, and constructing transformation parameters of the transformation model; S500, using the three-dimensional coordinates of the reference point as input to the transformation model and solving to obtain target transformation parameters; S600, transforming the standard human head point cloud model using the target transformation parameters combined with the standard three-dimensional coordinates, and obtaining the transformed model and its transformed three-dimensional coordinates; S700, using the transformed model again as the point cloud model of the target human head, and using the three-dimensional coordinates of the acupoints in the transformed model as the three-dimensional coordinates of the acupoints on the target human head, thereby completing the recognition and localization of acupoints on the target human head point cloud.
[0041] In this embodiment, the obtained standard human head point cloud model is a human head point cloud model without hair occlusion from Morgan McGuire's Computer Graphics Archive model library. The world 3D coordinates of each point in the standard human head point cloud model are obtained using existing methods. Under the guidance of medical experts, the acupoints on the human head of the standard human head point cloud model are identified, located, and labeled. The standard human head point cloud model with 3D coordinates and acupoint labels is stored for subsequent reuse.
[0042] After obtaining the point cloud model of the target human head, reference points are selected. These reference points are point cloud points chosen based on the proportional relationship of the human head size, allowing the calculation of the relative positions between points to determine the proportions of the human head. In this embodiment, the selected reference points for the human head are point cloud points located at the left and right outer canthi, the tip of the nose, the left and right corners of the mouth, and the mental protuberance. After selecting the reference points, their world 3D coordinates are obtained.
[0043] In this embodiment, the three-dimensional coordinates of the reference point corresponding to the point cloud model of the target human head in world coordinates are obtained using the binocular ranging principle. The specific calculation process is as follows:
[0044] Binocular ranging primarily employs triangulation, a method that uses the geometric relationships of triangles to determine unknown distances. Assume we have two observation points with known locations (the coordinates of two cameras), and we can measure the angle or displacement difference between these two observation points and the target object. Using this information, we can calculate the actual distance to the target object using basic triangle formulas.
[0045] Assume x l and x r Let be the pixel coordinates of the target object P in the left and right camera images, respectively; B be the baseline between the two cameras; and f be the focal length of the two cameras. Based on triangle similarity, we can deduce:
[0046]
[0047] Simplification yields the distance z between the object and the camera:
[0048]
[0049] Where d represents parallax.
[0050] In this embodiment, the mathematical expression of the constructed non-rigid transformation model is:
[0051] Q i =s·R·P i +t (3)
[0052] Where P i For the point cloud in a standard human head point cloud model, Q i Let s be the point cloud of the target human head point cloud model, s be the scaling parameter, and R and t be the rotation and translation matrices, respectively. The definition of the scaling parameter s affects the non-rigid transformation model's representation of the differences between the standard human head point cloud model and the target human head point cloud model. To better describe the differences between the standard human head point cloud model and the target human head point cloud model, based on the proportional relationship of the human head size, the scaling parameter is defined as a scaling matrix. This matrix not only scales at different ratios in three directions but also introduces six additional coefficients to change the outer contour of the standard human head point cloud model, thus making it more closely resemble the target model. The scaling matrix expression is:
[0053]
[0054] Among them, s x ′、s′ y and s z Let ' be 1×3 matrices, and their expressions are:
[0055]
[0056] Among them, s x s y and s z Let be the transformations of the model point cloud in the x, y, and z directions, located on the main diagonal of the scaling matrix. The other six coefficients s 12 s 13 s 21 s 23 s 31 s 32 Together, they are used to describe the outer contour of the standard human head point cloud model, thereby making it more closely match the target human head point cloud model.
[0057] Furthermore, the rotation matrix R typically has three parameters: rotation angles α, β, and γ relative to the x, y, and z axes, respectively, and the corresponding rotation matrix R. x (α), R y (β) and R z The formula for (γ) is shown below:
[0058]
[0059]
[0060] Therefore, the total rotation matrix is:
[0061] R(α,β,γ)=R z (γ)·R y (β)·R x (α) (11)
[0062] Finally, the translation matrix t also consists of three parameters: t x t y and t z They represent displacements along the x, y, and z axes, respectively. Therefore, the translation matrix t is generally expressed by the following formula:
[0063] t=(t x ,t y ,t z (12)
[0064] The transformation parameters of the non-rigid transformation model include the scaling parameters, rotation matrix, and translation matrix mentioned above. The calculation process is as follows:
[0065] Since the scaling parameter s is primarily used to adjust the overall size or shape of the point set, but it does not change the relative orientation relationships between points, the scaling parameter does not need to be considered when calculating the rotation matrix R. Therefore, the model point cloud P can be... i With the target point cloud Q iThe correspondence between them can be simplified to the following formula:
[0066] Q i =R·P i +t (13)
[0067] At this point, calculating the transformation matrix using the existing SVD (Singular Value Decomposition) method is equivalent to calculating the point cloud P of a standard human head point cloud model. i After rotation and translation, the point cloud Q of the target human head point cloud model is compared. i The shortest distance between corresponding points in the equation results in the smallest error function, as shown in the formula:
[0068]
[0069] To minimize the above expression, it can be solved in the following way. First, obtain the point cloud P. i and target point cloud Q i Substituting the centroid coordinates p0 and q0 into formula (14), we get:
[0070]
[0071] As can be seen from the above equation, for any R, there will always exist a t such that this term is zero. Therefore, to minimize the error function, the first term must be minimized, which can be simplified to:
[0072]
[0073] Where, q i ′、p i ′ are respectively P i Q i Since R is an orthogonal matrix, the centroid coordinates can be simplified to:
[0074]
[0075] At this point, H (the complex conjugate transpose in SVD) can be obtained using the following formula:
[0076]
[0077] By performing singular value decomposition (SVD) on H, we can obtain U (the left singular vector in SVD) and V (the right singular vector in SVD):
[0078] [U,S,V]=SVD(H) (19)
[0079] The rotation matrix R can be obtained using U and V, as well as formula (16).
[0080] R = V·UT (20)
[0081] Then, after calculating the rotation matrix and rotating the standard human head point cloud model, based on the target point cloud and the rotated model, the data information required for the scaling matrix and the corresponding translation matrix can be calculated simultaneously using the least squares method. The calculation formula is as follows:
[0082]
[0083] Among them, Q′ R For P i After rotation, the point cloud, x i ′、y i ′ and z i ′ represents the point cloud Q i x, y, and z coordinates, x i y i and z i They are point cloud Q′ R The x, y, and z coordinates are then determined. The scaling matrix and the corresponding translation matrix can then be derived.
[0084] Finally, the point cloud of the standard human head point cloud model is transformed according to the actual transformation parameters using the formula:
[0085] Q rt =s·R·P i +t (24)
[0086] The transformed model serves as the target human head model, where the three-dimensional coordinates corresponding to the acupoints on the head are the point cloud coordinates of the acupoints on the target human head, thereby enabling the identification and localization of the acupoints on the target human head.
[0087] In this embodiment, a standard human head point cloud model is identified and located against multiple sets of different target human head point cloud models. The differences in performance between different identification and localization methods are determined by comparing the MAE (mean absolute error) and MAE_max between the final determined model and the actual target human head point cloud model using existing methods and the method proposed in this embodiment. Experimental data are shown below. Figure 2 As shown in Table 1, rigid positioning is a positioning method using the ICP transformation model. The other three are positioning methods using non-rigid transformation models that only use scaling factor, scaling scale, or scaling matrix as scaling parameters. The scaling factor refers to scaling the standard human head point cloud model at the same ratio as the whole. The scaling scale refers to scaling the standard human head point cloud model at different ratios in the x, y, and z directions. The scaling matrix is the method described in this paper.
[0088] Table 1 shows that the localization method using the scaling matrix transformation model, whether MAE or MAE_max, is superior to other methods. Since experiments 1, 2, and 7-10 all used target point cloud data with different heads (specifically in head contour, head circumference, etc.), the table shows that the localization method using the ICP transformation model has a larger error and is no longer suitable for point cloud localization in this case. For the localization method using the scaling factor and scaling scale transformation model, since the target model used in the experiments differs from the standard human head point cloud model in the scaling ratios in the three directions, the localization method using the scaling scale transformation model has a smaller error compared to the scaling factor. However, since the target model differs from the standard human head point cloud model not only in the scaling ratios in the three directions but also in some details, such as the more obvious differences in detail between the target model used in experiment 9 and the standard human head point cloud model (e.g., a sharper head and rounder cheeks), the error between the point cloud obtained by the scaling factor and scaling scale transformation model and the target model is larger. The error between the point cloud obtained by the scaling matrix transformation model and the target model is around 0.9mm, regardless of the difference between the details of the target model and the standard human head point cloud model. The MAE is also around 3.1mm. This is a significant improvement compared to the transformation model using scaling factor and scaling scale. Therefore, the transformation model using the scaling matrix is more suitable for the localization of human head 3D point clouds.
[0089] To evaluate the accuracy of acupoint identification and localization in automatic acupoint recognition, the experiment, under the guidance of professional physicians, compared and analyzed the errors between the coordinates of the identified acupoints on the target human head and the actual coordinates of the acupoints. The errors for some acupoints were as follows: Figure 3 As shown in Table 2.
[0090] The absolute error refers to the Euclidean distance between the predicted acupoint coordinates and the actual coordinates. As can be seen from the data in Table 2, the automatic acupoint identification and positioning method for human head designed in this invention has high accuracy in acupoint positioning in two dimensions. In terms of pixels, the acupoint identification error is within three pixels, and the absolute error does not exceed three pixels.
[0091] To evaluate the positioning accuracy of the automatic acupoint recognition and positioning method in practical applications, an experiment was conducted under the guidance of a professional physician. The error between the acupoints indicated by the robotic arm of the physiotherapy device based on the recognition and positioning results and those indicated by the physician was compared and analyzed. To ensure the accuracy of acupoint coordinates obscured by hair, the error was defined as the distance between the actual coordinates of the acupoint indicated by the physician on the skin and the coordinates indicated by the robotic arm. The error for some acupoints is shown below. Figure 4 As shown in Table 3.
[0092] Table 3 shows the acupoint errors indicated by some robotic arms. Several representative acupoint error data are listed in the table. This error includes the error of acupoint location and the error of robotic arm indication. The acupoints with the largest indication errors in the table are Youqubin, Wangu, and Youtongtian. These acupoints are all located at the farthest position from the robotic arm, and all three acupoints are obscured by hair. Wangu is also obscured by the ear, which causes the indication error of these acupoints to be large.
[0093] To verify the generalization ability of the proposed method, robotic arm pointing tests were conducted on two targets with significantly different head shapes. The average, maximum, and standard deviation of the positioning errors for the three poses of the two targets were compared to verify the generalization ability of the proposed method. Error analysis is as follows: Figure 5 As shown in Table 4.
[0094] As can be seen, compared with existing methods, the human head point cloud localization method proposed in this invention has the following improvements: First, the method proposed in this invention replaces large model point clouds with small model point clouds with proportional relationships for localization, reducing the amount of computation and improving the computation speed; in addition, the method proposed in this invention achieves higher localization accuracy for non-standard human head models and standard human head point cloud models with different head shapes and face shapes under different occlusion conditions. Therefore, it can be considered that the method proposed in this invention can quickly and accurately localize non-standard human head models and standard human head point cloud models with different head shapes and face shapes under different occlusion conditions.
[0095] Based on the above method, the three-dimensional standard human head point cloud model with acupoints marked on the head can be located with the three-dimensional model of the human head being tested, thereby obtaining the three-dimensional spatial coordinates of all acupoints on the human head being tested.
[0096] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of equivalents of the claims be included within the present invention.
Claims
1. A method for acupoint identification and localization based on human head point clouds, characterized in that, Including the following steps: S100. Obtain a standard human head point cloud model, including acupoints on the head, and obtain the standard three-dimensional coordinates of the standard human head point cloud model. S200. Obtain the target human head point cloud model and obtain the human head reference point on the target human head point cloud model. S300. Obtain the three-dimensional coordinates of the reference point; S400. Construct a non-rigid transformation model and construct the transformation parameters of the transformation model; S500. Use the three-dimensional coordinates of the reference point as input to the transformation model and solve to obtain the target transformation parameters; S600. Transform the standard human head point cloud model by combining the target transformation parameters with the standard three-dimensional coordinates, and obtain the transformed model and its transformed three-dimensional coordinates. S700. The transformed model is used again as the point cloud model of the target human head, and the three-dimensional coordinates of the acupoints in the transformed model are used as the three-dimensional coordinates of the acupoints on the target human head, thereby completing the identification and positioning of the point cloud acupoints on the target human head.
2. The acupoint identification and localization method based on human head point clouds according to claim 1, characterized in that: The selected reference points for the human head are the left and right outer canthi, the tip of the nose, the left and right corners of the mouth, and the point cloud points corresponding to the mental process.
3. The acupoint identification and localization method based on human head point clouds according to claim 1, characterized in that: The standard three-dimensional coordinates and the three-dimensional coordinates of the reference point are world coordinates, respectively.
4. The acupoint identification and localization method based on human head point clouds according to claim 1, characterized in that, In step S400, the expression for the transformation model is: Q i =s·R·P i +t Where P i The point cloud of a standard human head point cloud model, Q i is the point cloud of the target human head point cloud model, s is the scaling parameter, and R and t are the rotation matrix and translation matrix, respectively; The transformation parameters include scaling parameters, rotation matrix, and translation matrix.
5. The acupoint identification and localization method based on human head point clouds according to claim 4, characterized in that: The scaling parameter is a scaling matrix, and the expression for the scaling matrix s is: Among them, s x ′、s′ y and s z ' and ' are 1×3 matrices, and their expressions are respectively: Among them, s x s y and s z Let s be the transformations of the model point cloud in the x, y, and z directions, located on the main diagonal of the scaling matrix; 12 s 13 s 21 s 23 s 31 s 32 Six coefficients are used together to describe the external contour of a standard human head point cloud model.
6. The acupoint identification and localization method based on human head point clouds according to claim 4, characterized in that: The rotation matrix R is: R(a,b,c)=R z (c)·R y (b)·R x (a) Wherein, the rotation angles α, β, and γ relative to the x, y, and z axes, and the corresponding rotation matrix R x (α), R y (β) and R z The formula for (γ) is as follows:
7. The acupoint identification and localization method based on human head point clouds according to claim 4, characterized in that, The translation matrix t is: t=(t x ,t y ,t z ) Among them, t x t y and t z These represent displacements along the x, y, and z axes, respectively.
8. The acupoint identification and localization method based on human head point clouds according to claim 4, characterized in that: The rotation matrix data is calculated and obtained using the singular value decomposition algorithm.
9. The acupoint identification and localization method based on human head point clouds according to claim 8, characterized in that: After calculating the rotation matrix and rotating the standard human head point cloud model, the scaling matrix and the corresponding translation matrix are calculated simultaneously using the least squares method based on the point cloud of the target human head point cloud model and the rotated model.